# Aginies — full site text Generated at build time from the public pages of https://www.aginies.com. English pages first, then Turkish. Pages behind an access code are not included. --- # Aginies — The orchestration layer that takes AI from pilot to production URL: https://www.aginies.com/en Language: en Summary: Enterprise agentic AI orchestration platform. Supervisor agents plan, specialised agents act in core systems; every step is traced, approved where needed and costed. On-prem, hybrid or cloud. Agentic orchestration · Sovereign by design # The layer that takes AI from pilot to production. Supervisor agents plan, specialised agents act inside your core systems; every step is traced, passes human approval where needed, and is costed. Model-agnostic; on-prem, hybrid or cloud. Book a 30-minute discovery session → Watch a live agent The agents below are simulations derived from real agent definitions. Pick a sector to watch a different scenario. Banking ### Dialogue banking: voice and text transaction assistant Open agent → [Dialogue banking: voice and text transaction assistant] Audit trail00:00.00 - ◦App chat: "pay my electricity bill, raise my limit to 40k, I don't recognise yesterday's 1,250" - ∿Three intents separated: bill payment · limit increase · transaction dispute; order and dependencies set run #41,289,010elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Limit increase > 60%, dispute amount > 5,000 or transfer to a new payee → operations approval Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Numbers from production · live in retail, insurance, automotive and banking enterprises 13,120,000+executions · last 30 days 97.4%success rate · all agents 1,920+agents active in production 13,100+total agents · all deployments From the observability layer · October 2026 The problem ## The hard part of enterprise AI is getting to production. Most enterprises are trying something with AI; very few can run it in production, at scale, under control. ### The PoC gap Pilots impress, but the path to production is undefined. ### Scattered initiatives Every unit trials its own tool; no enterprise coherence. ### Unmeasured value The gap between “it works” and “it pays off” goes unreported. ### Governance gap Security, compliance and auditability are afterthoughts. ### Outsourcing dependency Person-months are bought; knowledge leaves with the person, timesheets do not show what was done. Result: budget spent, transformation missing. One accountable partner ## From consulting to platform, from infrastructure to operations. A partner that owns enterprise AI transformation and runs it end-to-end. Four components under one roof; accountability in one place. 01 ### Expert team Strategy, data, engineering and change management in one experienced team. Consultants do not come and go; we operate the product with you. 02 ### The Aginies platform An enterprise-grade agentic orchestration layer that runs every AI workload from one place. Every AI scenario in the organisation runs on the same execution layer, acts inside core systems and leaves an audit trail. 03 ### Sovereign infrastructure Independent operation within national borders with GPU-as-a-Service. Data and model control stay with you; usage-based capacity with no capital outlay. 04 ### Agent workforce Outsourced agents instead of outsourced people: roles such as iOS developer, business analyst, accountant and tester work in your tools; every task in the work log, decisions with you. Autonomous Agent ## Every agent starts as a dialogue. Describe what you need; the Autonomous Agent sets up the supervisor and specialist agents, binds tools and the knowledge base, runs the test corpus and deploys with your approval. Every agent in the library can be produced this way. Autonomous Agent and the product family → Autonomous AgentAginies Workspace · Dialogue banking: voice and text transaction assistant Message the Autonomous Agent… AgentTestsLive The agent will appear here Product familyPlatform Aginies is the umbrella; the others are its modules. - Platform AginiesEverything comes together here What is included - Visual agent builder, workspace and tenant management, roles via SSO - Execution engine, approval cards, kill-switch, dashboards and reports - Knowledge-base management and the Autonomous Agent Integration With the organisation’s identity provider (SAML/OIDC), SIEM, data warehouse and ticketing. 77 / 77 scenarios use it See the screen and details → - Voice AginiesSTT · TTS · ASR What is included - Streaming Turkish ASR and STT (aginies-voice-tr-v3), natural TTS (tts-tr-v2) - Voiceprint verification, spoken OTP, warm hand-off - Call quality analysis: transcript, sentiment, compliance checks Integration SIP with Genesys, Avaya and open-source switches; API with recording systems. 23 / 77 scenarios use it See the screen and details → - Code AginiesIDE Registry · IDE extensions · code completion What is included - IDE extensions (VS Code, JetBrains) and in-house code completion - IDE Registry: approved extension and model distribution, version pinning - SDLC agent team: analysis, design, development, test, review Integration With Jira, Git (GitHub, GitLab, Bitbucket, self-hosted), CI/CD and static analysis tools. 8 / 77 scenarios use it See the screen and details → - Hub AginiesModel Hub · MCP Hub · Model proxy What is included - Model catalogue: tulpar, talay, sunkar, konrul families and current open releases - Model proxy and key management; usage and cost tracking - MCP Hub: approved tool servers, invocation policies, connection tests Integration With in-country GPU pools; with the organisation’s chosen cloud model providers through the proxy. 77 / 77 scenarios use it See the screen and details → - Vision AginiesDocuments · images · masking What is included - Invoice, contract, form and ID document extraction; per-field confidence - Damage-photo assessment and visual quality control - Personal-data detection and document masking Integration With document stores (S3, SharePoint), scanners and ECM; results written to ERP and claims systems through tools. 21 / 77 scenarios use it See the screen and details → - Workforce AginiesAgent workforce, on demand What is included - Role catalogue: engineering, business analysis, finance, support, HR, legal - Work log: task, output, hour-equivalent, quality score, review status - Shadow period → sampled review → autonomous; a named team lead Integration With the organisation’s own tools: Jira, Git, ERP, ticketing, e-mail. 17 / 77 scenarios use it See the screen and details → - Dialog AginiesDialogue UI What is included - Web and mobile dialogue UI; embeddable widget - WhatsApp, SMS and e-mail channels with the same agent - Approval cards, file and image sharing, warm hand-off Integration With the organisation’s identity and OTP stack; with the WhatsApp Business API and SMS gateways. 60 / 77 scenarios use it See the screen and details → Agent Outsourcing · Workforce Aginies ## Hire agents, not headcount. Instead of hiring an iOS developer, business analyst, accountant or tester from an outsourcing firm, you hire agents that hold the same role. The agent works in your tools and reports to your team; what it did, how long it took and what it cost are in the work log. Role catalogue and engagement model → - iOS Developer Agentreplaces: Outsourced iOS developerPRs / week · First-review pass rate - Business Analyst Agentreplaces: Outsourced business analystStories / week · Return rate from development - Accountant Agentreplaces: Outsourced accountantDays to close · Reconciliation breaks - QA Engineer Agentreplaces: Outsourced testerEscaped defects / release · Bug validity rate Workforce Aginies· roster and work log · fintech-opsMon 09:00 hired agents · 4 ios-dev-01iOS Developer ready mobile0 h this week qa-01QA Engineer ready mobile0 h this week ba-01Business Analyst ready payments0 h this week acct-01Accountant ready finance0 h this week Unified observability and reporting ## Track agent activity, platform health and model costs from one dashboard. Every run is recorded step by step; every request can be masked and anonymised. Separated by workspace and tenant; teams share and co-develop agents inside the organisation. - Per-agent success heatmap and error timeline - Platform health, model usage and cost - Proxy to cloud models with your own keys, observed from one point - Masking, anonymisation, full audit trail Explore the platform → Platform Aginies · Dashboard · last 30 daysall workspaces · all tenants executions13,120,000 success97.4% p50 latency1,4 s platform healthhealthy busiest agents · each cell ≈ 1 day dialog_banking_assistant%97.4 card_debt_payment%99.1 kvkk_data_detection%96.2 kvkk_document_masking%98.8 motor_claim_end_to_end%98.1 merchant_operations%100 model usage aginies-tulpar-438% aginies-talay-229% aginies-sunkar-119% aginies-voice-tr-v39% cloud · proxy$2.960 cost · by scenario Dialogue banking26% Credit & onboarding24% Claims operations18% KVKK masking14% Merchant ops10% Other8% PII masking: onanonymisation: onaudit trail: full1,920+ active agents How it works ## Agentic AI executes real business processes in four steps. - 01 ### Understand signals Inputs from applications, APIs, documents, voice channels and events become execution-ready tasks. ApplicationsAPIsDocumentsVoiceEvents - 02 ### Supervisor plans Complex objectives are decomposed into structured steps; responsibilities are assigned to specialised agents. SupervisorPlanningDecompositionDelegation - 03 ### Agents act Knowledge is retrieved, documents processed, tools invoked, and real operations triggered in enterprise systems. RetrievalTools / MCPActionsIntegrations - 04 ### Everything is observed Agent decisions, latency, cost and tool usage are monitored and optimised with full visibility across environments. TracesLatencyGuardrailsAudit Platform ## Eight integrated capability layers. Built for enterprise scale, governance and resilience. The layers run as one platform. Explore the platform → A ### Orchestration & Execution Intelligence Runs complex work end-to-end; routes requests intelligently and keeps multi-step processes stable even when dependencies fail. B ### AI / Model Orchestration Uses the right model for each task; reduces vendor lock-in and keeps operations running when a provider is unavailable. C ### Knowledge & Retrieval Turns fragmented enterprise information into governed, searchable, trustworthy knowledge; answers link back to sources. D ### Governance, Risk & Compliance Makes AI auditable, safe and fit for regulated environments; every decision is traceable. E ### Observability & Optimisation Provides visibility into accuracy, speed, cost and health; produces the data for continuous improvement. F ### Integration & Connectivity Connects to enterprise systems and customer channels without rebuilding the existing technology landscape. G ### Multichannel Experience & Voice Delivers AI across digital, messaging and voice channels on one consistent orchestration underneath. H ### Deployment & Enterprise Architecture Supports the on-premise, hybrid and sovereignty requirements of large enterprises. Real solutions, running in production ## Workloads running live on Aginies today. Retail & E-commerce ### E-commerce support & action Where is my order, returns, delivery changes: the agent completes the transaction in the order system. Read more → Enterprise Functions ### Invoice & document extraction OCR → validation → ERP: clean records without manual entry. Read more → Insurance ### Voice AI for claims intake A voice agent takes the loss notice and opens the claim in the system. Read more → Software Delivery ### SDLC agent team PM, analyst, architect, developer and tester agents working together in the CI/CD pipeline. Read more → Enterprise Functions ### Agent outsourcing (agent workforce) Outsourced agents instead of outsourced people: iOS developer, business analyst, accountant, tester; with a work log. Read more → Banking ### Conversational banking Complete mobile banking operations by typing or talking. Read more → 77 live agents in the agent library → 77live scenarios 30solution packages 14sectors 200systems connected 73approval rules 657tool definitions Derived at build time from the agent library’s scenario definitions. IndustriesFourteen sectors, one platform. - Banking9 - Fintech & Payments5 - Insurance6 - Retail & E-commerce6 - Telecom5 - Healthcare4 - Manufacturing & Automotive4 - Logistics3 - Energy & Utilities5 - Government & Public Sector5 - Enterprise Functions9 - Software Delivery8 - Defence & Aerospace4 - IoT & Connected Devices4 What sets us apart ## A disciplined transformation methodology. We decide with numbers which AI ideas are worth building and when to start. Every use case passes a five-layer assessment; each stage ends at a clear decision point: proceed, fix or stop. - 01 Value Why should this be done at all? - 02 Readiness Do we have the inputs we need? - 03 Feasibility Can we actually build this? - 04 Operations How will it live after go-live? - 05 Governance Is it secure, controlled and auditable? Assess your scenario in 3 minutes → Methodology Security, compliance and sovereignty ## Designed for regulated environments. AI must meet the same standards as every other critical system. Built to the requirements of regulated sectors such as finance, insurance, healthcare, government and defence; closed-network and air-gapped deployment included. SOC 2 Type IIcompliant platform ISO 27001compliant platform GDPR & KVKKdata residency guaranteed On-prem / air-gapdata never leaves See the architecture and deployment options → ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Agent Outsourcing URL: https://www.aginies.com/en/agent-outsourcing Language: en Summary: Outsourced agents instead of outsourced people: roles such as iOS developer, business analyst, accountant and test engineer hired as agents that work inside your tools; every task in the work log, decisions with you. Agent Outsourcing · Workforce Aginies # Hire agents, not headcount. Instead of hiring an iOS developer, business analyst, accountant or tester from an outsourcing firm, you hire agents that hold the same role. The agent works in your tools and reports to your team; what it did, how long it took and what it cost are in the work log. Decisions stay with your approver, with a named team lead on the Aginies side. 12+ready roles · unlimited custom roles 2 weeksshadow period, 100% review 24/7available; no payment for idle time 1named Aginies team lead Workforce Aginies· roster and work log · fintech-opsMon 09:00 hired agents · 4 ios-dev-01iOS Developer ready mobile0 h this week qa-01QA Engineer ready mobile0 h this week ba-01Business Analyst ready payments0 h this week acct-01Accountant ready finance0 h this week Simulation: one week of four hired agents. Use the tabs for the daily digest, weekly report, release bulletin and work log; give an agent an instruction from the box below. Why ## Three chronic problems of the outsourcing model 01 ### Turnover and lost knowledge The person leaves and so does what they knew; the next one starts from zero. An agent’s knowledge stays in the workspace; the next one starts with the same corpus. 02 ### Invisibility A timesheet shows only the hours logged. The work log shows every task, output, quality score and approval. 03 ### Slow scaling A new hire takes months; a second agent of the same role takes minutes. When demand drops, capacity drops and so does the invoice. Role catalogue ## Roles you can hire today Every role is an agent spec: tasks, tools, outputs, KPIs and approval points. Roles outside the catalogue are defined with the Autonomous Agent within days in a discovery session. Engineering ### iOS Developer Agent replaces: Outsourced iOS developer aginies-sunkar-1 · Code Aginies What it does - Takes the Jira story, implements it in SwiftUI, writes unit and UI tests - Opens the PR, applies review comments, gets CI green - Drafts release notes and App Store metadata Where it works JiraGitHub / GitLabXcode CloudTestFlightFigma Outputs Merged PRsStory points closedTest coverage delta Weekly KPIs PRs / weekFirst-review pass rateBuild success rateHour-equivalent Human control: Every PR is reviewed by a person; production deployment under the team lead’s approval Example in the library → ### Backend Developer Agent replaces: Outsourced backend developer aginies-sunkar-1 · Code Aginies What it does - Writes API endpoints, data model and migrations - Fits the existing codebase: lint, architecture rules, observability - Produces integration tests and API docs Where it works JiraGitHubCI/CDPostgreSQLOpenAPI Outputs Merged PRsEndpoints / servicesBugs closed Weekly KPIs PRs / weekBug reopen ratep95 latency impactHour-equivalent Human control: Schema changes and production deployment human-approved Example in the library → ### QA Engineer Agent replaces: Outsourced tester aginies-talay-2 · on-prem What it does - Derives test cases from acceptance criteria, maintains the regression suite - Runs regression on every release, files bugs with evidence and steps - Reports coverage and escaped defects Where it works Jira / XrayPlaywrightXCUITestCI/CDBrowserStack Outputs Tests run / releaseBugs filedRegression coverage Weekly KPIs Escaped defects / releaseBug validity rateRegression durationHour-equivalent Human control: The release “go” decision stays with the release manager Example in the library → ### DevOps / SRE Agent replaces: Outsourced DevOps engineer aginies-tulpar-4 · on-prem What it does - Writes and maintains pipelines, IaC and monitoring rules - First response in incidents, applies runbooks, rolls back - Produces cloud cost and capacity reports Where it works GitHub ActionsTerraformKubernetesGrafanaPagerDuty Outputs Incidents resolvedPipeline changesCost-saving proposals Weekly KPIs MTTRDeployment frequencyChange failure rateHour-equivalent Human control: Production infrastructure changes and rollbacks approved Example in the library → Business analysis and data ### Business Analyst Agent replaces: Outsourced business analyst aginies-tulpar-4 · on-prem What it does - Writes user stories and acceptance criteria (Gherkin) from requests and meeting notes - Produces process maps, data dictionary and impact analysis - Grooms the backlog, makes it estimation-ready Where it works JiraConfluenceMiroTeams / Slack Outputs Ready storiesProcess documentsImpact analyses Weekly KPIs Stories / weekReturn rate from developmentCriteria clarity scoreHour-equivalent Human control: The product owner approves the stories Example in the library → ### Data Analyst Agent replaces: Outsourced data analyst aginies-tulpar-4 · on-prem What it does - Produces SQL queries, dashboards and weekly reports - Explains anomalies and trends, linked to source data - Runs data-quality checks Where it works BigQuery / SnowflakedbtPower BI / LookerSlack Outputs ReportsDashboardsAnalysis requests answered Weekly KPIs Request-to-report timeRework rateData-quality findingsHour-equivalent Human control: Externally shared reports under the analytics lead’s approval Finance and operations ### Accountant Agent replaces: Outsourced accountant aginies-tulpar-4 · Vision Aginies What it does - Processes invoices and statements, prepares journal entries - Performs bank and vendor reconciliation, explains differences - Runs the month-end close checklist, prepares VAT Where it works ERP (SAP / Logo / Netsis)e-FaturaBanka APIExcel Outputs Entries postedReconciliationsClose tasks completed Weekly KPIs Days to closeReconciliation breaksCorrection rateHour-equivalent Human control: Entries posted with the accountant’s approval; filings stay with people Example in the library → ### Procurement Specialist Agent replaces: Outsourced procurement specialist aginies-talay-2 · on-prem What it does - Collects requests, prepares RFQs, compares supplier responses - Runs three-way matching of order, receipt and invoice - Produces supplier performance reports Where it works ERPE-postaTedarikçi portalıExcel Outputs RFQsMatched invoicesSavings reports Weekly KPIs Request-to-order timeMatch exceptionsSavingsHour-equivalent Human control: Order approval and supplier selection with the procurement manager Example in the library → Customer, sales and HR ### Customer Support Rep Agent replaces: Outsourced contact-centre agent aginies-talay-2 · on-prem What it does - Answers tickets and chats, acts in the system - Hands out-of-policy requests to a person with rationale - Writes the weekly topic and root-cause report Where it works Zendesk / SalesforceDialog AginiesWhatsAppCRM Outputs Tickets resolvedFirst-contact resolutionsEscalations Weekly KPIs CSATAverage resolution timeEscalation rateHour-equivalent Human control: Refunds, cancellations and above-limit actions approved; 10% sampled quality review Example in the library → ### Sales Development Rep Agent replaces: Outsourced SDR aginies-tulpar-4 · on-prem What it does - Researches and scores leads, updates the CRM - Drafts personalised outreach; sending is human-approved - Schedules meetings and prepares briefings Where it works HubSpot / SalesforceLinkedInE-postaTakvim Outputs Qualified leadsMeetings bookedBriefings Weekly KPIs Meetings / weekQualification accuracyReply rateHour-equivalent Human control: External outreach always sent with human approval Example in the library → ### Recruiter Agent replaces: Outsourced recruiter aginies-tulpar-4 · on-prem What it does - Pre-screens applications against criteria, writes rationale - Schedules interviews, runs candidate communication - Produces the hiring funnel report Where it works ATS (Greenhouse / Kariyer.net)TakvimE-postaHRIS Outputs Candidates screenedInterviews scheduledFunnel reports Weekly KPIs Time to fillInterview-to-hire ratioCandidate satisfactionHour-equivalent Human control: Screening and offer decisions with HR; the agent proposes with rationale Example in the library → Legal and compliance ### Contract Analyst Agent replaces: Outsourced legal assistant aginies-tulpar-4 · on-prem What it does - Extracts clauses, compares with the playbook, flags deviations - Drafts redlines and summaries - Maintains the contract calendar and renewal alerts Where it works CLMWordE-postaDMS Outputs Contracts reviewedRedlinesSummaries Weekly KPIs Review turnaroundMissed deviationsCounsel acceptance rateHour-equivalent Human control: Every redline goes to the counterparty with counsel approval Example in the library → + ### A role outside the catalogue? There is no limit on the number of roles. Write the job description; the Autonomous Agent composes the supervisor and specialists, connects them to your tools and runs the test corpus. After the two-week shadow period the role enters the catalogue. Typical examples: content editor, supply-chain planner, legal secretary, marketing analyst, technical writer. Let us scope the role together → The path of one instruction ## From instruction to bulletin in seven steps The agent takes work from a Jira story, an e-mail or a direct message; its plan is visible; it works in your tools; checks run; outputs over threshold go to a person; the result is written to the work log; digest, report and bulletin are produced from that record. - 01InstructionA Jira story, an e-mail or a direct message: “fix the dark-mode bug in PAY-119” - 02PlanScope, tools to touch, approval point; a 3-step plan visible to the team lead - 03WorkIn your tools: code, tests, entries, reports; every step traced - 04ChecksLint, tests, policy, PII; check families scored - 05ApprovalOutputs over threshold go to a person: PR review, above-limit entries - 06Work logTask · output · hour-equivalent · quality · approval · cost - 07BulletinsDaily digest 09:00 · weekly report Friday · release bulletin on release Routines ## Like an employee: morning digest, Friday report, bulletin on release A hired agent’s day flows like an employee’s. A new instruction can be given at any moment; the agent shows its plan and progress, and stops to ask on steps outside its authority. | when | what | content | to whom | | --- | --- | --- | --- | | Every morning 09:00 | Daily digest | Finished yesterday, planned today, blockers; per agent | team lead · Slack / e-mail | | Continuous | Work log | Every task a row: output, hour-equivalent, quality, approval, cost | approver · approval queue | | Friday 17:00 | Weekly report | Outputs, KPIs, quality, cost; timesheet export | team lead · management | | Every release | Release bulletin | What shipped, who did what, tests, approvals, rollout stage | team · stakeholders | | Month end | Close and invoice | Accounting close tasks; invoice line from the work log | finance · procurement | | On demand | Instruction | New work by direct message, Jira or e-mail; plan and progress visible | anyone · within permissions | Examples ## A weekly report and a release bulletin, written by the agents Weekly report · week 38 · fintech-opsFriday 17:00 · timesheet exported · invoice line ready | agent | role | outputs | what | quality | hour-eq. | cost | | --- | --- | --- | --- | --- | --- | --- | | ios-dev-01 | iOS Developer | 5 | 4 PRs · 21 points · release notes | 4.7 | 20 h | $6.7 | | qa-01 | QA Engineer | 6 | 1,245 tests · 1 bug · go recommendation | 4.8 | 24 h | $4.8 | | ba-01 | Business Analyst | 5 | 14 stories · 2 processes · retro | 4.6 | 20 h | $4.5 | | acct-01 | Accountant | 5 | 236 entries · 3 reconciliations · close | 4.9 | 23.5 h | $4.6 | 21outputs 87.5 hhour-equivalent 9human approvals $20.6total cost - Release 2.14.0 live; escaped defects 0; crash rate 0.08% (threshold 0.3%). - September close finished on working day 5 (previous month 9); 3 items with the accountant’s approval. - Human equivalent of the same work ~88 hours; the agent-month cost of four roles sits below it. Figures are illustrative. release bulletin2.14.0 · fintech-ops · mobilThu 13:00 · e-mailed to team and stakeholders What shipped Payment summary screen (PAY-117), dark-mode fixes (PAY-119), 6 bug fixes ios-dev-01 PR #4818, #4823 merged · 10 points · release notes and App Store metadata qa-01 rc2 → rc4: 421/421 · BUG-2291 found, fix verified · escaped defects 0 ba-01 PAY-118 acceptance criteria were the source of 9 tests in the release; 3 unclear stories with the product owner Approvals d.kaya (PR review) · e.demir (release go) · recorded: who, when, rationale Rollout 10% staged → 24 h monitoring (crash threshold 0.3%) → 50% → 100% Next PAY-121 and PAY-122 ready for development; PAY-124 in review The bulletin is written by the agents from the work log; the daily digest (every morning 09:00) and the weekly report (Friday 17:00) are produced the same way. Tracking ## What it did and how much: the work log Every task is a row: ticket id, output, hour-equivalent, quality score, approval and cost. The weekly timesheet and the monthly invoice come from the same record; no surprises. TaskTicket / story id, start and finish, status OutputPR, report, entry, ticket: concrete and countable Hour-equivalentEstimated human time for the same work; for billing and comparison QualityReview result, rework, check-family scores ApprovalWho, when, rationale; every step over threshold CostTokens, tool calls, model; per task and role dailywork log · approval queue weeklytimesheet · KPIs · quality monthlyinvoice · review · scale decision Every row is a task; the weekly timesheet and the monthly invoice are produced from the same record. What was done, how long it took, who approved it and what it cost are answered in one place. Engagement model ## Starts in five steps, then scales From role scoping to the monthly review, responsibility is clear at every step. In the shadow period every output is reviewed; once the quality threshold is cleared, review becomes sampled. The kill-switch is always on. - 01Role scopingThe job description becomes an agent spec: tasks, tools, outputs, KPIs, approval points.you + the Aginies team lead - 02Access and onboardingThe agent gets accounts in your tools: SSO, roles, read / write permissions, data boundaries. Code and data stay with you.IT + security - 03Shadow periodFor the first two weeks every output is reviewed by a person; mistakes are added to the test corpus, the agent is corrected.your team lead + Aginies - 04Sampled reviewOnce the quality score clears the threshold, review drops to 10% sampling and threshold-based approval; the kill-switch is always on.your team lead - 05Monthly review and scaleWork log, KPIs and cost are read together; the role is widened, a new role added, or the contract ended.you + Aginies Comparison ## Human outsourcing and agent outsourcing side by side Two models doing the same work; the difference shows in time to start, traceability, knowledge retention and pricing. In both, the final decision belongs to your team. | | outsourced people | outsourced agents · Workforce Aginies | | --- | --- | --- | | Time to start | 4–8 weeks of search, contract, onboarding | Days: role spec, access, shadow period | | Availability | Office hours, leave, turnover | 24/7; idles when demand drops, scales when it rises | | Knowledge | Arrives and leaves with the person | Stays in the workspace; the next agent starts with the same corpus | | Traceability | Timesheet and weekly report | Step-by-step work log: task, output, quality, approval, cost | | Quality consistency | Varies by person and day | Versioned with a test corpus and check families | | Data security | Laptop, home network, third-party office | Inside your tenant; code and data never leave | | Scaling | New hires, months | A second agent of the same role in minutes | | Pricing | Person-month, regardless of use | Agent-month or per output; no payment for idle capacity | | Human accountability | The vendor’s project manager | A named Aginies team lead; decisions with your approver | Commercial model ## Three pricing models, no list prices Set together in discovery according to role, volume and approval intensity. You own the outputs; contracts can be ended monthly. ### Agent-month A fixed monthly fee per role; unlimited tasks, defined KPIs and SLA. Best fitRoles with continuous workload ### Output-based Fee per merged PR, resolved ticket or processed invoice; the work log is the invoice line. Best fitVariable or project-based work ### Agent pod Several roles + an Aginies team lead; works like a product team with weekly planning and retrospectives. Best fitTeams that want end-to-end delivery Measured targets and rollout ## Your first agent leaves the shadow period in 4 weeks. Targets as ranges from comparable deployments; baseline and target fixed together in the first week. Measured targets | Accepted output share | 85% and above | | Cost per role | 30–50% of an outsourced person-month | | Knowledge loss at turnover | Zero; knowledge stays in the workspace | What you need to start - Tool access the role needs (Jira, Git, ERP, ticketing) - Task definition and acceptance criteria - A review owner on the organisation side Rollout - 1 · Policy and role catalogue · 2 weeksHR policies, process owners and system access defined; personal-data boundaries written. - 2 · Pilot · 4 weeksOne department or one role; quality score and user feedback tracked weekly. - 3 · Scale-outAll departments; thresholds, review sampling and kill-switch owned by HR management. Governance ## Managed like an employee, audited like a system - Accounts in your tools, in your tenant; code and data never leave - Read / write permission per tool; steps over threshold go to approval - Who approved every output, and why, is recorded - A named Aginies team lead; weekly planning, monthly review - Kill-switch: immediate stop per role or task - You own the outputs and the knowledge produced; the corpus stays with you when you leave In the library ## Live examples of hired roles Software DeliveryHITL ### iOS feature implementation (hired developer agent) The Jira story is taken, built in SwiftUI, tested and opened as a PR; review and release decisions stay with the team. 380 runs / month 94.4% success 4 agents Software DeliveryHITL ### Feature delivery pipeline A Jira epic turns into criteria, design, code and tests via agents; humans only review and approve the PR. 1,200 runs / month 91.2% success 5 agents Software DeliveryHITL ### Regression test cycle (hired QA agent) Regression runs on every release candidate, new bugs are filed with evidence, a coverage report is written; the “go” decision stays with the release manager. 260 runs / month 95.8% success 4 agents Software DeliveryHITL ### Incident response and hotfix Alert fires, bug reproduced, patch and regression run ready; on-call approves, post-mortem already drafted. 320 runs / month 90.5% success 5 agents Software DeliveryHITL ### Backlog grooming and estimation Incoming requests are triaged, clarified and sized; a sprint proposal goes to the product owner for approval. 340 runs / month 92.8% success 4 agents Enterprise FunctionsHITL ### Month-end close and reconciliation (hired accountant agent) Invoices are processed, bank and vendor reconciliations done, the close checklist run; entries are posted with the accountant’s approval. 60 runs / month 96.2% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Agent Library URL: https://www.aginies.com/en/agents Language: en Summary: 77 live agent scenarios across banking, fintech, insurance, retail, telecom, healthcare, manufacturing, logistics, energy, government, defence, IoT, software delivery and enterprise functions. Every agent is simulated with its supervisor, specialised sub-agents, systems and approval points. Agent library # Enterprise agent scenarios from production, as live simulations. 77 agents, each with its supervisor, specialised sub-agents, the systems it touches, its approval rule and its cost. Collected from global patterns, rebuilt on Aginies primitives. Pick a sector, watch the agent work, compare with your own scenario. 77live scenarios 14sectors 261specialist agents 200systems connected 73approval rules 3,320,970simulated runs / month Banking ### Dialogue banking: voice and text transaction assistant Open agent → [Dialogue banking: voice and text transaction assistant] Audit trail00:00.00 - ◦App chat: "pay my electricity bill, raise my limit to 40k, I don't recognise yesterday's 1,250" - ∿Three intents separated: bill payment · limit increase · transaction dispute; order and dependencies set run #41,289,242elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Limit increase > 60%, dispute amount > 5,000 or transfer to a new payee → operations approval Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus BankingHITL ### Dialogue banking: voice and text transaction assistant The customer talks or types; bill, limit, dispute and transfer are completed in core banking in one session. 186,000 runs / month 94.7% success 5 agents Fintech & PaymentsHITL ### Merchant operations assistant: POS, settlement, refund and limit The merchant writes on WhatsApp; POS fault, settlement query, refund and limit increase are resolved in one dialogue. 64,000 runs / month 93.8% success 5 agents InsuranceHITL ### End-to-end motor claim: from notice to payment Voice notice, photo assessment, repair-shop assignment and payment order in one flow; the adjuster steps in only above the threshold. 21,000 runs / month 91.2% success 5 agents Retail & E-commerceHITL ### Outfit stylist: a complete look by size, stock, budget and style The customer sends a photo and a sentence; the agent builds three looks that are in stock, fit and within budget, and adds the chosen one to the cart. 142,000 runs / month 96.1% success 5 agents TelecomHITL ### Subscriber care assistant: bill, plan, fault and contract The subscriber calls or writes; bill dispute, plan change and line fault are resolved in BSS and network systems in one session. 410,000 runs / month 94.3% success 5 agents Software DeliveryHITL ### Legacy module modernisation: analysis, tests, rewrite, PR Starts from a Jira epic; the old module is read, its behaviour pinned with tests, rewritten and opened as a PR with evidence. 1,900 runs / month 90.4% success 5 agents HealthcareHITL ### Patient journey coordinator: appointment, preparation, approval and results The patient writes or calls; appointment, preparation instructions, insurer pre-approval and result notice come together in one coordinator. 58,000 runs / month 95.1% success 5 agents Manufacturing & AutomotiveHITL ### Line stoppage response: from alarm to work order and shift report An MES alarm lands; root cause found from history, spare part checked, work order and plan update ready within minutes. 3,100 runs / month 92.2% success 5 agents LogisticsHITL ### Control tower: delay prediction, rerouting, customer notice, SLA The telematics signal catches the delay early; alternative route and carrier chosen, customer informed, SLA compensation computed. 96,000 runs / month 93.4% success 5 agents Energy & UtilitiesHITL ### Utility customer assistant: bill, outage, consumption and instalments in one dialogue The customer calls or writes; the high bill is explained, outage ETA given, meter data shown, an instalment plan set up. 520,000 runs / month 94.6% success 5 agents Government & Public SectorHITL ### Citizen services assistant: application, documents, appointment and status The citizen writes or calls; eligibility checked, documents read, application opened, appointment given, every step traceable. 230,000 runs / month 93.8% success 5 agents Enterprise FunctionsHITL ### Procure-to-pay desk: request, quotes, contract and invoice in one place The request arrives by email; quotes compared, contract clauses checked, PO opened, invoice three-way matched and payment scheduled. 14,000 runs / month 92.7% success 5 agents Defence & AerospaceHITL ### Technical publication assistant: maintenance manual, parts, procedure and revision control The technician asks in the hangar; the right-revision procedure, part number and safety note arrive with clearance checks, every access traced. 9,800 runs / month 95.7% success 5 agents IoT & Connected DevicesHITL ### Field device support assistant: telemetry, remote diagnosis, OTA and technician The customer says "the device is dead"; telemetry read, remote diagnosis made, update pushed, technician dispatched with the right part if needed. 37,000 runs / month 94.1% success 5 agents BankingHITL ### Credit card limit increase The customer types in the app or talks to the call centre; the decision takes seconds, and anything above threshold goes to a manager. 18,400 runs / month 96.4% success 4 agents BankingHITL ### Digital account opening (KYC) ID, liveness and sanctions checks in one flow; clean applications open an account in minutes, no branch visit. 42,000 runs / month 93.8% success 5 agents BankingHITL ### Card fraud alert triage The card is blocked in seconds and the customer asked on WhatsApp; the reply releases or cancels the card. 65,000 runs / month 97.1% success 4 agents BankingHITL ### Chargeback dispute handling Dispute captured, evidence assembled and filed with the card scheme; large provisional credits go to approval. 9,800 runs / month 95.2% success 4 agents BankingHITL ### Early-stage collections outreach Past-due accounts are segmented and reminded on the right channel; customers set up their own payment plan. 96,000 runs / month 91.2% success 4 agents BankingHITL ### Mortgage document review Payslips, social security records and appraisal read automatically; income, LTV and missing documents checked. 6,200 runs / month 94.7% success 4 agents BankingHITL ### Voice banking money transfer Customer calls and talks; voiceprint, security question and OTP verify them; transfer completes agent-free. 27,000 runs / month 93.1% success 5 agents BankingHITL ### Credit card debt payment The customer types or talks; debt is fetched, an account chosen, OTP verified and the payment made in the core. 52,000 runs / month 97.8% success 4 agents Fintech & PaymentsHITL ### Merchant KYB onboarding Registry, UBO and sanctions checks in one flow; low-risk merchants start accepting payments the same day. 1,400 runs / month 93.4% success 5 agents Fintech & PaymentsHITL ### AML transaction monitoring alert triage Alerts arrive enriched with a draft narrative; the analyst always decides and the filing goes out in minutes. 8,500 runs / month 95.8% success 5 agents Fintech & PaymentsHITL ### BNPL instant instalment decision Checkout instalment decision in seconds from bureau and open-banking data; large tickets go to underwriting. 54,000 runs / month 97.3% success 5 agents Fintech & PaymentsHITL ### Payment reconciliation exceptions PSP settlement files matched to the ledger; small breaks fix themselves, large ones go to finance approval. 620 runs / month 95.1% success 4 agents InsuranceHITL ### Voice claims intake (FNOL) The customer calls and describes it; policy found by plate, cover checked, claim opened and adjuster assigned. 34,000 runs / month 95.8% success 5 agents InsuranceHITL ### Policy renewal offer Policies re-priced, sharp increases go to approval; customer accepts with one tap and the policy is issued. 22,000 runs / month 94.4% success 4 agents InsuranceHITL ### Underwriting document extraction Medical reports and labs read and assessed against the manual; underwriters only see files above threshold. 4,800 runs / month 94.9% success 4 agents InsuranceHITL ### Motor claim photo assessment Damage detected from photos, estimate built from parts and labour, adjuster only approves; repair slot booked. 15,500 runs / month 93.3% success 4 agents InsuranceHITL ### Subrogation recovery Closed claims scanned, the at-fault party's insurer identified, and the demand sent after counsel approval. 3,400 runs / month 95.5% success 4 agents Retail & E-commerce ### Where is my order? (WISMO) Customer asks, order and carrier events join instantly; on delay a new ETA is given and a trace opened. 110,000 runs / month 98.1% success 4 agents Retail & E-commerceHITL ### Returns and refunds Policy and history checked in seconds, label issued; the refund executes when the carrier scans the parcel. 48,000 runs / month 96.7% success 4 agents Retail & E-commerce ### Product Q&A assistant Size, compatibility and spec questions answered from catalogue and return data; stock and price verified. 120,000 runs / month 98.7% success 4 agents Retail & E-commerceHITL ### Catalogue enrichment (batch) Attributes pulled from supplier PDFs and images, TR/EN copy written; hundreds of SKUs ready in minutes. 900 runs / month 95.8% success 4 agents Retail & E-commerceHITL ### Review response and quality escalation Reviews classified and answered in brand tone; recurring defects joined with return data and escalated to QA. 26,000 runs / month 97.2% success 4 agents TelecomHITL ### Bill dispute resolution Invoice lines and usage explained in seconds; goodwill credit applied by policy, larger amounts need a lead. 88,000 runs / month 96.3% success 4 agents TelecomHITL ### SIM swap fraud check Every SIM swap is risk-scored before provisioning; suspicious requests are held and banks are signalled. 41,000 runs / month 97.7% success 4 agents TelecomHITL ### Network outage communications When an outage ticket opens, affected subscribers are found and the approved message sent before they call. 1,400 runs / month 98.5% success 4 agents TelecomHITL ### Plan upgrade advisor Usage analysed, the right plan offered with a comparison; on acceptance the change provisions instantly. 52,000 runs / month 95.8% success 4 agents Software DeliveryHITL ### Feature delivery pipeline A Jira epic turns into criteria, design, code and tests via agents; humans only review and approve the PR. 1,200 runs / month 91.2% success 5 agents Software DeliveryHITL ### CI/CD release pipeline Build, tests, scans and staging run unattended; production ships only on release-manager approval, via ArgoCD. 950 runs / month 94.7% success 5 agents Software DeliveryHITL ### Incident response and hotfix Alert fires, bug reproduced, patch and regression run ready; on-call approves, post-mortem already drafted. 320 runs / month 90.5% success 5 agents Software DeliveryHITL ### Code review and security gate Agents review, scan and coverage-check every PR; fixes land first, a human approves only clean PRs. 4,200 runs / month 93.6% success 5 agents Software DeliveryHITL ### Backlog grooming and estimation Incoming requests are triaged, clarified and sized; a sprint proposal goes to the product owner for approval. 340 runs / month 92.8% success 4 agents Software DeliveryHITL ### iOS feature implementation (hired developer agent) The Jira story is taken, built in SwiftUI, tested and opened as a PR; review and release decisions stay with the team. 380 runs / month 94.4% success 4 agents Software DeliveryHITL ### Regression test cycle (hired QA agent) Regression runs on every release candidate, new bugs are filed with evidence, a coverage report is written; the “go” decision stays with the release manager. 260 runs / month 95.8% success 4 agents HealthcareHITL ### Prior authorization Payer rules searched, clinical evidence gathered and the packet ready to submit within minutes of the order. 6,200 runs / month 94.1% success 4 agents HealthcareHITL ### Patient intake and triage The patient types or talks; intake and urgency scoring take seconds, and a nurse always makes the final call. 24,500 runs / month 95.8% success 5 agents Healthcare ### Appointment scheduling and no-show reduction High no-show-risk patients are reached on WhatsApp in advance; freed slots are refilled from the waiting list. 62,000 runs / month 98.2% success 4 agents Manufacturing & AutomotiveHITL ### Supplier invoice three-way match Invoice, PO and goods receipt matched in SAP automatically; only out-of-tolerance lines reach a buyer. 21,000 runs / month 97.1% success 4 agents Manufacturing & AutomotiveHITL ### Maintenance ticket triage Sensor alarm and technician voice note become one work order, spare parts and a safe stop window included. 3,400 runs / month 95.2% success 5 agents Manufacturing & AutomotiveHITL ### Warranty claim processing Dealer claims validated for coverage, parts and labour in minutes; big amounts go to the regional manager. 8,600 runs / month 96.3% success 4 agents LogisticsHITL ### Shipment exception handling When a delay or damage lands, impact is costed, the carrier rebooked and the customer hears before they ask. 12,800 runs / month 95.5% success 4 agents LogisticsHITL ### Customs document preparation HS codes, invoice and packing list prepared in minutes; only low-confidence lines reach a customs specialist. 4,300 runs / month 94.7% success 4 agents Energy & UtilitiesHITL ### Outage report handling Reports are clustered by feeder, one OSS ticket is raised and everyone affected hears before the crew leaves. 38,000 runs / month 97.7% success 5 agents Energy & UtilitiesHITL ### Meter reading dispute Customer sends a meter photo; the index is read by vision, the estimated bill fixed, big credits approved. 9,200 runs / month 94.9% success 4 agents Energy & UtilitiesHITL ### Rooftop solar connection application Application documents are read, transformer capacity and regulation checked, gaps reported the same day; the connection opinion belongs to the engineer. 5,400 runs / month 95.3% success 5 agents Energy & UtilitiesHITL ### Demand response event Under grid stress eligible consumers are selected, the offer sent by channel, participation measured and settlement computed. 140 runs / month 96.2% success 4 agents Government & Public SectorHITL ### Citizen request routing e-Government requests classified in seconds, given a legal deadline and routed to the right unit with an SLA. 54,000 runs / month 98.1% success 4 agents Government & Public SectorHITL ### Permit application check Missing documents and regulatory fit reported in minutes; the licensing officer always makes the decision. 3,100 runs / month 95.8% success 4 agents Government & Public SectorHITL ### Public procurement bid file pre-check Bid files are compared with the specification, missing and non-conforming documents listed; the evaluation decision belongs to the commission. 380 runs / month 94.8% success 4 agents Government & Public SectorHITL ### Social assistance eligibility pre-assessment The application is verified against agency records, eligibility scored, gaps reported to the citizen; the decision belongs to the case officer. 21,000 runs / month 96.6% success 4 agents Enterprise Functions ### Email classification and attachment check Every e-mail in the shared mailbox is classified, attachments checked against a list, gaps asked for at once. 86,000 runs / month 97.6% success 4 agents Enterprise FunctionsHITL ### Call centre quality monitoring Every call scored against the rubric, not a sample; compliance breaches and coaching flags ready by morning. 90,000 runs / month 98.8% success 4 agents Enterprise FunctionsHITL ### Contract clause extraction Clauses are extracted and compared with the legal playbook; deviations reach counsel with redline suggestions. 1,400 runs / month 95.1% success 4 agents Enterprise FunctionsHITL ### Employee onboarding Accounts, kit and training plan ready the moment HR creates the hire; off-template access goes to the manager. 1,900 runs / month 97.2% success 5 agents Enterprise FunctionsHITL ### IT helpdesk L1 Password resets and access requests solved in chat; access to finance groups asks the owner via a Teams card. 47,000 runs / month 98.5% success 4 agents Enterprise FunctionsHITL ### Accounts payable inbox Every invoice in the ap@ mailbox is captured, coded and posted to the ERP; only exceptions reach an analyst. 28,000 runs / month 96.8% success 4 agents Enterprise FunctionsHITL ### Sales lead qualification Inbound leads enriched, scored and in the CRM within seconds; the SDR starts the call with a context brief. 15,600 runs / month 97.9% success 4 agents Enterprise FunctionsHITL ### Month-end close and reconciliation (hired accountant agent) Invoices are processed, bank and vendor reconciliations done, the close checklist run; entries are posted with the accountant’s approval. 60 runs / month 96.2% success 4 agents Defence & AerospaceHITL ### Export control and end-user screening Part, buyer and end use are classified in one flow; licence need is detected, the decision always stays with the compliance officer. 2,600 runs / month 96.1% success 4 agents Defence & AerospaceHITL ### Platform maintenance (MRO) work order preparation The fault report is matched to technical publications, task card and parts list are produced, the work order opens with planner approval. 4,100 runs / month 95.7% success 5 agents Defence & AerospaceHITL ### Supplier security and compliance assessment Supplier questionnaires and documents are read, compared against facility security requirements, gaps listed; the qualification decision belongs to the security manager. 900 runs / month 95.2% success 4 agents IoT & Connected DevicesHITL ### Device fleet anomaly triage Telemetry anomalies from thousands of devices are clustered, root-cause candidates listed, one incident reaches the right team. 18,000 runs / month 96.9% success 4 agents IoT & Connected DevicesHITL ### Staged OTA update rollout Firmware rolls out in 1% → 10% → 100% stages; a health gate at every stage, human approval for critical fleets. 320 runs / month 98.1% success 4 agents IoT & Connected DevicesHITL ### Cold-chain excursion response A temperature excursion is assessed instantly, product risk computed, a field crew dispatched, the quality decision recorded. 2,700 runs / month 97.4% success 4 agents Simulations are derived from real agent definitions; figures are representative. Real metrics from our reference customers are shared in discovery sessions. ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Accounts payable inbox URL: https://www.aginies.com/en/agents/accounts-payable-inbox Language: en Summary: Every invoice in the ap@ mailbox is captured, coded and posted to the ERP; only exceptions reach an analyst. E-mail with a PDF invoice in the ap@ mailbox → Invoice posted in the ERP, archived, supplier informed, exceptions queued Enterprise Functionsemailbatch # Accounts payable inbox Every invoice in the ap@ mailbox is captured, coded and posted to the ERP; only exceptions reach an analyst. In short Accounts payable inbox is an agent scenario in the Enterprise Functions sector running on the email, batch channels. Trigger: E-mail with a PDF invoice in the ap@ mailbox. The AP Supervisor coordinates 3 specialist agents (Capture Agent, Coding Agent, Posting Agent) across 4 systems (Exchange / M365 Mail, Oracle EBS, Document Store (S3), Vendor Master / Procurement); of 9 steps, 2 write to systems and 2 are checks or approvals. Outcome: Invoice posted in the ERP, archived, supplier informed, exceptions queued. Approval rule: Non-PO invoice > 25,000 TL → cost centre owner approval. Enterprise Functions ### Accounts payable inbox [Accounts payable inbox] run #41,287,996elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Non-PO invoice > 25,000 TL → cost centre owner approval Agents in this run AP Supervisorsupervisorplans, routes, approves Capture Agentaginies-konrul-1 · on-premExtracts fields with OCR and catches duplicates Coding Agentaginies-tulpar-4 · on-premSuggests GL account, cost centre and VAT coding Posting Agentaginies-talay-2 · on-premPosts to the ERP, archives and notifies Audit trail00:00.00 - ◦ap@ mailbox: 312 e-mails overnight; 287 PDF invoices, 25 statements - ∿Invoice INV-77120 from Yıldız Lojistik: 42,600 TL + 18% VAT, no PO (services) - ✓Duplicate check: no number/amount/date match in 180 days; IBAN matches vendor master - ⚙Vendor 20418 active, payment term 45 days, contract "freight 2026" on file - ∿GL 6300 (freight) · cost centre 2210 · VAT 18% deductible · confidence 0.91 - ⏸Non-PO invoice > 25,000 TL → cost centre owner approved the same day - ⇢Invoice posted (AP-2026-118440), due 23 Oct, payment run 21 Oct - ⇢PDF and extraction JSON archived, 10-year retention flag set - ➤Supplier confirmation sent; 14 exceptions of 287 routed to the AP analyst queue - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger E-mail with a PDF invoice in the ap@ mailbox Outcome Invoice posted in the ERP, archived, supplier informed, exceptions queued ## Agents SUPERVISOR ### AP Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Capture Agent Extracts fields with OCR and catches duplicates ocr_documentextract_invoice_fieldsdetect_duplicate aginies-tulpar-4 · on-prem ### Coding Agent Suggests GL account, cost centre and VAT coding get_vendor_mastersuggest_gl_accountget_cost_centrecheck_vat aginies-talay-2 · on-prem ### Posting Agent Posts to the ERP, archives and notifies post_invoicepark_invoicearchive_documentssend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | AP Supervisor | ap@ mailbox: 312 e-mails overnight; 287 PDF invoices, 25 statements | — | 300 | — | | 02 | reason | Capture Agent | Invoice INV-77120 from Yıldız Lojistik: 42,600 TL + 18% VAT, no PO (services) | — | 860 | 680 | | 03 | verify | Capture Agent | Duplicate check: no number/amount/date match in 180 days; IBAN matches vendor master | — | 380 | 220 | | 04 | tool | Coding Agent | Vendor 20418 active, payment term 45 days, contract "freight 2026" on file | Vendor Master / Procurement | 400 | — | | 05 | reason | Coding Agent | GL 6300 (freight) · cost centre 2210 · VAT 18% deductible · confidence 0.91 | — | 720 | 760 | | 06 | approval | AP Supervisor | Non-PO invoice > 25,000 TL → cost centre owner approved the same day | — | 2,600 | — | | 07 | write | Posting Agent | Invoice posted (AP-2026-118440), due 23 Oct, payment run 21 Oct | Oracle EBS | 540 | — | | 08 | write | Posting Agent | PDF and extraction JSON archived, 10-year retention flag set | Document Store (S3) | 260 | — | | 09 | notify | Posting Agent | Supplier confirmation sent; 14 exceptions of 287 routed to the AP analyst queue | Exchange / M365 Mail | 240 | — | Simulation metrics (estimated) 28,000runs / month 96.8%success 6.3 sp50 $0.0026est. cost / run Approval rule Non-PO invoice > 25,000 TL → cost centre owner approval Step profile ingest · 1reason · 2tool · 1verify · 1approval · 1write · 2notify · 1 SYSTEMS - Exchange / M365 Mailemail - Oracle EBSerp - Document Store (S3)storage - Vendor Master / Procurementdb What you need to run it - Mailbox access (Microsoft 365 or Google Workspace) - Posting and query API access to ERP/MES - Read access to the document store and the document types - Read access to the database or vector store; a data owner - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors AP inbox automation at global shared-service centres Packages that include this scenario Invoice & document extraction →Email classification & attachment verification → Related agents Enterprise FunctionsHITL ### Procure-to-pay desk: request, quotes, contract and invoice in one place The request arrives by email; quotes compared, contract clauses checked, PO opened, invoice three-way matched and payment scheduled. 14,000 runs / month 92.7% success 5 agents Enterprise Functions ### Email classification and attachment check Every e-mail in the shared mailbox is classified, attachments checked against a list, gaps asked for at once. 86,000 runs / month 97.6% success 4 agents Enterprise FunctionsHITL ### Call centre quality monitoring Every call scored against the rubric, not a sample; compliance breaches and coaching flags ready by morning. 90,000 runs / month 98.8% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # AML transaction monitoring alert triage URL: https://www.aginies.com/en/agents/aml-transaction-monitoring Language: en Summary: Alerts arrive enriched with a draft narrative; the analyst always decides and the filing goes out in minutes. Rule alert from the monitoring engine (structuring / rapid pass-through) → Alert decided, case closed or STR filed with MASAK, narrative and evidence trail stored Fintech & Paymentsapibatch # AML transaction monitoring alert triage Alerts arrive enriched with a draft narrative; the analyst always decides and the filing goes out in minutes. In short AML transaction monitoring alert triage is an agent scenario in the Fintech & Payments sector running on the api, batch channels. Trigger: Rule alert from the monitoring engine (structuring / rapid pass-through). The AML Supervisor coordinates 4 specialist agents (Enrichment Agent, Screening Agent, Narrative Agent, Filing Agent) across 4 systems (Core Ledger, Sanctions Screening, Case Management, MASAK Reporting Portal); of 10 steps, 2 write to systems and 2 are checks or approvals. Outcome: Alert decided, case closed or STR filed with MASAK, narrative and evidence trail stored. Approval rule: Every alert → compliance analyst decision mandatory; the agent never files an STR or closes a case alone. Fintech & Payments ### AML transaction monitoring alert triage [AML transaction monitoring alert triage] run #41,289,252elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Every alert → compliance analyst decision mandatory; the agent never files an STR or closes a case alone Agents in this run AML Supervisorsupervisorplans, routes, approves Enrichment Agentaginies-talay-2 · on-premGathers customer profile, 90-day transactions and counterparty data Screening Agentaginies-sunkar-1 · on-premScreens counterparties against sanctions/PEP lists, finds prior cases Narrative Agentaginies-tulpar-4 · on-premAssesses the typology, drafts the STR narrative Filing Agentaginies-talay-2 · on-premUpdates the case, files the report with MASAK Audit trail00:00.00 - ◦Alert #TM-51290: 9 cash deposits in 6 days totalling 1,140,000, rule R-14 - ⚙Customer profile and 90-day history pulled: 47 transactions, 3 counterparties - ⌘Plan: counterparty screening → typology → draft narrative → analyst decision → filing - ⚙3 counterparties screened: no sanctions hit, 1 PEP link (relative) - ⌕Prior cases: 1 alert on the same customer closed in 2024 (false positive) - ∿Typology: sub-threshold splitting, 4× declared income → suspicion score 0.82 - ✓Draft narrative prepared with 5 evidence references, MASAK format check passed - ⏸Mandatory analyst decision: compliance analyst edited the narrative, chose "file STR" - ⇢STR filed on the MASAK portal, reference #STR-2026-08812 - ⇢Case #AML-7734 set to "filed", evidence pack and decision trail attached - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Rule alert from the monitoring engine (structuring / rapid pass-through) Outcome Alert decided, case closed or STR filed with MASAK, narrative and evidence trail stored ## Agents SUPERVISOR ### AML Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Enrichment Agent Gathers customer profile, 90-day transactions and counterparty data get_alertget_customer_profileget_transactionsget_counterparties aginies-sunkar-1 · on-prem ### Screening Agent Screens counterparties against sanctions/PEP lists, finds prior cases screen_sanctionsscreen_pepsearch_prior_cases aginies-tulpar-4 · on-prem ### Narrative Agent Assesses the typology, drafts the STR narrative score_typologydraft_narrativesummarise_evidence aginies-talay-2 · on-prem ### Filing Agent Updates the case, files the report with MASAK update_casefile_strnotify_analyst ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | AML Supervisor | Alert #TM-51290: 9 cash deposits in 6 days totalling 1,140,000, rule R-14 | — | 180 | — | | 02 | tool | Enrichment Agent | Customer profile and 90-day history pulled: 47 transactions, 3 counterparties | Core Ledger | 520 | — | | 03 | plan | AML Supervisor | Plan: counterparty screening → typology → draft narrative → analyst decision → filing | — | 300 | 220 | | 04 | tool | Screening Agent | 3 counterparties screened: no sanctions hit, 1 PEP link (relative) | Sanctions Screening | 560 | — | | 05 | retrieve | Screening Agent | Prior cases: 1 alert on the same customer closed in 2024 (false positive) | Case Management | 410 | — | | 06 | reason | Narrative Agent | Typology: sub-threshold splitting, 4× declared income → suspicion score 0.82 | — | 1,200 | 1,600 | | 07 | verify | Narrative Agent | Draft narrative prepared with 5 evidence references, MASAK format check passed | — | 900 | 1,100 | | 08 | approval | AML Supervisor | Mandatory analyst decision: compliance analyst edited the narrative, chose "file STR" | — | 3,800 | — | | 09 | write | Filing Agent | STR filed on the MASAK portal, reference #STR-2026-08812 | MASAK Reporting Portal | 740 | — | | 10 | write | Filing Agent | Case #AML-7734 set to "filed", evidence pack and decision trail attached | Case Management | 460 | — | Simulation metrics (estimated) 8,500runs / month 95.8%success 9.2 sp50 $0.0093est. cost / run Approval rule Every alert → compliance analyst decision mandatory; the agent never files an STR or closes a case alone Step profile ingest · 1plan · 1retrieve · 1reason · 1tool · 2verify · 1approval · 1write · 2 SYSTEMS - Core Ledgercore - Sanctions Screeningapi - Case Managementticketing - MASAK Reporting Portalapi What you need to run it - Read/write API or MCP server and a service account for the core system - Credentials and a test environment for the enterprise APIs involved - API access and assignment rules for the ticketing or work-order system - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors AML alert enrichment and narrative drafting at global banks and fintechs, with the analyst in the loop Packages that include this scenario Merchant and payment operations → Sector: Fintech & Payments → Related agents Fintech & PaymentsHITL ### Merchant operations assistant: POS, settlement, refund and limit The merchant writes on WhatsApp; POS fault, settlement query, refund and limit increase are resolved in one dialogue. 64,000 runs / month 93.8% success 5 agents Fintech & PaymentsHITL ### Merchant KYB onboarding Registry, UBO and sanctions checks in one flow; low-risk merchants start accepting payments the same day. 1,400 runs / month 93.4% success 5 agents Fintech & PaymentsHITL ### BNPL instant instalment decision Checkout instalment decision in seconds from bureau and open-banking data; large tickets go to underwriting. 54,000 runs / month 97.3% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Appointment scheduling and no-show reduction URL: https://www.aginies.com/en/agents/appointment-scheduling-noshow Language: en Summary: High no-show-risk patients are reached on WhatsApp in advance; freed slots are refilled from the waiting list. Nightly batch: appointment list for the next 72 hours → Appointment moved or confirmed, old slot released to the waiting list, reminder scheduled Healthcarewhatsappsmsbatch # Appointment scheduling and no-show reduction High no-show-risk patients are reached on WhatsApp in advance; freed slots are refilled from the waiting list. In short Appointment scheduling and no-show reduction is an agent scenario in the Healthcare sector running on the whatsapp, sms, batch channels. Trigger: Nightly batch: appointment list for the next 72 hours. The Appointment Supervisor coordinates 3 specialist agents (No-show Risk Agent, Dialogue Agent, Booking Agent) across 3 systems (Hospital Information System (HIS), WhatsApp Business API, SMS Gateway); of 9 steps, 1 write to systems and 1 are checks or approvals. Outcome: Appointment moved or confirmed, old slot released to the waiting list, reminder scheduled. Healthcare ### Appointment scheduling and no-show reduction [Appointment scheduling and no-show reduction] run #41,290,281elapsed 0 mstokens 0est. cost $0.000status — Agents in this run Appointment Supervisorsupervisorplans, routes, approves No-show Risk Agentaginies-sunkar-1 · on-premPredicts no-show risk from history Dialogue Agentaginies-sunkar-1 · on-premTalks on WhatsApp and understands reschedule intent Booking Agentaginies-talay-2 · on-premFinds a free slot in the HIS and moves the appointment Audit trail00:00.00 - ◦Nightly batch: 1,240 appointments in the next 72 hours loaded - ⚙History pulled: patient P-77120, 2 no-shows in the last 12 months - ∿No-show risk 0.62 (threshold 0.40) → proactive confirmation channel selected - ➤WhatsApp sent: "Dr Aksoy, Thu 10:20 — confirm / reschedule?" - ∿Reply: "Thursday is bad, next week?" → intent: reschedule, preference: next week AM - ⚙3 free slots found: Mon 09:00, Tue 09:40, Wed 11:20 (same physician) - ✓Rule check: same physician, ≤ 14 days delay, no prep conflict → OK - ⇢Appointment moved to Tue 09:40, original slot released to the waiting list - ➤Confirmation SMS sent, 24 h reminder scheduled, 12 slots refilled from the list - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Nightly batch: appointment list for the next 72 hours Outcome Appointment moved or confirmed, old slot released to the waiting list, reminder scheduled ## Agents SUPERVISOR ### Appointment Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### No-show Risk Agent Predicts no-show risk from history get_appointment_historyscore_noshow_risk aginies-sunkar-1 · on-prem ### Dialogue Agent Talks on WhatsApp and understands reschedule intent send_whatsappparse_replypropose_slots aginies-talay-2 · on-prem ### Booking Agent Finds a free slot in the HIS and moves the appointment find_free_slotsreschedule_appointmentrelease_slotsend_sms ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Appointment Supervisor | Nightly batch: 1,240 appointments in the next 72 hours loaded | — | 300 | — | | 02 | tool | No-show Risk Agent | History pulled: patient P-77120, 2 no-shows in the last 12 months | Hospital Information System (HIS) | 420 | — | | 03 | reason | No-show Risk Agent | No-show risk 0.62 (threshold 0.40) → proactive confirmation channel selected | — | 540 | 380 | | 04 | notify | Dialogue Agent | WhatsApp sent: "Dr Aksoy, Thu 10:20 — confirm / reschedule?" | WhatsApp Business API | 260 | — | | 05 | reason | Dialogue Agent | Reply: "Thursday is bad, next week?" → intent: reschedule, preference: next week AM | — | 620 | 410 | | 06 | tool | Booking Agent | 3 free slots found: Mon 09:00, Tue 09:40, Wed 11:20 (same physician) | Hospital Information System (HIS) | 480 | — | | 07 | verify | Booking Agent | Rule check: same physician, ≤ 14 days delay, no prep conflict → OK | — | 380 | 240 | | 08 | write | Booking Agent | Appointment moved to Tue 09:40, original slot released to the waiting list | Hospital Information System (HIS) | 520 | — | | 09 | notify | Booking Agent | Confirmation SMS sent, 24 h reminder scheduled, 12 slots refilled from the list | SMS Gateway | 240 | — | Simulation metrics (estimated) 62,000runs / month 98.2%success 3.8 sp50 $0.0008est. cost / run Step profile ingest · 1reason · 2tool · 2verify · 1write · 1notify · 2 SYSTEMS - Hospital Information System (HIS)core - WhatsApp Business APIchannel - SMS Gatewaychannel What you need to run it - Read/write API or MCP server and a service account for the core system - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors WhatsApp-based reminder and no-show reduction programmes at large hospital networks Packages that include this scenario Healthcare operations & clinical guidance → Sector: Healthcare → Related agents HealthcareHITL ### Patient journey coordinator: appointment, preparation, approval and results The patient writes or calls; appointment, preparation instructions, insurer pre-approval and result notice come together in one coordinator. 58,000 runs / month 95.1% success 5 agents HealthcareHITL ### Prior authorization Payer rules searched, clinical evidence gathered and the packet ready to submit within minutes of the order. 6,200 runs / month 94.1% success 4 agents HealthcareHITL ### Patient intake and triage The patient types or talks; intake and urgency scoring take seconds, and a nurse always makes the final call. 24,500 runs / month 95.8% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Backlog grooming and estimation URL: https://www.aginies.com/en/agents/backlog-grooming-estimation Language: en Summary: Incoming requests are triaged, clarified and sized; a sprint proposal goes to the product owner for approval. Weekly batch: new requests in the "Triage" column in Jira → Requests sized and prioritised, sprint 42 proposal approved, Jira and Slack up to date Software Deliverybatchwebapi # Backlog grooming and estimation Incoming requests are triaged, clarified and sized; a sprint proposal goes to the product owner for approval. In short Backlog grooming and estimation is an agent scenario in the Software Delivery sector running on the batch, web, api channels. Trigger: Weekly batch: new requests in the "Triage" column in Jira. The Planning Supervisor coordinates 3 specialist agents (Project Manager Agent, Analyst Agent, Architect Agent) across 4 systems (Jira, Confluence, GitHub, Slack / Teams); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: Requests sized and prioritised, sprint 42 proposal approved, Jira and Slack up to date. Approval rule: Sprint scope → product owner approval; agents only propose, never commit items to a sprint. Software Delivery ### Backlog grooming and estimation [Backlog grooming and estimation] run #41,289,835elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Sprint scope → product owner approval; agents only propose, never commit items to a sprint Agents in this run Planning Supervisorsupervisorplans, routes, approves Project Manager Agentaginies-tulpar-4 · on-premTriages requests, merges duplicates, prepares the sprint proposal Analyst Agentaginies-tulpar-4 · on-premClarifies vague requests, writes acceptance criteria Architect Agentaginies-tulpar-4 · on-premEstimates complexity and risk, proposes story points Audit trail00:00.00 - ◦Weekly triage: 27 new requests (14 features, 9 bugs, 4 tech debt) - ∿3 duplicates merged (PROJ-4901 ↔ 4887), 24 requests ranked; velocity 46 points/sprint - ⌘Plan: clarify vague requests → estimate → propose sprint → product owner approval - ⌕PRD and prior decisions read; clarifying questions drafted for 6 requests - ➤6 questions asked in #product, 5 answered; PROJ-4910 parked - ⚙Affected modules and change frequency pulled: 4 requests touch checkout - ∿Estimate: 23 requests = 118 points; 2 above 13 points → split proposed - ✓Sprint 42 proposal: 11 requests, 44 points (velocity 46), no dependency clash - ⏸Product owner reviewed the proposal, swapped 1 request, approved sprint 42 - ⇢Sprint 42 created, 11 requests moved with points, the rest re-prioritised - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Weekly batch: new requests in the "Triage" column in Jira Outcome Requests sized and prioritised, sprint 42 proposal approved, Jira and Slack up to date ## Agents SUPERVISOR ### Planning Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-tulpar-4 · on-prem ### Project Manager Agent Triages requests, merges duplicates, prepares the sprint proposal list_jira_issuesdetect_duplicatesrank_by_valuepropose_sprintupdate_jira_issue aginies-tulpar-4 · on-prem ### Analyst Agent Clarifies vague requests, writes acceptance criteria search_confluenceask_clarifying_questionwrite_acceptance_criteria aginies-tulpar-4 · on-prem ### Architect Agent Estimates complexity and risk, proposes story points search_codebaseestimate_complexityflag_dependencies ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Planning Supervisor | Weekly triage: 27 new requests (14 features, 9 bugs, 4 tech debt) | — | 240 | — | | 02 | reason | Project Manager Agent | 3 duplicates merged (PROJ-4901 ↔ 4887), 24 requests ranked; velocity 46 points/sprint | — | 1,800 | 3,600 | | 03 | plan | Planning Supervisor | Plan: clarify vague requests → estimate → propose sprint → product owner approval | — | 280 | 200 | | 04 | retrieve | Analyst Agent | PRD and prior decisions read; clarifying questions drafted for 6 requests | Confluence | 1,400 | 2,800 | | 05 | notify | Analyst Agent | 6 questions asked in #product, 5 answered; PROJ-4910 parked | Slack / Teams | 3,200 | 900 | | 06 | tool | Architect Agent | Affected modules and change frequency pulled: 4 requests touch checkout | GitHub | 620 | — | | 07 | reason | Architect Agent | Estimate: 23 requests = 118 points; 2 above 13 points → split proposed | — | 2,000 | 3,900 | | 08 | verify | Project Manager Agent | Sprint 42 proposal: 11 requests, 44 points (velocity 46), no dependency clash | — | 900 | 1,100 | | 09 | approval | Planning Supervisor | Product owner reviewed the proposal, swapped 1 request, approved sprint 42 | — | 4,000 | — | | 10 | write | Project Manager Agent | Sprint 42 created, 11 requests moved with points, the rest re-prioritised | Jira | 560 | — | Simulation metrics (estimated) 340runs / month 92.8%success 14.6 sp50 $0.0189est. cost / run Approval rule Sprint scope → product owner approval; agents only propose, never commit items to a sprint Step profile ingest · 1plan · 1retrieve · 1reason · 2tool · 1verify · 1approval · 1write · 1notify · 1 SYSTEMS - Jiraticketing - Confluencestorage - GitHubapi - Slack / Teamschannel What you need to run it - API access and assignment rules for the ticketing or work-order system - Read access to the document store and the document types - Credentials and a test environment for the enterprise APIs involved - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors AI-assisted backlog refinement at large software organisations scaling agile delivery Packages that include this scenario Agent outsourcing (agent workforce) → Sector: Software Delivery → Related agents Software DeliveryHITL ### Legacy module modernisation: analysis, tests, rewrite, PR Starts from a Jira epic; the old module is read, its behaviour pinned with tests, rewritten and opened as a PR with evidence. 1,900 runs / month 90.4% success 5 agents Software DeliveryHITL ### Feature delivery pipeline A Jira epic turns into criteria, design, code and tests via agents; humans only review and approve the PR. 1,200 runs / month 91.2% success 5 agents Software DeliveryHITL ### CI/CD release pipeline Build, tests, scans and staging run unattended; production ships only on release-manager approval, via ArgoCD. 950 runs / month 94.7% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Bill dispute resolution URL: https://www.aginies.com/en/agents/bill-dispute-resolution Language: en Summary: Invoice lines and usage explained in seconds; goodwill credit applied by policy, larger amounts need a lead. "Why is my bill so high?" on app, WhatsApp or a call → Customer understands the charges, credit applied, case closed Telecomappvoicewhatsapp # Bill dispute resolution Invoice lines and usage explained in seconds; goodwill credit applied by policy, larger amounts need a lead. In short Bill dispute resolution is an agent scenario in the Telecom sector running on the app, voice, whatsapp channels. Trigger: "Why is my bill so high?" on app, WhatsApp or a call. The Billing Supervisor coordinates 3 specialist agents (Intent Agent, Bill Analysis Agent, Adjustment Agent) across 3 systems (Billing (Amdocs), Usage Mediation / CDR Store, SMS Gateway); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: Customer understands the charges, credit applied, case closed. Approval rule: Goodwill credit > 75 → team lead approval. Telecom ### Bill dispute resolution [Bill dispute resolution] run #41,288,546elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Goodwill credit > 75 → team lead approval Agents in this run Billing Supervisorsupervisorplans, routes, approves Intent Agentaginies-sunkar-1 · on-premUnderstands the dispute and period, verifies the line Bill Analysis Agentaginies-tulpar-4 · on-premExplains invoice lines against usage records and tariff Adjustment Agentaginies-talay-2 · on-premApplies goodwill policy and posts the credit Audit trail00:00.00 - ◦App: "Why is my bill 640 this month? It's normally 420" - ∿Intent: bill dispute · period: August · line 0532 *** 4471 verified - ⚙Invoice #INV-2608-99120: 640 = plan 420 + roaming 185 + SMS 35 - ⚙CDRs: 380 MB of roaming data in Bulgaria, 14–16 Aug - ⌘Plan: explain lines → goodwill policy → approval → credit → notify - ∿Roaming charge matches the tariff; 4-year customer, first dispute - ✓Goodwill policy: first dispute + loyal customer → 50% credit (92.50), above threshold - ⏸Credit > 75 → team lead approved the 92.50 credit - ⇢92.50 credit applied to the next invoice, case #BD-40211 closed - ➤SMS: explanation of charges, credit and a roaming add-on suggestion - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger "Why is my bill so high?" on app, WhatsApp or a call Outcome Customer understands the charges, credit applied, case closed ## Agents SUPERVISOR ### Billing Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Intent Agent Understands the dispute and period, verifies the line detect_intentextract_bill_periodverify_customer aginies-tulpar-4 · on-prem ### Bill Analysis Agent Explains invoice lines against usage records and tariff get_invoiceget_usage_detailget_tariffexplain_charges aginies-talay-2 · on-prem ### Adjustment Agent Applies goodwill policy and posts the credit check_goodwill_policyapply_creditclose_casesend_sms ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Billing Supervisor | App: "Why is my bill 640 this month? It's normally 420" | — | 180 | — | | 02 | reason | Intent Agent | Intent: bill dispute · period: August · line 0532 *** 4471 verified | — | 400 | 250 | | 03 | tool | Bill Analysis Agent | Invoice #INV-2608-99120: 640 = plan 420 + roaming 185 + SMS 35 | Billing (Amdocs) | 420 | — | | 04 | tool | Bill Analysis Agent | CDRs: 380 MB of roaming data in Bulgaria, 14–16 Aug | Usage Mediation / CDR Store | 480 | — | | 05 | plan | Billing Supervisor | Plan: explain lines → goodwill policy → approval → credit → notify | — | 240 | 170 | | 06 | reason | Bill Analysis Agent | Roaming charge matches the tariff; 4-year customer, first dispute | — | 760 | 820 | | 07 | verify | Adjustment Agent | Goodwill policy: first dispute + loyal customer → 50% credit (92.50), above threshold | — | 440 | 320 | | 08 | approval | Billing Supervisor | Credit > 75 → team lead approved the 92.50 credit | — | 2,300 | — | | 09 | write | Adjustment Agent | 92.50 credit applied to the next invoice, case #BD-40211 closed | Billing (Amdocs) | 480 | — | | 10 | notify | Adjustment Agent | SMS: explanation of charges, credit and a roaming add-on suggestion | SMS Gateway | 260 | — | Simulation metrics (estimated) 88,000runs / month 96.3%success 6 sp50 $0.0009est. cost / run Approval rule Goodwill credit > 75 → team lead approval Step profile ingest · 1plan · 1reason · 2tool · 2verify · 1approval · 1write · 1notify · 1 SYSTEMS - Billing (Amdocs)core - Usage Mediation / CDR Storedb - SMS Gatewaychannel What you need to run it - Read/write API or MCP server and a service account for the core system - Read access to the database or vector store; a data owner - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the "explain my bill" assistant pattern at large operators, extended to apply credits Packages that include this scenario Call-centre quality monitoring →Subscriber care and network support → Sector: Telecom → Related agents TelecomHITL ### Subscriber care assistant: bill, plan, fault and contract The subscriber calls or writes; bill dispute, plan change and line fault are resolved in BSS and network systems in one session. 410,000 runs / month 94.3% success 5 agents TelecomHITL ### SIM swap fraud check Every SIM swap is risk-scored before provisioning; suspicious requests are held and banks are signalled. 41,000 runs / month 97.7% success 4 agents TelecomHITL ### Network outage communications When an outage ticket opens, affected subscribers are found and the approved message sent before they call. 1,400 runs / month 98.5% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # BNPL instant instalment decision URL: https://www.aginies.com/en/agents/bnpl-credit-decision Language: en Summary: Checkout instalment decision in seconds from bureau and open-banking data; large tickets go to underwriting. "Pay in 3 instalments" selected at checkout (API) → Instalment plan posted to the ledger, merchant payout committed, customer informed Fintech & Paymentsapiappsms # BNPL instant instalment decision Checkout instalment decision in seconds from bureau and open-banking data; large tickets go to underwriting. In short BNPL instant instalment decision is an agent scenario in the Fintech & Payments sector running on the api, app, sms channels. Trigger: "Pay in 3 instalments" selected at checkout (API). The Credit Decision Supervisor coordinates 4 specialist agents (Intake Agent, Bureau Agent, Decision Agent, Booking Agent) across 5 systems (Merchant Checkout API, Credit Bureau (KKB / Findeks), Open Banking API, Core Ledger, SMS Gateway); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: Instalment plan posted to the ledger, merchant payout committed, customer informed. Approval rule: Basket > 15,000 or bureau score < 1,200 → underwriter approval. Fintech & Payments ### BNPL instant instalment decision [BNPL instant instalment decision] run #41,287,663elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Basket > 15,000 or bureau score < 1,200 → underwriter approval Agents in this run Credit Decision Supervisorsupervisorplans, routes, approves Intake Agentaginies-sunkar-1 · on-premChecks basket, customer and fraud signals Bureau Agentaginies-talay-2 · on-premPulls the bureau score and open-banking cash flow Decision Agentaginies-tulpar-4 · on-premDecides and sets the limit from affordability and policy Booking Agentaginies-talay-2 · on-premPosts the plan to the ledger, notifies merchant and customer Audit trail00:00.00 - ◦Basket 18,750, 3-instalment request; customer E.Y., device and email known - ⚙Basket and merchant data pulled: MCC 5732, merchant chargeback rate 0.4% - ⌘Plan: bureau score → open banking → affordability → policy → booking - ⚙Bureau score 1,480, 2 open loans, no delinquency, 1 enquiry in 30 days - ⚙Open banking: 3-month average income 62,000, free monthly cash flow 14,200 - ✓Policy v3.0: instalment 6,250 ≤ 50% of free cash → eligible, above the 15,000 limit - ∿Decision explanation generated: 3 × 6,250, 0% interest, first instalment today - ⏸Amount > 15,000 → underwriter approved within 90 seconds - ⇢Instalment plan #BN-204118 posted to the ledger, T+1 merchant payout committed - ➤SMS to customer: plan summary and due dates (8 Sep, 8 Oct, 8 Nov) - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger "Pay in 3 instalments" selected at checkout (API) Outcome Instalment plan posted to the ledger, merchant payout committed, customer informed ## Agents SUPERVISOR ### Credit Decision Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Intake Agent Checks basket, customer and fraud signals get_checkoutget_customercheck_fraud_signals aginies-talay-2 · on-prem ### Bureau Agent Pulls the bureau score and open-banking cash flow get_bureau_scorefetch_open_bankingcompute_affordability aginies-tulpar-4 · on-prem ### Decision Agent Decides and sets the limit from affordability and policy apply_credit_policyset_instalment_planexplain_decision aginies-talay-2 · on-prem ### Booking Agent Posts the plan to the ledger, notifies merchant and customer post_journalconfirm_merchant_payoutsend_sms ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Credit Decision Supervisor | Basket 18,750, 3-instalment request; customer E.Y., device and email known | — | 160 | — | | 02 | tool | Intake Agent | Basket and merchant data pulled: MCC 5732, merchant chargeback rate 0.4% | Merchant Checkout API | 320 | — | | 03 | plan | Credit Decision Supervisor | Plan: bureau score → open banking → affordability → policy → booking | — | 260 | 190 | | 04 | tool | Bureau Agent | Bureau score 1,480, 2 open loans, no delinquency, 1 enquiry in 30 days | Credit Bureau (KKB / Findeks) | 440 | — | | 05 | tool | Bureau Agent | Open banking: 3-month average income 62,000, free monthly cash flow 14,200 | Open Banking API | 680 | — | | 06 | verify | Decision Agent | Policy v3.0: instalment 6,250 ≤ 50% of free cash → eligible, above the 15,000 limit | — | 580 | 640 | | 07 | reason | Decision Agent | Decision explanation generated: 3 × 6,250, 0% interest, first instalment today | — | 420 | 380 | | 08 | approval | Credit Decision Supervisor | Amount > 15,000 → underwriter approved within 90 seconds | — | 2,200 | — | | 09 | write | Booking Agent | Instalment plan #BN-204118 posted to the ledger, T+1 merchant payout committed | Core Ledger | 480 | — | | 10 | notify | Booking Agent | SMS to customer: plan summary and due dates (8 Sep, 8 Oct, 8 Nov) | SMS Gateway | 220 | — | Simulation metrics (estimated) 54,000runs / month 97.3%success 5.6 sp50 $0.0015est. cost / run Approval rule Basket > 15,000 or bureau score < 1,200 → underwriter approval Step profile ingest · 1plan · 1reason · 1tool · 3verify · 1approval · 1write · 1notify · 1 SYSTEMS - Merchant Checkout APIapi - Credit Bureau (KKB / Findeks)api - Open Banking APIapi - Core Ledgercore - SMS Gatewaychannel What you need to run it - Credentials and a test environment for the enterprise APIs involved - Read/write API or MCP server and a service account for the core system - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the checkout-time, open-banking-backed credit decisioning of global BNPL providers Packages that include this scenario Financial analysis & operational intelligence → Sector: Fintech & Payments → Related agents Fintech & PaymentsHITL ### Merchant operations assistant: POS, settlement, refund and limit The merchant writes on WhatsApp; POS fault, settlement query, refund and limit increase are resolved in one dialogue. 64,000 runs / month 93.8% success 5 agents Fintech & PaymentsHITL ### Merchant KYB onboarding Registry, UBO and sanctions checks in one flow; low-risk merchants start accepting payments the same day. 1,400 runs / month 93.4% success 5 agents Fintech & PaymentsHITL ### AML transaction monitoring alert triage Alerts arrive enriched with a draft narrative; the analyst always decides and the filing goes out in minutes. 8,500 runs / month 95.8% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Call centre quality monitoring URL: https://www.aginies.com/en/agents/call-center-quality-monitoring Language: en Summary: Every call scored against the rubric, not a sample; compliance breaches and coaching flags ready by morning. Nightly batch: the day's call recordings from Genesys → Scorecard filled for every call, breaches with the team leader, weekly coaching summary in Workday Enterprise Functionsvoicebatch # Call centre quality monitoring Every call scored against the rubric, not a sample; compliance breaches and coaching flags ready by morning. In short Call centre quality monitoring is an agent scenario in the Enterprise Functions sector running on the voice, batch channels. Trigger: Nightly batch: the day's call recordings from Genesys. The Quality Supervisor coordinates 3 specialist agents (Transcription Agent, Scoring Agent, Coaching Agent) across 4 systems (Genesys SIP, QM Scorecard / Case Management, QA Rubric Vector Store, Workday); of 9 steps, 1 write to systems and 2 are checks or approvals. Outcome: Scorecard filled for every call, breaches with the team leader, weekly coaching summary in Workday. Approval rule: Score < 60 or compliance breach → team leader review. Enterprise Functions ### Call centre quality monitoring [Call centre quality monitoring] run #41,288,609elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Score < 60 or compliance breach → team leader review Agents in this run Quality Supervisorsupervisorplans, routes, approves Transcription Agentaginies-voice-tr-v3 · on-premTranscribes, diarises speakers and redacts PII Scoring Agentaginies-tulpar-4 · on-premScores against rubric criteria and checks compliance phrases Coaching Agentaginies-sunkar-1 · on-premRaises coaching flags and produces summaries Audit trail00:00.00 - ◦Nightly batch: 4,120 recordings from Genesys queued (100% coverage) - ⚙Call C-9921834 (6:42 min) fetched with agent and queue metadata - ∿Transcript with 2 speakers, 94% word accuracy, card and ID numbers redacted - ⌕Rubric v7 retrieved: 12 criteria, 3 mandatory compliance phrases - ∿Score 78/100 · missing: disclosure phrase · empathy 4/5 · resolved: yes - ✓Calibration: 2% sample double-scored by QA, drift vs human 3.1 pts → OK - ⇢Scorecard updated; coaching flag "compliance phrase" opened for agent A-2231 - ⏸Compliance breach → team leader reviewed before it reached the file - ➤Weekly coaching summary pushed to Workday: 3 agents flagged, 41 praised - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Nightly batch: the day's call recordings from Genesys Outcome Scorecard filled for every call, breaches with the team leader, weekly coaching summary in Workday ## Agents SUPERVISOR ### Quality Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-voice-tr-v3 · on-prem ### Transcription Agent Transcribes, diarises speakers and redacts PII get_recordingtranscribe_calldiarize_speakersredact_pii aginies-tulpar-4 · on-prem ### Scoring Agent Scores against rubric criteria and checks compliance phrases search_rubricscore_rubricdetect_compliance_phrasesdetect_sentiment aginies-sunkar-1 · on-prem ### Coaching Agent Raises coaching flags and produces summaries summarise_callcreate_coaching_flagupdate_scorecardpush_hr_summary ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Quality Supervisor | Nightly batch: 4,120 recordings from Genesys queued (100% coverage) | — | 300 | — | | 02 | tool | Transcription Agent | Call C-9921834 (6:42 min) fetched with agent and queue metadata | Genesys SIP | 380 | — | | 03 | reason | Transcription Agent | Transcript with 2 speakers, 94% word accuracy, card and ID numbers redacted | — | 1,400 | 520 | | 04 | retrieve | Scoring Agent | Rubric v7 retrieved: 12 criteria, 3 mandatory compliance phrases | QA Rubric Vector Store | 320 | 300 | | 05 | reason | Scoring Agent | Score 78/100 · missing: disclosure phrase · empathy 4/5 · resolved: yes | — | 960 | 1,300 | | 06 | verify | Scoring Agent | Calibration: 2% sample double-scored by QA, drift vs human 3.1 pts → OK | — | 340 | 220 | | 07 | write | Coaching Agent | Scorecard updated; coaching flag "compliance phrase" opened for agent A-2231 | QM Scorecard / Case Management | 460 | — | | 08 | approval | Quality Supervisor | Compliance breach → team leader reviewed before it reached the file | — | 2,200 | — | | 09 | notify | Coaching Agent | Weekly coaching summary pushed to Workday: 3 agents flagged, 41 praised | Workday | 260 | — | Simulation metrics (estimated) 90,000runs / month 98.8%success 6.6 sp50 $0.0016est. cost / run Approval rule Score < 60 or compliance breach → team leader review Step profile ingest · 1retrieve · 1reason · 2tool · 1verify · 1approval · 1write · 1notify · 1 SYSTEMS - Genesys SIPvoice - QM Scorecard / Case Managementticketing - QA Rubric Vector Storedb - Workdayhr What you need to run it - SIP integration with the switch and recording access - API access and assignment rules for the ticketing or work-order system - Read access to the database or vector store; a data owner - HR system API access and personal-data boundaries - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors 100% QA coverage programmes at large BPOs and bank contact centres Packages that include this scenario Call-centre quality monitoring →KVKK / GDPR personal data protection →Voice analysis & quality → Related agents Enterprise FunctionsHITL ### Procure-to-pay desk: request, quotes, contract and invoice in one place The request arrives by email; quotes compared, contract clauses checked, PO opened, invoice three-way matched and payment scheduled. 14,000 runs / month 92.7% success 5 agents Enterprise Functions ### Email classification and attachment check Every e-mail in the shared mailbox is classified, attachments checked against a list, gaps asked for at once. 86,000 runs / month 97.6% success 4 agents Enterprise FunctionsHITL ### Contract clause extraction Clauses are extracted and compared with the legal playbook; deviations reach counsel with redline suggestions. 1,400 runs / month 95.1% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Credit card debt payment URL: https://www.aginies.com/en/agents/card-debt-payment Language: en Summary: The customer types or talks; debt is fetched, an account chosen, OTP verified and the payment made in the core. "Pay my card debt" request in the app or by voice → Payment executed in the core, customer informed, decision trail stored Bankingappvoice # Credit card debt payment The customer types or talks; debt is fetched, an account chosen, OTP verified and the payment made in the core. In short Credit card debt payment is an agent scenario in the Banking sector running on the app, voice channels. Trigger: "Pay my card debt" request in the app or by voice. The Card Supervisor coordinates 3 specialist agents (Intent Agent, Payment Agent, Voice Agent) across 4 systems (Card Management, Core Banking, Knowledge Base · kart_sss_v3, OTP / SMS Gateway); of 9 steps, 1 write to systems and 1 are checks or approvals. Outcome: Payment executed in the core, customer informed, decision trail stored. Approval rule: Amount > 100,000 → branch manager approval. Banking ### Credit card debt payment [Credit card debt payment] run #41,288,311elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Amount > 100,000 → branch manager approval Agents in this run Card Supervisorsupervisorplans, routes, approves Intent Agentaginies-sunkar-1 · on-premUnderstands the request; extracts card, amount and payment type Payment Agentaginies-tulpar-4 · on-premChecks balances, requests OTP, executes the payment in the core Voice Agentaginies-voice-tr-v3 · on-premTranscribes speech and voices the reply on the voice channel Audit trail00:00.00 - ◦Request received, session identity verified · channel: app - ∿Intent: debt payment · Card: ****2210 · Type: full statement balance - ⚙Card debt fetched: 24,860 · due 14 September - ⌘Plan: account selection → OTP → payment → notify - ⚙Accounts listed: current ****7741 has sufficient balance - ✓Policy: amount ≤ 100,000 → no approval; OTP mandatory - ➤OTP sent and verified (2 attempts left) - ⇢24,860 paid from ****7741 to the card · txn #TX-88120 - ➤Spoken and written confirmation given, decision trail stored - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger "Pay my card debt" request in the app or by voice Outcome Payment executed in the core, customer informed, decision trail stored ## Agents SUPERVISOR ### Card Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Intent Agent Understands the request; extracts card, amount and payment type detect_intentextract_entitiesget_card_debt aginies-tulpar-4 · on-prem ### Payment Agent Checks balances, requests OTP, executes the payment in the core get_accountssend_otppay_from_account aginies-voice-tr-v3 · on-prem ### Voice Agent Transcribes speech and voices the reply on the voice channel transcribe_streamspeak ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Card Supervisor | Request received, session identity verified · channel: app | — | 160 | — | | 02 | reason | Intent Agent | Intent: debt payment · Card: ****2210 · Type: full statement balance | — | 520 | 380 | | 03 | tool | Intent Agent | Card debt fetched: 24,860 · due 14 September | Card Management | 340 | — | | 04 | plan | Card Supervisor | Plan: account selection → OTP → payment → notify | — | 280 | 190 | | 05 | tool | Payment Agent | Accounts listed: current ****7741 has sufficient balance | Core Banking | 410 | — | | 06 | verify | Payment Agent | Policy: amount ≤ 100,000 → no approval; OTP mandatory | — | 300 | 210 | | 07 | notify | Payment Agent | OTP sent and verified (2 attempts left) | OTP / SMS Gateway | 2,200 | — | | 08 | write | Payment Agent | 24,860 paid from ****7741 to the card · txn #TX-88120 | Core Banking | 540 | — | | 09 | notify | Voice Agent | Spoken and written confirmation given, decision trail stored | OTP / SMS Gateway | 260 | 120 | Simulation metrics (estimated) 52,000runs / month 97.8%success 5 sp50 $0.001est. cost / run Approval rule Amount > 100,000 → branch manager approval Step profile ingest · 1plan · 1reason · 1tool · 2verify · 1write · 1notify · 2 SYSTEMS - Card Managementcore - Core Bankingcore - Knowledge Base · kart_sss_v3db - OTP / SMS Gatewaychannel What you need to run it - Read/write API or MCP server and a service account for the core system - Read access to the database or vector store; a data owner - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Conversational banking: the in-app assistant pattern that completes a payment inside the conversation Packages that include this scenario Credit and onboarding operations → Sector: Banking → Related agents BankingHITL ### Dialogue banking: voice and text transaction assistant The customer talks or types; bill, limit, dispute and transfer are completed in core banking in one session. 186,000 runs / month 94.7% success 5 agents BankingHITL ### Credit card limit increase The customer types in the app or talks to the call centre; the decision takes seconds, and anything above threshold goes to a manager. 18,400 runs / month 96.4% success 4 agents BankingHITL ### Digital account opening (KYC) ID, liveness and sanctions checks in one flow; clean applications open an account in minutes, no branch visit. 42,000 runs / month 93.8% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Catalogue enrichment (batch) URL: https://www.aginies.com/en/agents/catalogue-enrichment Language: en Summary: Attributes pulled from supplier PDFs and images, TR/EN copy written; hundreds of SKUs ready in minutes. Supplier catalogue files landing in the S3 drop folder → SKUs publish-ready in the PIM, exceptions queued, catalogue team has the summary report Retail & E-commercebatchemail # Catalogue enrichment (batch) Attributes pulled from supplier PDFs and images, TR/EN copy written; hundreds of SKUs ready in minutes. In short Catalogue enrichment (batch) is an agent scenario in the Retail & E-commerce sector running on the batch, email channels. Trigger: Supplier catalogue files landing in the S3 drop folder. The Catalogue Supervisor coordinates 3 specialist agents (Document Agent, Normalisation Agent, Content Agent) across 4 systems (Document Store (S3), PIM / Catalogue, SAP ERP, Email (Exchange)); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: SKUs publish-ready in the PIM, exceptions queued, catalogue team has the summary report. Approval rule: New category or supplier price deviation > 15% → category manager approval. Retail & E-commerce ### Catalogue enrichment (batch) [Catalogue enrichment (batch)] run #41,289,239elapsed 0 mstokens 0est. cost $0.000status —Approval rule: New category or supplier price deviation > 15% → category manager approval Agents in this run Catalogue Supervisorsupervisorplans, routes, approves Document Agentaginies-konrul-1 · on-premExtracts product attributes from PDFs and images Normalisation Agentaginies-talay-2 · on-premMaps to the taxonomy, normalises units, dedupes SKUs Content Agentaginies-tulpar-4 · on-premWrites TR/EN descriptions and SEO titles in brand tone Audit trail00:00.00 - ◦Supplier drop: 340-product PDF catalogue + 1,120 images landed in S3 - ⌕Files read; 14 attributes per SKU extracted (material, dimensions, colour), confidence 96% - ⌘Plan: taxonomy mapping → unit normalisation → dedupe → content → QC - ⚙Existing 2,150 SKUs and the supplier price list pulled from SAP - ∿340 SKUs → 312 new, 28 existing; sizes converted to cm, 3 colour codes unmapped - ∿TR/EN descriptions and SEO titles written for 312 products in brand tone - ✓QC: mandatory fields 99.1% complete; 9 SKUs missing dimensions → human queue - ⏸New category or price deviation > 15% → category manager approved 12 SKUs - ⇢303 SKUs saved to the PIM as "ready to publish", 9 as "incomplete" - ➤Summary report and exception list emailed to the catalogue team - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Supplier catalogue files landing in the S3 drop folder Outcome SKUs publish-ready in the PIM, exceptions queued, catalogue team has the summary report ## Agents SUPERVISOR ### Catalogue Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Document Agent Extracts product attributes from PDFs and images parse_supplier_pdfextract_attributesread_product_image aginies-talay-2 · on-prem ### Normalisation Agent Maps to the taxonomy, normalises units, dedupes SKUs get_existing_skusmap_to_taxonomynormalise_unitsdedupe_skurun_qc aginies-tulpar-4 · on-prem ### Content Agent Writes TR/EN descriptions and SEO titles in brand tone write_descriptiongenerate_seo_titletranslatesave_to_pimsend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Catalogue Supervisor | Supplier drop: 340-product PDF catalogue + 1,120 images landed in S3 | — | 300 | — | | 02 | retrieve | Document Agent | Files read; 14 attributes per SKU extracted (material, dimensions, colour), confidence 96% | Document Store (S3) | 1,800 | 3,400 | | 03 | plan | Catalogue Supervisor | Plan: taxonomy mapping → unit normalisation → dedupe → content → QC | — | 260 | 190 | | 04 | tool | Normalisation Agent | Existing 2,150 SKUs and the supplier price list pulled from SAP | SAP ERP | 520 | — | | 05 | reason | Normalisation Agent | 340 SKUs → 312 new, 28 existing; sizes converted to cm, 3 colour codes unmapped | — | 740 | 680 | | 06 | reason | Content Agent | TR/EN descriptions and SEO titles written for 312 products in brand tone | — | 2,400 | 5,200 | | 07 | verify | Normalisation Agent | QC: mandatory fields 99.1% complete; 9 SKUs missing dimensions → human queue | — | 480 | 350 | | 08 | approval | Catalogue Supervisor | New category or price deviation > 15% → category manager approved 12 SKUs | — | 3,200 | — | | 09 | write | Content Agent | 303 SKUs saved to the PIM as "ready to publish", 9 as "incomplete" | PIM / Catalogue | 620 | — | | 10 | notify | Content Agent | Summary report and exception list emailed to the catalogue team | Email (Exchange) | 240 | — | Simulation metrics (estimated) 900runs / month 95.8%success 10.5 sp50 $0.0198est. cost / run Approval rule New category or supplier price deviation > 15% → category manager approval Step profile ingest · 1plan · 1retrieve · 1reason · 2tool · 1verify · 1approval · 1write · 1notify · 1 SYSTEMS - Document Store (S3)storage - PIM / Cataloguedb - SAP ERPerp - Email (Exchange)email What you need to run it - Read access to the document store and the document types - Read access to the database or vector store; a data owner - Posting and query API access to ERP/MES - Mailbox access (Microsoft 365 or Google Workspace) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the PIM enrichment pattern marketplaces use to onboard supplier data at scale Packages that include this scenario Sector: Retail & E-commerce → Related agents Retail & E-commerceHITL ### Outfit stylist: a complete look by size, stock, budget and style The customer sends a photo and a sentence; the agent builds three looks that are in stock, fit and within budget, and adds the chosen one to the cart. 142,000 runs / month 96.1% success 5 agents Retail & E-commerce ### Where is my order? (WISMO) Customer asks, order and carrier events join instantly; on delay a new ETA is given and a trace opened. 110,000 runs / month 98.1% success 4 agents Retail & E-commerceHITL ### Returns and refunds Policy and history checked in seconds, label issued; the refund executes when the carrier scans the parcel. 48,000 runs / month 96.7% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Chargeback dispute handling URL: https://www.aginies.com/en/agents/chargeback-dispute Language: en Summary: Dispute captured, evidence assembled and filed with the card scheme; large provisional credits go to approval. "I don't recognise this / goods not received" dispute in app or branch → Chargeback filed with the scheme, provisional credit posted, customer knows the timeline Bankingappemailbranch # Chargeback dispute handling Dispute captured, evidence assembled and filed with the card scheme; large provisional credits go to approval. In short Chargeback dispute handling is an agent scenario in the Banking sector running on the app, email, branch channels. Trigger: "I don't recognise this / goods not received" dispute in app or branch. The Dispute Supervisor coordinates 3 specialist agents (Intake Agent, Evidence Agent, Filing Agent) across 4 systems (Card Management, Document Store (S3), Visa / Mastercard Dispute API, Email (Exchange)); of 10 steps, 2 write to systems and 2 are checks or approvals. Outcome: Chargeback filed with the scheme, provisional credit posted, customer knows the timeline. Approval rule: Amount > 2,500 → dispute team lead approval before provisional credit. Banking ### Chargeback dispute handling [Chargeback dispute handling] run #41,288,209elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Amount > 2,500 → dispute team lead approval before provisional credit Agents in this run Dispute Supervisorsupervisorplans, routes, approves Intake Agentaginies-sunkar-1 · on-premClassifies the dispute, identifies the transaction and reason code Evidence Agentaginies-tulpar-4 · on-premBuilds the evidence pack from transaction and merchant data Filing Agentaginies-talay-2 · on-premPosts provisional credit, files with the scheme, informs the customer Audit trail00:00.00 - ◦In-app dispute: 3,200 transaction dated 12 Mar, "goods not received" - ✓Reason code 13.1 (goods not received) · within the 120-day window → valid - ⌘Plan: transaction detail → merchant history → evidence pack → file with scheme - ⚙Transaction detail pulled; merchant MCC 5732, 14 disputes in the last 90 days - ⌕Customer's order email and courier tracking screenshot retrieved - ∿Evidence summary: no delivery record, merchant unresponsive → strong case - ⏸Amount > 2,500 → dispute team lead approved provisional credit - ⇢Provisional credit of 3,200 posted to the account - ⇢Chargeback filed with Visa, case #CB-20481 - ➤Email to customer: case number and 45-day timeline - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger "I don't recognise this / goods not received" dispute in app or branch Outcome Chargeback filed with the scheme, provisional credit posted, customer knows the timeline ## Agents SUPERVISOR ### Dispute Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Intake Agent Classifies the dispute, identifies the transaction and reason code classify_disputeextract_transactionmap_reason_code aginies-tulpar-4 · on-prem ### Evidence Agent Builds the evidence pack from transaction and merchant data get_transaction_detailget_merchant_historyfetch_customer_docsdraft_dispute_summary aginies-talay-2 · on-prem ### Filing Agent Posts provisional credit, files with the scheme, informs the customer post_provisional_creditfile_chargebacksend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Dispute Supervisor | In-app dispute: 3,200 transaction dated 12 Mar, "goods not received" | — | 200 | — | | 02 | verify | Intake Agent | Reason code 13.1 (goods not received) · within the 120-day window → valid | — | 560 | 520 | | 03 | plan | Dispute Supervisor | Plan: transaction detail → merchant history → evidence pack → file with scheme | — | 300 | 210 | | 04 | tool | Evidence Agent | Transaction detail pulled; merchant MCC 5732, 14 disputes in the last 90 days | Card Management | 440 | — | | 05 | retrieve | Evidence Agent | Customer's order email and courier tracking screenshot retrieved | Document Store (S3) | 380 | — | | 06 | reason | Evidence Agent | Evidence summary: no delivery record, merchant unresponsive → strong case | — | 1,100 | 1,450 | | 07 | approval | Dispute Supervisor | Amount > 2,500 → dispute team lead approved provisional credit | — | 2,800 | — | | 08 | write | Filing Agent | Provisional credit of 3,200 posted to the account | Card Management | 480 | — | | 09 | write | Filing Agent | Chargeback filed with Visa, case #CB-20481 | Visa / Mastercard Dispute API | 620 | — | | 10 | notify | Filing Agent | Email to customer: case number and 45-day timeline | Email (Exchange) | 240 | — | Simulation metrics (estimated) 9,800runs / month 95.2%success 7.1 sp50 $0.0048est. cost / run Approval rule Amount > 2,500 → dispute team lead approval before provisional credit Step profile ingest · 1plan · 1retrieve · 1reason · 1tool · 1verify · 1approval · 1write · 2notify · 1 SYSTEMS - Card Managementcore - Document Store (S3)storage - Visa / Mastercard Dispute APIapi - Email (Exchange)email What you need to run it - Read/write API or MCP server and a service account for the core system - Read access to the document store and the document types - Credentials and a test environment for the enterprise APIs involved - Mailbox access (Microsoft 365 or Google Workspace) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors end-to-end dispute automation aligned with Visa/Mastercard scheme rules at card issuers Packages that include this scenario Fraud and dispute operations → Sector: Banking → Related agents BankingHITL ### Dialogue banking: voice and text transaction assistant The customer talks or types; bill, limit, dispute and transfer are completed in core banking in one session. 186,000 runs / month 94.7% success 5 agents BankingHITL ### Credit card limit increase The customer types in the app or talks to the call centre; the decision takes seconds, and anything above threshold goes to a manager. 18,400 runs / month 96.4% success 4 agents BankingHITL ### Digital account opening (KYC) ID, liveness and sanctions checks in one flow; clean applications open an account in minutes, no branch visit. 42,000 runs / month 93.8% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # CI/CD release pipeline URL: https://www.aginies.com/en/agents/cicd-release-pipeline Language: en Summary: Build, tests, scans and staging run unattended; production ships only on release-manager approval, via ArgoCD. Merge to main or a v2.14.0 release tag → v2.14.0 in production, health checks green, release notes published, rollback plan ready Software Deliveryapibatch # CI/CD release pipeline Build, tests, scans and staging run unattended; production ships only on release-manager approval, via ArgoCD. In short CI/CD release pipeline is an agent scenario in the Software Delivery sector running on the api, batch channels. Trigger: Merge to main or a v2.14.0 release tag. The Release Supervisor coordinates 4 specialist agents (Build Agent, Security Agent, Tester Agent, Release Agent) across 4 systems (CI Runner (GitHub Actions), SonarQube + Snyk, Kubernetes / ArgoCD, Slack); of 10 steps, 2 write to systems and 3 are checks or approvals. Outcome: v2.14.0 in production, health checks green, release notes published, rollback plan ready. Approval rule: Every production deploy → release-manager approval mandatory; a critical CVE closes the gate automatically. Software Delivery ### CI/CD release pipeline [CI/CD release pipeline] run #41,287,954elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Every production deploy → release-manager approval mandatory; a critical CVE closes the gate automatically Agents in this run Release Supervisorsupervisorplans, routes, approves Build Agentaginies-sunkar-1 · on-premBuilds, runs unit/integration tests, pushes the image Security Agentaginies-tulpar-4 · on-premEvaluates SAST and dependency scans, gates by policy Tester Agentaginies-tulpar-4 · on-premRuns smoke tests on staging, compares results with the baseline Release Agentaginies-talay-2 · on-premDeploys via ArgoCD, writes release notes, informs the team Audit trail00:00.00 - ◦Tag v2.14.0 on main: 23 commits, 4 PRs, 6 days since the last release - ⚙CI: build 4 min 12 s, 1,284 unit + 96 integration tests passed, image pushed - ⚙SAST: 0 critical, 2 medium; dependencies: 1 high CVE (lodash 4.17.20) - ✓High CVE sits on an unreachable code path → policy: pass, follow-up PROJ-4877 raised - ⌘Plan: staging deploy → smoke tests → release-manager approval → production → notes - ⇢Deployed to staging (argocd sync billing-api-staging), 3/3 pods ready - ✓Smoke tests 42/42 passed; p95 latency 184 ms, 3% better than baseline - ⏸Production deploy → release manager approved change record CHG-2107 - ⇢Production rollout canary 10% → 100%, health checks green, rollback ready - ➤Release notes published; summary and 4 PR links posted to #releases - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Merge to main or a v2.14.0 release tag Outcome v2.14.0 in production, health checks green, release notes published, rollback plan ready ## Agents SUPERVISOR ### Release Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Build Agent Builds, runs unit/integration tests, pushes the image run_buildrun_testspush_image aginies-tulpar-4 · on-prem ### Security Agent Evaluates SAST and dependency scans, gates by policy run_sastscan_dependenciestriage_findings aginies-tulpar-4 · on-prem ### Tester Agent Runs smoke tests on staging, compares results with the baseline run_smoke_testscheck_healthcompare_baseline aginies-talay-2 · on-prem ### Release Agent Deploys via ArgoCD, writes release notes, informs the team deploy_argocddraft_release_notespost_slack ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Release Supervisor | Tag v2.14.0 on main: 23 commits, 4 PRs, 6 days since the last release | — | 180 | — | | 02 | tool | Build Agent | CI: build 4 min 12 s, 1,284 unit + 96 integration tests passed, image pushed | CI Runner (GitHub Actions) | 3,600 | — | | 03 | tool | Security Agent | SAST: 0 critical, 2 medium; dependencies: 1 high CVE (lodash 4.17.20) | SonarQube + Snyk | 2,400 | — | | 04 | verify | Security Agent | High CVE sits on an unreachable code path → policy: pass, follow-up PROJ-4877 raised | — | 900 | 1,100 | | 05 | plan | Release Supervisor | Plan: staging deploy → smoke tests → release-manager approval → production → notes | — | 280 | 200 | | 06 | write | Release Agent | Deployed to staging (argocd sync billing-api-staging), 3/3 pods ready | Kubernetes / ArgoCD | 1,800 | — | | 07 | verify | Tester Agent | Smoke tests 42/42 passed; p95 latency 184 ms, 3% better than baseline | — | 2,600 | 600 | | 08 | approval | Release Supervisor | Production deploy → release manager approved change record CHG-2107 | — | 4,000 | — | | 09 | write | Release Agent | Production rollout canary 10% → 100%, health checks green, rollback ready | Kubernetes / ArgoCD | 3,400 | — | | 10 | notify | Release Agent | Release notes published; summary and 4 PR links posted to #releases | Slack | 320 | 700 | Simulation metrics (estimated) 950runs / month 94.7%success 18.9 sp50 $0.0075est. cost / run Approval rule Every production deploy → release-manager approval mandatory; a critical CVE closes the gate automatically Step profile ingest · 1plan · 1tool · 2verify · 2approval · 1write · 2notify · 1 SYSTEMS - CI Runner (GitHub Actions)api - SonarQube + Snykapi - Kubernetes / ArgoCDapi - Slackchannel What you need to run it - Credentials and a test environment for the enterprise APIs involved - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors agent-augmented release pipelines at large software teams practising GitOps and progressive delivery Packages that include this scenario SDLC agent team → Sector: Software Delivery → Related agents Software DeliveryHITL ### Legacy module modernisation: analysis, tests, rewrite, PR Starts from a Jira epic; the old module is read, its behaviour pinned with tests, rewritten and opened as a PR with evidence. 1,900 runs / month 90.4% success 5 agents Software DeliveryHITL ### Feature delivery pipeline A Jira epic turns into criteria, design, code and tests via agents; humans only review and approve the PR. 1,200 runs / month 91.2% success 5 agents Software DeliveryHITL ### Incident response and hotfix Alert fires, bug reproduced, patch and regression run ready; on-call approves, post-mortem already drafted. 320 runs / month 90.5% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Citizen request routing URL: https://www.aginies.com/en/agents/citizen-request-routing Language: en Summary: e-Government requests classified in seconds, given a legal deadline and routed to the right unit with an SLA. New request via the e-Government gateway → Case opened in the right unit with an SLA, citizen told the case number and deadline Government & Public Sectorwebappemailvoice # Citizen request routing e-Government requests classified in seconds, given a legal deadline and routed to the right unit with an SLA. In short Citizen request routing is an agent scenario in the Government & Public Sector sector running on the web, app, email, voice channels. Trigger: New request via the e-Government gateway. The Request Supervisor coordinates 3 specialist agents (Classification Agent, SLA Agent, Case Agent) across 4 systems (e-Devlet Gateway, Case Management, Legislation Vector Store, SMS Gateway); of 8 steps, 1 write to systems and 2 are checks or approvals. Outcome: Case opened in the right unit with an SLA, citizen told the case number and deadline. Approval rule: Classification confidence < 0.85 or staff complaint → registry officer approval. Government & Public Sector ### Citizen request routing [Citizen request routing] run #41,286,976elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Classification confidence < 0.85 or staff complaint → registry officer approval Agents in this run Request Supervisorsupervisorplans, routes, approves Classification Agentaginies-sunkar-1 · on-premDetermines topic and unit, masks personal data SLA Agentaginies-talay-2 · on-premComputes the legal deadline from regulation Case Agentaginies-sunkar-1 · on-premOpens and assigns the case, informs the citizen Audit trail00:00.00 - ◦Request received via e-Government, identity verified by the gateway (TC) - ∿Topic: "water bill objection" · unit: Water & Sewerage · confidence 0.81 · PII masked - ⏸Confidence 0.81 < 0.85 → registry officer confirmed the unit - ⌕Petition Law 3071 and Right to Information Law 4982 → 15 working days - ∿Deadline: 29 Sep 2026; reminder on day 10, escalation on day 13 - ✓Duplicate check: no open case from the same citizen on this subject in 30 days - ⇢Case CM-2026-114882 opened, assigned to Water & Sewerage, SLA 15 days - ➤Citizen SMS: case number and 15-day deadline; status visible on e-Government - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger New request via the e-Government gateway Outcome Case opened in the right unit with an SLA, citizen told the case number and deadline ## Agents SUPERVISOR ### Request Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Classification Agent Determines topic and unit, masks personal data classify_requestextract_entitiesmask_pii aginies-talay-2 · on-prem ### SLA Agent Computes the legal deadline from regulation search_regulationscompute_deadline aginies-sunkar-1 · on-prem ### Case Agent Opens and assigns the case, informs the citizen check_duplicatecreate_caseassign_unitsend_sms ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Request Supervisor | Request received via e-Government, identity verified by the gateway (TC) | — | 200 | — | | 02 | reason | Classification Agent | Topic: "water bill objection" · unit: Water & Sewerage · confidence 0.81 · PII masked | — | 620 | 540 | | 03 | approval | Request Supervisor | Confidence 0.81 Simulation metrics (estimated) 54,000runs / month 98.1%success 4.9 sp50 $0.0009est. cost / run Approval rule Classification confidence < 0.85 or staff complaint → registry officer approval Step profile ingest · 1retrieve · 1reason · 2verify · 1approval · 1write · 1notify · 1 SYSTEMS - e-Devlet Gatewayapi - Case Managementticketing - Legislation Vector Storedb - SMS Gatewaychannel What you need to run it - Credentials and a test environment for the enterprise APIs involved - API access and assignment rules for the ticketing or work-order system - Read access to the database or vector store; a data owner - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the one-stop request routing of digital government services such as Estonia and GOV.UK Packages that include this scenario Email classification & attachment verification →Citizen services & file pre-review → Sector: Government & Public Sector → Related agents Government & Public SectorHITL ### Citizen services assistant: application, documents, appointment and status The citizen writes or calls; eligibility checked, documents read, application opened, appointment given, every step traceable. 230,000 runs / month 93.8% success 5 agents Government & Public SectorHITL ### Permit application check Missing documents and regulatory fit reported in minutes; the licensing officer always makes the decision. 3,100 runs / month 95.8% success 4 agents Government & Public SectorHITL ### Public procurement bid file pre-check Bid files are compared with the specification, missing and non-conforming documents listed; the evaluation decision belongs to the commission. 380 runs / month 94.8% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Citizen services assistant: application, documents, appointment and status URL: https://www.aginies.com/en/agents/citizen-services-assistant Language: en Summary: The citizen writes or calls; eligibility checked, documents read, application opened, appointment given, every step traceable. Application request via the municipal call centre, WhatsApp or e-Government referral → Application file opened, missing document completed, appointment given, tracking number sent Government & Public Sectorwhatsappvoiceweb # Citizen services assistant: application, documents, appointment and status The citizen writes or calls; eligibility checked, documents read, application opened, appointment given, every step traceable. In short Citizen services assistant: application, documents, appointment and status is an agent scenario in the Government & Public Sector sector running on the whatsapp, voice, web channels. Trigger: Application request via the municipal call centre, WhatsApp or e-Government referral. The Citizen Services Supervisor coordinates 4 specialist agents (Dialogue Agent, Eligibility Agent, Document Agent, Case Agent) across 5 systems (WhatsApp Business, e-Government Gateway, Case Management, Document Store, Appointment System); of 10 steps, 2 write to systems and 2 are checks or approvals. Outcome: Application file opened, missing document completed, appointment given, tracking number sent. Approval rule: Borderline eligibility, inconsistent document or another application in the household → officer review. Government & Public Sector ### Citizen services assistant: application, documents, appointment and status [Citizen services assistant: application, documents, appointment and status] run #41,287,381elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Borderline eligibility, inconsistent document or another application in the household → officer review Agents in this run Citizen Services Supervisorsupervisorplans, routes, approves Dialogue Agentaginies-talay-2 · on-premTalks with the citizen in plain language, clarifies request and identity Eligibility Agentaginies-tulpar-4 · on-premAssesses eligibility against regulation and criteria, writes the reasoning Document Agentaginies-konrul-1 · on-premReads uploaded documents, extracts fields, lists what is missing Case Agentaginies-sunkar-1 · on-premOpens the application, gives the appointment, assigns the officer Audit trail00:00.00 - ◦WhatsApp: "I want to apply for the gas allowance, sending my lease and payslip" - ✓Identity confirmed via the e-Government verification link; household data from the civil registry - ⌘Plan: read documents → criteria → eligibility → gaps → file → appointment → tracking number - ∿Lease: 9,500/month, 12 months; payslip: net 24,300; both legible and signed - ⌕Criteria: household income threshold, rent ratio, other aid at the address; social-aid registry queried - ∿Eligibility: income 8% below threshold, rent ratio 39% → eligible; reasoning text written - ✓Missing: residence certificate; fetched automatically from e-Government, nothing more asked of the citizen - ⇢Application #SA-2026-48811 opened, documents linked, officer assigned - ⇢Social review appointment: Tuesday 10:30, home visit; entered in the calendar - ➤Tracking number, appointment and process steps in plain language on WhatsApp; SMS backup sent - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Application request via the municipal call centre, WhatsApp or e-Government referral Outcome Application file opened, missing document completed, appointment given, tracking number sent ## Agents SUPERVISOR ### Citizen Services Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Dialogue Agent Talks with the citizen in plain language, clarifies request and identity verify_identitydetect_intentcompose_reply aginies-tulpar-4 · on-prem ### Eligibility Agent Assesses eligibility against regulation and criteria, writes the reasoning get_criteriaquery_registryassess_eligibility aginies-konrul-1 · on-prem ### Document Agent Reads uploaded documents, extracts fields, lists what is missing extract_fieldsvalidate_documentlist_missing aginies-sunkar-1 · on-prem ### Case Agent Opens the application, gives the appointment, assigns the officer open_casebook_appointmentassign_officersend_sms ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Citizen Services Supervisor | WhatsApp: "I want to apply for the gas allowance, sending my lease and payslip" | WhatsApp Business | 180 | — | | 02 | verify | Dialogue Agent | Identity confirmed via the e-Government verification link; household data from the civil registry | e-Government Gateway | 1,600 | 160 | | 03 | plan | Citizen Services Supervisor | Plan: read documents → criteria → eligibility → gaps → file → appointment → tracking number | — | 240 | 220 | | 04 | reason | Document Agent | Lease: 9,500/month, 12 months; payslip: net 24,300; both legible and signed | Document Store | 2,100 | 1,200 | | 05 | retrieve | Eligibility Agent | Criteria: household income threshold, rent ratio, other aid at the address; social-aid registry queried | e-Government Gateway | 900 | 420 | | 06 | reason | Eligibility Agent | Eligibility: income 8% below threshold, rent ratio 39% → eligible; reasoning text written | — | 1,300 | 700 | | 07 | verify | Document Agent | Missing: residence certificate; fetched automatically from e-Government, nothing more asked of the citizen | e-Government Gateway | 800 | — | | 08 | write | Case Agent | Application #SA-2026-48811 opened, documents linked, officer assigned | Case Management | 560 | — | | 09 | write | Case Agent | Social review appointment: Tuesday 10:30, home visit; entered in the calendar | Appointment System | 480 | — | | 10 | notify | Dialogue Agent | Tracking number, appointment and process steps in plain language on WhatsApp; SMS backup sent | WhatsApp Business | 440 | 260 | Simulation metrics (estimated) 230,000runs / month 93.8%success 8.2 sp50 $0.0021est. cost / run Approval rule Borderline eligibility, inconsistent document or another application in the household → officer review Step profile ingest · 1plan · 1retrieve · 1reason · 2verify · 2write · 2notify · 1 SYSTEMS - WhatsApp Businesschannel - e-Government Gatewayapi - Case Managementticketing - Document Storestorage - Appointment Systemapi What you need to run it - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - Credentials and a test environment for the enterprise APIs involved - API access and assignment rules for the ticketing or work-order system - Read access to the document store and the document types - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Digital government programmes' "one application, zero repeated documents" pattern Packages that include this scenario Citizen services & file pre-review → Sector: Government & Public Sector → Related agents Government & Public SectorHITL ### Citizen request routing e-Government requests classified in seconds, given a legal deadline and routed to the right unit with an SLA. 54,000 runs / month 98.1% success 4 agents Government & Public SectorHITL ### Permit application check Missing documents and regulatory fit reported in minutes; the licensing officer always makes the decision. 3,100 runs / month 95.8% success 4 agents Government & Public SectorHITL ### Public procurement bid file pre-check Bid files are compared with the specification, missing and non-conforming documents listed; the evaluation decision belongs to the commission. 380 runs / month 94.8% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Code review and security gate URL: https://www.aginies.com/en/agents/code-review-security-gate Language: en Summary: Agents review, scan and coverage-check every PR; fixes land first, a human approves only clean PRs. Pull request opened or updated on GitHub (webhook) → PR findings fixed, security and coverage gates green, ready to merge with human approval Software Deliveryapiweb # Code review and security gate Agents review, scan and coverage-check every PR; fixes land first, a human approves only clean PRs. In short Code review and security gate is an agent scenario in the Software Delivery sector running on the api, web channels. Trigger: Pull request opened or updated on GitHub (webhook). The Review Supervisor coordinates 4 specialist agents (Reviewer Agent, Security Agent, Tester Agent, Developer Agent) across 4 systems (GitHub, SonarQube, Dependency Scanner (Snyk), CI Runner (GitHub Actions)); of 10 steps, 2 write to systems and 2 are checks or approvals. Outcome: PR findings fixed, security and coverage gates green, ready to merge with human approval. Approval rule: Every PR → human CODEOWNERS approval; a high SAST finding or coverage < 80% blocks the gate. Software Delivery ### Code review and security gate [Code review and security gate] run #41,288,696elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Every PR → human CODEOWNERS approval; a high SAST finding or coverage < 80% blocks the gate Agents in this run Review Supervisorsupervisorplans, routes, approves Reviewer Agentaginies-tulpar-4 · on-premReads the diff, writes bug, design and style findings Security Agentaginies-sunkar-1 · on-premTriages SAST, secret and dependency findings Tester Agentaginies-tulpar-4 · on-premMeasures the coverage delta, proposes missing tests Developer Agentaginies-tulpar-4 · on-premFixes findings, adds tests, updates the PR Audit trail00:00.00 - ◦PR #2318 opened: "Coupon validation service", 9 files, +684 / −120, author M.A. - ⌘Plan: review + security + coverage in parallel → fix → re-check → human approval - ∿Diff and 4 calling modules read; 3 findings: race condition, null check, N+1 query - ⚙SAST: 1 high (SQL string concatenation, line 212), no secrets found - ⚙Dependencies: newly added jsonwebtoken 8.5.1 → 1 medium CVE, 9.0.2 suggested - ⚙Coverage: new code 64% (threshold 80%), 6 tests missing for CouponRules - ⇢5 findings fixed, jsonwebtoken 9.0.2, 6 tests added; commit 4e1b9c pushed - ✓Re-check: SAST 0 high, coverage 86%, 1,312 tests passed → gates green - ⏸Code owner (CODEOWNERS) reviewed PR #2318, left 1 comment, approved - ⇢PR #2318 merged to main, review trail and scan reports attached to the PR - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Pull request opened or updated on GitHub (webhook) Outcome PR findings fixed, security and coverage gates green, ready to merge with human approval ## Agents SUPERVISOR ### Review Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-tulpar-4 · on-prem ### Reviewer Agent Reads the diff, writes bug, design and style findings read_diffsearch_codebasepost_review_comment aginies-sunkar-1 · on-prem ### Security Agent Triages SAST, secret and dependency findings run_sastscan_secretsscan_dependenciestriage_findings aginies-tulpar-4 · on-prem ### Tester Agent Measures the coverage delta, proposes missing tests run_testsreport_coveragesuggest_tests aginies-tulpar-4 · on-prem ### Developer Agent Fixes findings, adds tests, updates the PR edit_filesrun_testspush_commit ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Review Supervisor | PR #2318 opened: "Coupon validation service", 9 files, +684 / −120, author M.A. | — | 160 | — | | 02 | plan | Review Supervisor | Plan: review + security + coverage in parallel → fix → re-check → human approval | — | 280 | 210 | | 03 | reason | Reviewer Agent | Diff and 4 calling modules read; 3 findings: race condition, null check, N+1 query | — | 2,200 | 4,800 | | 04 | tool | Security Agent | SAST: 1 high (SQL string concatenation, line 212), no secrets found | SonarQube | 1,800 | — | | 05 | tool | Security Agent | Dependencies: newly added jsonwebtoken 8.5.1 → 1 medium CVE, 9.0.2 suggested | Dependency Scanner (Snyk) | 900 | — | | 06 | tool | Tester Agent | Coverage: new code 64% (threshold 80%), 6 tests missing for CouponRules | CI Runner (GitHub Actions) | 2,400 | — | | 07 | write | Developer Agent | 5 findings fixed, jsonwebtoken 9.0.2, 6 tests added; commit 4e1b9c pushed | GitHub | 3,600 | 7,400 | | 08 | verify | Tester Agent | Re-check: SAST 0 high, coverage 86%, 1,312 tests passed → gates green | — | 2,600 | 500 | | 09 | approval | Review Supervisor | Code owner (CODEOWNERS) reviewed PR #2318, left 1 comment, approved | — | 4,000 | — | | 10 | write | Developer Agent | PR #2318 merged to main, review trail and scan reports attached to the PR | GitHub | 420 | — | Simulation metrics (estimated) 4,200runs / month 93.6%success 17.9 sp50 $0.0129est. cost / run Approval rule Every PR → human CODEOWNERS approval; a high SAST finding or coverage < 80% blocks the gate Step profile ingest · 1plan · 1reason · 1tool · 3verify · 1approval · 1write · 2 SYSTEMS - GitHubapi - SonarQubeapi - Dependency Scanner (Snyk)api - CI Runner (GitHub Actions)api What you need to run it - Credentials and a test environment for the enterprise APIs involved - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors AI-assisted code review and shift-left security gating at large technology companies Packages that include this scenario SDLC agent team → Sector: Software Delivery → Related agents Software DeliveryHITL ### Legacy module modernisation: analysis, tests, rewrite, PR Starts from a Jira epic; the old module is read, its behaviour pinned with tests, rewritten and opened as a PR with evidence. 1,900 runs / month 90.4% success 5 agents Software DeliveryHITL ### Feature delivery pipeline A Jira epic turns into criteria, design, code and tests via agents; humans only review and approve the PR. 1,200 runs / month 91.2% success 5 agents Software DeliveryHITL ### CI/CD release pipeline Build, tests, scans and staging run unattended; production ships only on release-manager approval, via ArgoCD. 950 runs / month 94.7% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Cold-chain excursion response URL: https://www.aginies.com/en/agents/cold-chain-excursion-response Language: en Summary: A temperature excursion is assessed instantly, product risk computed, a field crew dispatched, the quality decision recorded. Threshold breach from a cold-room or vehicle sensor → Affected lots identified, dispatch done, quality decision and customer notification completed IoT & Connected Devicesapiappwhatsapp # Cold-chain excursion response A temperature excursion is assessed instantly, product risk computed, a field crew dispatched, the quality decision recorded. In short Cold-chain excursion response is an agent scenario in the IoT & Connected Devices sector running on the api, app, whatsapp channels. Trigger: Threshold breach from a cold-room or vehicle sensor. The Cold-Chain Supervisor coordinates 3 specialist agents (Sensor Agent, Quality Agent, Field Agent) across 5 systems (IoT Platform, WMS / Lot Tracking, Stability Rules (vector), Field Service, WhatsApp Business API); of 9 steps, 1 write to systems and 2 are checks or approvals. Outcome: Affected lots identified, dispatch done, quality decision and customer notification completed. Approval rule: Product transfer or disposal decision → quality manager approval. IoT & Connected Devices ### Cold-chain excursion response [Cold-chain excursion response] run #41,290,388elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Product transfer or disposal decision → quality manager approval Agents in this run Cold-Chain Supervisorsupervisorplans, routes, approves Sensor Agentaginies-sunkar-1 · on-premConfirms the excursion and computes its duration Quality Agentaginies-tulpar-4 · on-premAssesses risk against product stability rules Field Agentaginies-talay-2 · on-premDispatches a technician and informs the customer Audit trail00:00.00 - ◦Cold room CR-4: 8.6 °C, threshold 8 °C; in excursion for 14 minutes - ⚙Last 2 hours pulled; second sensor confirms, compressor current low - ∿Excursion real, trend +0.3 °C / 10 min; 10 °C expected in 45 min - ⚙Lots in CR-4: 18 lots, 3 product families, most sensitive: vaccine lot L-2291 - ⌕Stability rule: 2–8 °C, cumulative excursion tolerance 4 hours @ ≤ 12 °C - ✓Cumulative 14 min + forecast 45 min < 4 hours → product risk low, monitoring continues - ⇢Service order FS-2026-8812: compressor check, priority urgent, nearest technician 25 min - ⏸Transfer decision if the vaccine lot is affected → quality manager approved “hold and monitor” - ➤WhatsApp to the site lead: status, technician ETA, monitoring window; record written to the quality file - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Threshold breach from a cold-room or vehicle sensor Outcome Affected lots identified, dispatch done, quality decision and customer notification completed ## Agents SUPERVISOR ### Cold-Chain Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Sensor Agent Confirms the excursion and computes its duration get_sensor_seriesvalidate_readingcompute_excursion aginies-tulpar-4 · on-prem ### Quality Agent Assesses risk against product stability rules get_lots_in_locationsearch_stability_rulesscore_product_risk aginies-talay-2 · on-prem ### Field Agent Dispatches a technician and informs the customer create_service_orderdispatch_techniciansend_whatsapp ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Cold-Chain Supervisor | Cold room CR-4: 8.6 °C, threshold 8 °C; in excursion for 14 minutes | IoT Platform | 200 | — | | 02 | tool | Sensor Agent | Last 2 hours pulled; second sensor confirms, compressor current low | IoT Platform | 420 | — | | 03 | reason | Sensor Agent | Excursion real, trend +0.3 °C / 10 min; 10 °C expected in 45 min | — | 600 | 520 | | 04 | tool | Quality Agent | Lots in CR-4: 18 lots, 3 product families, most sensitive: vaccine lot L-2291 | WMS / Lot Tracking | 440 | — | | 05 | retrieve | Quality Agent | Stability rule: 2–8 °C, cumulative excursion tolerance 4 hours @ ≤ 12 °C | Stability Rules (vector) | 480 | 460 | | 06 | verify | Quality Agent | Cumulative 14 min + forecast 45 min Simulation metrics (estimated) 2,700runs / month 97.4%success 5.9 sp50 $0.0026est. cost / run Approval rule Product transfer or disposal decision → quality manager approval Step profile ingest · 1retrieve · 1reason · 1tool · 2verify · 1approval · 1write · 1notify · 1 SYSTEMS - IoT Platformcore - WMS / Lot Trackingerp - Stability Rules (vector)db - Field Serviceticketing - WhatsApp Business APIchannel What you need to run it - Read/write API or MCP server and a service account for the core system - Posting and query API access to ERP/MES - Read access to the database or vector store; a data owner - API access and assignment rules for the ticketing or work-order system - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors GDP-aligned cold-chain monitoring and excursion handling in pharma and food logistics Packages that include this scenario Field service & dispatch automation →IoT device fleet operations →Logistics control tower → Sector: IoT & Connected Devices → Related agents IoT & Connected DevicesHITL ### Field device support assistant: telemetry, remote diagnosis, OTA and technician The customer says "the device is dead"; telemetry read, remote diagnosis made, update pushed, technician dispatched with the right part if needed. 37,000 runs / month 94.1% success 5 agents IoT & Connected DevicesHITL ### Device fleet anomaly triage Telemetry anomalies from thousands of devices are clustered, root-cause candidates listed, one incident reaches the right team. 18,000 runs / month 96.9% success 4 agents IoT & Connected DevicesHITL ### Staged OTA update rollout Firmware rolls out in 1% → 10% → 100% stages; a health gate at every stage, human approval for critical fleets. 320 runs / month 98.1% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Early-stage collections outreach URL: https://www.aginies.com/en/agents/collections-outreach Language: en Summary: Past-due accounts are segmented and reminded on the right channel; customers set up their own payment plan. Nightly batch: accounts 5–30 days past due → Payment plan recorded, late interest frozen, customer knows the instalment dates Bankingbatchwhatsappsmsvoice # Early-stage collections outreach Past-due accounts are segmented and reminded on the right channel; customers set up their own payment plan. In short Early-stage collections outreach is an agent scenario in the Banking sector running on the batch, whatsapp, sms, voice channels. Trigger: Nightly batch: accounts 5–30 days past due. The Collections Supervisor coordinates 3 specialist agents (Segmentation Agent, Dialogue Agent, Payment Plan Agent) across 4 systems (Core Banking, Collections Platform, WhatsApp Business API, SMS Gateway); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: Payment plan recorded, late interest frozen, customer knows the instalment dates. Approval rule: Plan balance > 25,000 or > 3 instalments → collections team lead approval. Banking ### Early-stage collections outreach [Early-stage collections outreach] run #41,290,103elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Plan balance > 25,000 or > 3 instalments → collections team lead approval Agents in this run Collections Supervisorsupervisorplans, routes, approves Segmentation Agentaginies-talay-2 · on-premScores propensity to pay, picks channel and tone Dialogue Agentaginies-tulpar-4 · on-premTalks to the customer, understands intent to pay and plan requests Payment Plan Agentaginies-sunkar-1 · on-premCalculates and records a policy-compliant plan Audit trail00:00.00 - ◦Nightly batch: 12,400 accounts 5–30 days past due listed - ⚙Balance, days past due and 6-month payment behaviour pulled - ∿Customer A.K.: 22 days past due, 32,600, propensity 0.68 → WhatsApp - ⌘Plan: soft reminder → reply → plan offer → approval → record - ➤WhatsApp reminder with "pay today / request a plan" options - ∿Customer: "salary on the 25th, can we split in three?" → plan request - ✓Policy: ≤ 3 instalments, first within 15 days → 3 × 10,867 eligible, above threshold - ⏸Balance > 25,000 → collections team lead approved the plan - ⇢Payment plan #PP-33120 created, late interest frozen - ➤Plan summary and first due date (25 Sep) sent by SMS - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Nightly batch: accounts 5–30 days past due Outcome Payment plan recorded, late interest frozen, customer knows the instalment dates ## Agents SUPERVISOR ### Collections Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Segmentation Agent Scores propensity to pay, picks channel and tone get_delinquent_accountsscore_propensitypick_channel aginies-tulpar-4 · on-prem ### Dialogue Agent Talks to the customer, understands intent to pay and plan requests send_whatsappparse_replypropose_plan aginies-sunkar-1 · on-prem ### Payment Plan Agent Calculates and records a policy-compliant plan calculate_plancreate_payment_planfreeze_late_interestsend_sms ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Collections Supervisor | Nightly batch: 12,400 accounts 5–30 days past due listed | — | 300 | — | | 02 | tool | Segmentation Agent | Balance, days past due and 6-month payment behaviour pulled | Core Banking | 520 | — | | 03 | reason | Segmentation Agent | Customer A.K.: 22 days past due, 32,600, propensity 0.68 → WhatsApp | — | 680 | 640 | | 04 | plan | Collections Supervisor | Plan: soft reminder → reply → plan offer → approval → record | — | 260 | 180 | | 05 | notify | Dialogue Agent | WhatsApp reminder with "pay today / request a plan" options | WhatsApp Business API | 250 | — | | 06 | reason | Dialogue Agent | Customer: "salary on the 25th, can we split in three?" → plan request | — | 540 | 470 | | 07 | verify | Payment Plan Agent | Policy: ≤ 3 instalments, first within 15 days → 3 × 10,867 eligible, above threshold | — | 610 | 520 | | 08 | approval | Collections Supervisor | Balance > 25,000 → collections team lead approved the plan | — | 2,700 | — | | 09 | write | Payment Plan Agent | Payment plan #PP-33120 created, late interest frozen | Collections Platform | 480 | — | | 10 | notify | Payment Plan Agent | Plan summary and first due date (25 Sep) sent by SMS | SMS Gateway | 220 | — | Simulation metrics (estimated) 96,000runs / month 91.2%success 6.5 sp50 $0.0008est. cost / run Approval rule Plan balance > 25,000 or > 3 instalments → collections team lead approval Step profile ingest · 1plan · 1reason · 2tool · 1verify · 1approval · 1write · 1notify · 2 SYSTEMS - Core Bankingcore - Collections Platformdb - WhatsApp Business APIchannel - SMS Gatewaychannel What you need to run it - Read/write API or MCP server and a service account for the core system - Read access to the database or vector store; a data owner - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors digital-first collections platforms in the US/EU where customers self-serve payment plans Packages that include this scenario Fraud and dispute operations → Sector: Banking → Related agents BankingHITL ### Dialogue banking: voice and text transaction assistant The customer talks or types; bill, limit, dispute and transfer are completed in core banking in one session. 186,000 runs / month 94.7% success 5 agents BankingHITL ### Credit card limit increase The customer types in the app or talks to the call centre; the decision takes seconds, and anything above threshold goes to a manager. 18,400 runs / month 96.4% success 4 agents BankingHITL ### Digital account opening (KYC) ID, liveness and sanctions checks in one flow; clean applications open an account in minutes, no branch visit. 42,000 runs / month 93.8% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Contract clause extraction URL: https://www.aginies.com/en/agents/contract-clause-extraction Language: en Summary: Clauses are extracted and compared with the legal playbook; deviations reach counsel with redline suggestions. A vendor contract arrives from procurement for review → Deviation register and risk score in the CLM, redline v1 ready, counsel and procurement informed Enterprise Functionsemailweb # Contract clause extraction Clauses are extracted and compared with the legal playbook; deviations reach counsel with redline suggestions. In short Contract clause extraction is an agent scenario in the Enterprise Functions sector running on the email, web channels. Trigger: A vendor contract arrives from procurement for review. The Contract Supervisor coordinates 3 specialist agents (Clause Agent, Playbook Agent, CLM Agent) across 4 systems (CLM (Contract Lifecycle Management), Legal Playbook Vector Store, Document Store (S3), Exchange / M365 Mail); of 9 steps, 2 write to systems and 2 are checks or approvals. Outcome: Deviation register and risk score in the CLM, redline v1 ready, counsel and procurement informed. Approval rule: High-risk deviation or contract value > 500,000 € → counsel approval. Enterprise Functions ### Contract clause extraction [Contract clause extraction] run #41,287,821elapsed 0 mstokens 0est. cost $0.000status —Approval rule: High-risk deviation or contract value > 500,000 € → counsel approval Agents in this run Contract Supervisorsupervisorplans, routes, approves Clause Agentaginies-tulpar-4 · on-premExtracts clauses and classifies them by type Playbook Agentaginies-tulpar-4 · on-premCompares with playbook positions and drafts redlines CLM Agentaginies-talay-2 · on-premUpdates the CLM record and informs stakeholders Audit trail00:00.00 - ◦Vendor MSA (38 pages, DOCX) received from procurement for review - ∿27 clauses extracted: liability, indemnity, termination, data protection, IP… - ⌕Playbook v2026.2: standard and fallback positions for 27 clause types retrieved - ∿4 deviations: liability cap 1× (std 2×), 30-day termination (std 90), no DPA, auto-renewal - ✓Risk: 2 high, 2 medium; jurisdiction and governing law match the playbook - ⏸High-risk deviation → counsel reviewed; redline attached, not sent to the counterparty - ⇢Contract record updated: deviation register, risk score 62/100, redline v1 - ⇢Source document, clause map and redline archived - ➤Procurement and counsel notified with a summary, review task assigned - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger A vendor contract arrives from procurement for review Outcome Deviation register and risk score in the CLM, redline v1 ready, counsel and procurement informed ## Agents SUPERVISOR ### Contract Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-tulpar-4 · on-prem ### Clause Agent Extracts clauses and classifies them by type extract_clausesclassify_clause_type aginies-tulpar-4 · on-prem ### Playbook Agent Compares with playbook positions and drafts redlines search_playbookcompare_positionsdraft_redline aginies-talay-2 · on-prem ### CLM Agent Updates the CLM record and informs stakeholders update_contract_recordarchive_documentscreate_review_tasksend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Contract Supervisor | Vendor MSA (38 pages, DOCX) received from procurement for review | — | 200 | — | | 02 | reason | Clause Agent | 27 clauses extracted: liability, indemnity, termination, data protection, IP… | — | 1,600 | 2,400 | | 03 | retrieve | Playbook Agent | Playbook v2026.2: standard and fallback positions for 27 clause types retrieved | Legal Playbook Vector Store | 480 | 900 | | 04 | reason | Playbook Agent | 4 deviations: liability cap 1× (std 2×), 30-day termination (std 90), no DPA, auto-renewal | — | 1,300 | 1,800 | | 05 | verify | Playbook Agent | Risk: 2 high, 2 medium; jurisdiction and governing law match the playbook | — | 420 | 380 | | 06 | approval | Contract Supervisor | High-risk deviation → counsel reviewed; redline attached, not sent to the counterparty | — | 3,000 | — | | 07 | write | CLM Agent | Contract record updated: deviation register, risk score 62/100, redline v1 | CLM (Contract Lifecycle Management) | 520 | — | | 08 | write | CLM Agent | Source document, clause map and redline archived | Document Store (S3) | 260 | — | | 09 | notify | CLM Agent | Procurement and counsel notified with a summary, review task assigned | Exchange / M365 Mail | 240 | — | Simulation metrics (estimated) 1,400runs / month 95.1%success 8 sp50 $0.0189est. cost / run Approval rule High-risk deviation or contract value > 500,000 € → counsel approval Step profile ingest · 1retrieve · 1reason · 2verify · 1approval · 1write · 2notify · 1 SYSTEMS - CLM (Contract Lifecycle Management)core - Legal Playbook Vector Storedb - Document Store (S3)storage - Exchange / M365 Mailemail What you need to run it - Read/write API or MCP server and a service account for the core system - Read access to the database or vector store; a data owner - Read access to the document store and the document types - Mailbox access (Microsoft 365 or Google Workspace) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors playbook-based contract review at global law firms and in-house legal ops teams Packages that include this scenario Information management & decision support →Public procurement & tender compliance → Related agents Enterprise FunctionsHITL ### Procure-to-pay desk: request, quotes, contract and invoice in one place The request arrives by email; quotes compared, contract clauses checked, PO opened, invoice three-way matched and payment scheduled. 14,000 runs / month 92.7% success 5 agents Enterprise Functions ### Email classification and attachment check Every e-mail in the shared mailbox is classified, attachments checked against a list, gaps asked for at once. 86,000 runs / month 97.6% success 4 agents Enterprise FunctionsHITL ### Call centre quality monitoring Every call scored against the rubric, not a sample; compliance breaches and coaching flags ready by morning. 90,000 runs / month 98.8% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Control tower: delay prediction, rerouting, customer notice, SLA URL: https://www.aginies.com/en/agents/control-tower-exception-desk Language: en Summary: The telematics signal catches the delay early; alternative route and carrier chosen, customer informed, SLA compensation computed. Deviation event from telematics or the carrier API (ETA +4 hours) → Route changed, customer informed with new ETA, SLA record and compensation file ready Logisticsapiemailwhatsapp # Control tower: delay prediction, rerouting, customer notice, SLA The telematics signal catches the delay early; alternative route and carrier chosen, customer informed, SLA compensation computed. In short Control tower: delay prediction, rerouting, customer notice, SLA is an agent scenario in the Logistics sector running on the api, email, whatsapp channels. Trigger: Deviation event from telematics or the carrier API (ETA +4 hours). The Control Tower Supervisor coordinates 4 specialist agents (Prediction Agent, Routing Agent, Customer Agent, Finance Agent) across 5 systems (TMS, Telematics, Carrier API, Customer Portal, ERP Billing); of 10 steps, 2 write to systems and 1 are checks or approvals. Outcome: Route changed, customer informed with new ETA, SLA record and compensation file ready. Approval rule: Route cost increase > 15% or compensation > 10,000 → operations manager approval. Logistics ### Control tower: delay prediction, rerouting, customer notice, SLA [Control tower: delay prediction, rerouting, customer notice, SLA] run #41,289,493elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Route cost increase > 15% or compensation > 10,000 → operations manager approval Agents in this run Control Tower Supervisorsupervisorplans, routes, approves Prediction Agentaginies-tulpar-4 · on-premPredicts the shipment and SLA impact of the deviation Routing Agentaginies-talay-2 · on-premCompares alternative route, transfer and carrier options with cost Customer Agentaginies-sunkar-1 · on-premInforms consignee and shipper with the new ETA and options Finance Agentaginies-sunkar-1 · on-premRecords the SLA breach, computes compensation and carrier recourse Audit trail00:00.00 - ◦Telematics: vehicle 34 TR 7781 stationary on the motorway for 90 min; 3 shipments, 2 cold chain - ∿Prediction: ETA +4.5 h; shipment 2 SLA breach certain, shipment 3 marginal; cold chain tolerance 6 h - ⌘Plan: route/transfer options → cost → rebooking → customer → SLA record - ⚙3 options: transfer to backup vehicle (+2,400, ETA +1.5 h), partner carrier (+3,900), wait - ✓Option 1 cost increase 11% < 15% threshold; inside cold-chain tolerance → auto-approved - ⇢Backup vehicle assigned, transfer point and time written to the TMS, drivers instructed - ➤New ETA and live tracking link to consignees; delivery window choice offered for shipment 2 - ⇢SLA breach recorded; compensation 1,850 per contract, carrier recourse file opened - ⚙Transfer completed, new ETA confirmed; temperature log continuous, no excursion - ➤Incident summary and cost impact emailed to the shipper; incident closed on the customer portal - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Deviation event from telematics or the carrier API (ETA +4 hours) Outcome Route changed, customer informed with new ETA, SLA record and compensation file ready ## Agents SUPERVISOR ### Control Tower Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-tulpar-4 · on-prem ### Prediction Agent Predicts the shipment and SLA impact of the deviation get_telematicspredict_etasla_impact aginies-talay-2 · on-prem ### Routing Agent Compares alternative route, transfer and carrier options with cost route_optionscarrier_quoterebook_leg aginies-sunkar-1 · on-prem ### Customer Agent Informs consignee and shipper with the new ETA and options notify_customeroffer_optionslog_contact aginies-sunkar-1 · on-prem ### Finance Agent Records the SLA breach, computes compensation and carrier recourse record_slacalc_penaltycreate_claim ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Control Tower Supervisor | Telematics: vehicle 34 TR 7781 stationary on the motorway for 90 min; 3 shipments, 2 cold chain | Telematics | 160 | — | | 02 | reason | Prediction Agent | Prediction: ETA +4.5 h; shipment 2 SLA breach certain, shipment 3 marginal; cold chain tolerance 6 h | — | 1,600 | 800 | | 03 | plan | Control Tower Supervisor | Plan: route/transfer options → cost → rebooking → customer → SLA record | — | 240 | 220 | | 04 | tool | Routing Agent | 3 options: transfer to backup vehicle (+2,400, ETA +1.5 h), partner carrier (+3,900), wait | Carrier API | 1,200 | — | | 05 | verify | Routing Agent | Option 1 cost increase 11% Simulation metrics (estimated) 96,000runs / month 93.4%success 7.9 sp50 $0.0028est. cost / run Approval rule Route cost increase > 15% or compensation > 10,000 → operations manager approval Step profile ingest · 1plan · 1reason · 1tool · 2verify · 1write · 2notify · 2 SYSTEMS - TMSerp - Telematicsapi - Carrier APIapi - Customer Portalchannel - ERP Billingerp What you need to run it - Posting and query API access to ERP/MES - Credentials and a test environment for the enterprise APIs involved - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Global logistics providers' control tower pattern, with automatic rebooking and SLA finance Packages that include this scenario Operational automation & demand management →Field service & dispatch automation →Logistics control tower → Sector: Logistics → Related agents LogisticsHITL ### Shipment exception handling When a delay or damage lands, impact is costed, the carrier rebooked and the customer hears before they ask. 12,800 runs / month 95.5% success 4 agents LogisticsHITL ### Customs document preparation HS codes, invoice and packing list prepared in minutes; only low-confidence lines reach a customs specialist. 4,300 runs / month 94.7% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Credit card limit increase URL: https://www.aginies.com/en/agents/credit-limit-increase Language: en Summary: The customer types in the app or talks to the call centre; the decision takes seconds, and anything above threshold goes to a manager. "I want to raise my limit" message in the mobile app → New limit written to core banking, customer informed, decision trail stored Bankingappvoice # Credit card limit increase The customer types in the app or talks to the call centre; the decision takes seconds, and anything above threshold goes to a manager. In short Credit card limit increase is an agent scenario in the Banking sector running on the app, voice channels. Trigger: "I want to raise my limit" message in the mobile app. The Card Supervisor coordinates 3 specialist agents (Intent Agent, Risk Agent, Operations Agent) across 3 systems (Core Banking, KYC / Customer 360, Push & SMS Gateway); of 8 steps, 1 write to systems and 2 are checks or approvals. Outcome: New limit written to core banking, customer informed, decision trail stored. Approval rule: Limit > 100,000 → branch manager approval. Banking ### Credit card limit increase [Credit card limit increase] run #41,288,031elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Limit > 100,000 → branch manager approval Agents in this run Card Supervisorsupervisorplans, routes, approves Intent Agentaginies-sunkar-1 · on-premUnderstands the request, extracts amount and card Risk Agentaginies-tulpar-4 · on-premApplies policy against score, income and utilisation Operations Agentaginies-talay-2 · on-premWrites the approved limit to core banking Audit trail00:00.00 - ◦Message received, session identity verified - ∿Intent: limit increase · Amount: 150,000 · Card: ****4412 - ⌘Plan: risk check → policy → write → notify - ⚙Customer profile and 12-month utilisation pulled - ✓Policy v1.3: score 742, utilisation 38% → eligible, above threshold - ⏸Limit > 100,000 → routed to branch manager approval - ⇢Limit updated to 150,000 in core banking - ➤Push and SMS sent, decision trail stored - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger "I want to raise my limit" message in the mobile app Outcome New limit written to core banking, customer informed, decision trail stored ## Agents SUPERVISOR ### Card Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Intent Agent Understands the request, extracts amount and card detect_intentextract_entities aginies-tulpar-4 · on-prem ### Risk Agent Applies policy against score, income and utilisation get_customer_profileget_utilisationpolicy_check aginies-talay-2 · on-prem ### Operations Agent Writes the approved limit to core banking update_card_limitsend_notification ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Card Supervisor | Message received, session identity verified | — | 180 | — | | 02 | reason | Intent Agent | Intent: limit increase · Amount: 150,000 · Card: ****4412 | — | 640 | 410 | | 03 | plan | Card Supervisor | Plan: risk check → policy → write → notify | — | 320 | 220 | | 04 | tool | Risk Agent | Customer profile and 12-month utilisation pulled | KYC / Customer 360 | 410 | — | | 05 | verify | Risk Agent | Policy v1.3: score 742, utilisation 38% → eligible, above threshold | — | 780 | 960 | | 06 | approval | Card Supervisor | Limit > 100,000 → routed to branch manager approval | — | 2,400 | — | | 07 | write | Operations Agent | Limit updated to 150,000 in core banking | Core Banking | 520 | — | | 08 | notify | Operations Agent | Push and SMS sent, decision trail stored | Push & SMS Gateway | 260 | — | Simulation metrics (estimated) 18,400runs / month 96.4%success 5.2 sp50 $0.0019est. cost / run Approval rule Limit > 100,000 → branch manager approval Step profile ingest · 1plan · 1reason · 1tool · 1verify · 1approval · 1write · 1notify · 1 SYSTEMS - Core Bankingcore - KYC / Customer 360db - Push & SMS Gatewaychannel What you need to run it - Read/write API or MCP server and a service account for the core system - Read access to the database or vector store; a data owner - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Conversational banking: the action-taking version of in-app assistants at global banks Packages that include this scenario Conversational banking →Credit and onboarding operations → Sector: Banking → Related agents BankingHITL ### Dialogue banking: voice and text transaction assistant The customer talks or types; bill, limit, dispute and transfer are completed in core banking in one session. 186,000 runs / month 94.7% success 5 agents BankingHITL ### Digital account opening (KYC) ID, liveness and sanctions checks in one flow; clean applications open an account in minutes, no branch visit. 42,000 runs / month 93.8% success 5 agents BankingHITL ### Card fraud alert triage The card is blocked in seconds and the customer asked on WhatsApp; the reply releases or cancels the card. 65,000 runs / month 97.1% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Customs document preparation URL: https://www.aginies.com/en/agents/customs-document-preparation Language: en Summary: HS codes, invoice and packing list prepared in minutes; only low-confidence lines reach a customs specialist. An export order is released as ready to ship in the WMS → Document set generated, handed to the customs broker, shipment marked "docs ready" Logisticsapiemailbatch # Customs document preparation HS codes, invoice and packing list prepared in minutes; only low-confidence lines reach a customs specialist. In short Customs document preparation is an agent scenario in the Logistics sector running on the api, email, batch channels. Trigger: An export order is released as ready to ship in the WMS. The Customs Supervisor coordinates 3 specialist agents (Tariff Agent, Document Agent, Broker Agent) across 5 systems (SAP S/4HANA, WMS, Tariff & Regulation Vector Store, Customs Broker Portal, Document Store (S3)); of 9 steps, 3 write to systems and 2 are checks or approvals. Outcome: Document set generated, handed to the customs broker, shipment marked "docs ready". Approval rule: HS confidence < 0.90 or duty difference > 2,000 € → customs specialist approval. Logistics ### Customs document preparation [Customs document preparation] run #41,287,225elapsed 0 mstokens 0est. cost $0.000status —Approval rule: HS confidence < 0.90 or duty difference > 2,000 € → customs specialist approval Agents in this run Customs Supervisorsupervisorplans, routes, approves Tariff Agentaginies-tulpar-4 · on-premDetermines HS codes against the tariff schedule Document Agentaginies-sunkar-1 · on-premPrepares invoice, packing list and origin documents Broker Agentaginies-talay-2 · on-premHands documents to the broker and updates the shipment Audit trail00:00.00 - ◦Export order EX-20931 to Germany released from the WMS, 22 line items - ⚙Order lines, Incoterm DAP and invoice value of 148,600 € pulled - ⌕Tariff chapters 84 and 85 plus the 2026 EU explanatory notes retrieved - ∿22 lines classified; 19 high-confidence, 3 flagged (e.g. 8479.89 vs 8428.90) - ✓Origin check: 22/22 lines meet EU–TR customs union rules, A.TR eligible - ⏸3 low-confidence HS lines → customs specialist confirmed the codes - ⇢Commercial invoice, packing list and A.TR generated (PDF, 9 pages) - ⇢Document set submitted to the broker, declaration pre-check reference BR-55120 - ⇢Shipment status "customs docs ready", pickup slot Thu 14:00 confirmed - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger An export order is released as ready to ship in the WMS Outcome Document set generated, handed to the customs broker, shipment marked "docs ready" ## Agents SUPERVISOR ### Customs Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-tulpar-4 · on-prem ### Tariff Agent Determines HS codes against the tariff schedule search_tariff_scheduleclassify_hs_codeget_duty_rate aginies-sunkar-1 · on-prem ### Document Agent Prepares invoice, packing list and origin documents get_order_linesbuild_commercial_invoicebuild_packing_listcheck_origin_rules aginies-talay-2 · on-prem ### Broker Agent Hands documents to the broker and updates the shipment submit_to_brokerupdate_shipmentsend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Customs Supervisor | Export order EX-20931 to Germany released from the WMS, 22 line items | — | 200 | — | | 02 | tool | Document Agent | Order lines, Incoterm DAP and invoice value of 148,600 € pulled | SAP S/4HANA | 420 | — | | 03 | retrieve | Tariff Agent | Tariff chapters 84 and 85 plus the 2026 EU explanatory notes retrieved | Tariff & Regulation Vector Store | 480 | 720 | | 04 | reason | Tariff Agent | 22 lines classified; 19 high-confidence, 3 flagged (e.g. 8479.89 vs 8428.90) | — | 1,100 | 1,450 | | 05 | verify | Document Agent | Origin check: 22/22 lines meet EU–TR customs union rules, A.TR eligible | — | 520 | 480 | | 06 | approval | Customs Supervisor | 3 low-confidence HS lines → customs specialist confirmed the codes | — | 2,600 | — | | 07 | write | Document Agent | Commercial invoice, packing list and A.TR generated (PDF, 9 pages) | Document Store (S3) | 640 | — | | 08 | write | Broker Agent | Document set submitted to the broker, declaration pre-check reference BR-55120 | Customs Broker Portal | 560 | — | | 09 | write | Broker Agent | Shipment status "customs docs ready", pickup slot Thu 14:00 confirmed | WMS | 380 | — | Simulation metrics (estimated) 4,300runs / month 94.7%success 6.9 sp50 $0.0105est. cost / run Approval rule HS confidence < 0.90 or duty difference > 2,000 € → customs specialist approval Step profile ingest · 1retrieve · 1reason · 1tool · 1verify · 1approval · 1write · 3 SYSTEMS - SAP S/4HANAerp - WMScore - Tariff & Regulation Vector Storedb - Customs Broker Portalapi - Document Store (S3)storage What you need to run it - Posting and query API access to ERP/MES - Read/write API or MCP server and a service account for the core system - Read access to the database or vector store; a data owner - Credentials and a test environment for the enterprise APIs involved - Read access to the document store and the document types - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors export documentation automation at global freight forwarders and customs brokers Packages that include this scenario Information management & decision support →Logistics control tower → Sector: Logistics → Related agents LogisticsHITL ### Control tower: delay prediction, rerouting, customer notice, SLA The telematics signal catches the delay early; alternative route and carrier chosen, customer informed, SLA compensation computed. 96,000 runs / month 93.4% success 5 agents LogisticsHITL ### Shipment exception handling When a delay or damage lands, impact is costed, the carrier rebooked and the customer hears before they ask. 12,800 runs / month 95.5% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Demand response event URL: https://www.aginies.com/en/agents/demand-response-event Language: en Summary: Under grid stress eligible consumers are selected, the offer sent by channel, participation measured and settlement computed. Load-reduction call from the market operator or SCADA → Participants selected and informed for the target MW; realised reduction measured, settlement computed Energy & Utilitiesapiappwhatsappbatch # Demand response event Under grid stress eligible consumers are selected, the offer sent by channel, participation measured and settlement computed. In short Demand response event is an agent scenario in the Energy & Utilities sector running on the api, app, whatsapp, batch channels. Trigger: Load-reduction call from the market operator or SCADA. The Demand Response Supervisor coordinates 3 specialist agents (Selection Agent, Offer Agent, Settlement Agent) across 4 systems (SCADA / DMS, Meter Data Management, Billing / Settlement, WhatsApp Business API); of 9 steps, 1 write to systems and 2 are checks or approvals. Outcome: Participants selected and informed for the target MW; realised reduction measured, settlement computed. Approval rule: Event > 10 MW → grid operator approval; settlement with an audit trail. Energy & Utilities ### Demand response event [Demand response event] run #41,288,047elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Event > 10 MW → grid operator approval; settlement with an audit trail Agents in this run Demand Response Supervisorsupervisorplans, routes, approves Selection Agentaginies-tulpar-4 · on-premSelects consumers by flexibility potential Offer Agentaginies-talay-2 · on-premSends the offer by channel and collects responses Settlement Agentaginies-sunkar-1 · on-premMeasures reduction from meter data and computes payment Audit trail00:00.00 - ◦Call: 12 MW reduction 17:00–19:00, region: Aegean industrial feeders - ⚙Flexibility portfolio pulled: 214 sites, contracted potential 31 MW - ∿63 sites selected by forecast and past participation; expected reduction 14.8 MW - ⏸Event > 10 MW → grid operator approved the participant list - ➤WhatsApp and app notification to 63 sites: hours, target kW, tariff - ⚙57 accepts, 4 declines, 2 no reply in 20 minutes; committed 13.6 MW - ⚙Post-event 15-minute interval data pulled; baseline computed - ✓Realised reduction 12.9 MW; 51 sites met commitment, 6 partial - ⇢Payments computed and posted to settlement; event report attached to the SCADA record - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Load-reduction call from the market operator or SCADA Outcome Participants selected and informed for the target MW; realised reduction measured, settlement computed ## Agents SUPERVISOR ### Demand Response Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-tulpar-4 · on-prem ### Selection Agent Selects consumers by flexibility potential get_flex_portfolioforecast_reductionrank_participants aginies-talay-2 · on-prem ### Offer Agent Sends the offer by channel and collects responses send_whatsappsend_pushcollect_responses aginies-sunkar-1 · on-prem ### Settlement Agent Measures reduction from meter data and computes payment get_interval_datacompute_baselinecompute_settlementpost_settlement ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Demand Response Supervisor | Call: 12 MW reduction 17:00–19:00, region: Aegean industrial feeders | SCADA / DMS | 200 | — | | 02 | tool | Selection Agent | Flexibility portfolio pulled: 214 sites, contracted potential 31 MW | Meter Data Management | 480 | — | | 03 | reason | Selection Agent | 63 sites selected by forecast and past participation; expected reduction 14.8 MW | — | 820 | 1,100 | | 04 | approval | Demand Response Supervisor | Event > 10 MW → grid operator approved the participant list | — | 2,400 | — | | 05 | notify | Offer Agent | WhatsApp and app notification to 63 sites: hours, target kW, tariff | WhatsApp Business API | 300 | — | | 06 | tool | Offer Agent | 57 accepts, 4 declines, 2 no reply in 20 minutes; committed 13.6 MW | WhatsApp Business API | 420 | — | | 07 | tool | Settlement Agent | Post-event 15-minute interval data pulled; baseline computed | Meter Data Management | 520 | — | | 08 | verify | Settlement Agent | Realised reduction 12.9 MW; 51 sites met commitment, 6 partial | — | 560 | 480 | | 09 | write | Settlement Agent | Payments computed and posted to settlement; event report attached to the SCADA record | Billing / Settlement | 480 | — | Simulation metrics (estimated) 140runs / month 96.2%success 6.2 sp50 $0.0169est. cost / run Approval rule Event > 10 MW → grid operator approval; settlement with an audit trail Step profile ingest · 1reason · 1tool · 3verify · 1approval · 1write · 1notify · 1 SYSTEMS - SCADA / DMScore - Meter Data Managementdb - Billing / Settlementerp - WhatsApp Business APIchannel What you need to run it - Read/write API or MCP server and a service account for the core system - Read access to the database or vector store; a data owner - Posting and query API access to ERP/MES - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors aggregator and DSO demand-response flows in European flexibility markets Packages that include this scenario Sector: Energy & Utilities → Related agents Energy & UtilitiesHITL ### Utility customer assistant: bill, outage, consumption and instalments in one dialogue The customer calls or writes; the high bill is explained, outage ETA given, meter data shown, an instalment plan set up. 520,000 runs / month 94.6% success 5 agents Energy & UtilitiesHITL ### Outage report handling Reports are clustered by feeder, one OSS ticket is raised and everyone affected hears before the crew leaves. 38,000 runs / month 97.7% success 5 agents Energy & UtilitiesHITL ### Meter reading dispute Customer sends a meter photo; the index is read by vision, the estimated bill fixed, big credits approved. 9,200 runs / month 94.9% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Device fleet anomaly triage URL: https://www.aginies.com/en/agents/device-fleet-anomaly-triage Language: en Summary: Telemetry anomalies from thousands of devices are clustered, root-cause candidates listed, one incident reaches the right team. Anomaly alarm or threshold-breach event from the IoT platform → One incident record, affected device list and recommended action; critical commands approved IoT & Connected Devicesapibatchapp # Device fleet anomaly triage Telemetry anomalies from thousands of devices are clustered, root-cause candidates listed, one incident reaches the right team. In short Device fleet anomaly triage is an agent scenario in the IoT & Connected Devices sector running on the api, batch, app channels. Trigger: Anomaly alarm or threshold-breach event from the IoT platform. The Fleet Health Supervisor coordinates 3 specialist agents (Signal Agent, Diagnosis Agent, Operations Agent) across 5 systems (IoT Platform (MQTT / Digital Twin), Time-series DB, Known Issues (vector), ITSM / Incident, Slack); of 9 steps, 1 write to systems and 2 are checks or approvals. Outcome: One incident record, affected device list and recommended action; critical commands approved. Approval rule: Bulk device command > 100 devices → fleet operations lead approval. IoT & Connected Devices ### Device fleet anomaly triage [Device fleet anomaly triage] run #41,287,338elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Bulk device command > 100 devices → fleet operations lead approval Agents in this run Fleet Health Supervisorsupervisorplans, routes, approves Signal Agentaginies-sunkar-1 · on-premNormalises and clusters telemetry Diagnosis Agentaginies-tulpar-4 · on-premEvaluates root cause with firmware, location and event history Operations Agentaginies-talay-2 · on-premOpens the incident, prepares commands, informs the team Audit trail00:00.00 - ◦Alarm: temperature deviation from 312 devices in 20 minutes, region: Marmara warehouses - ⚙Last 6 hours of telemetry pulled; 312 devices split into 3 clusters, largest 284 - ∿Main cluster: same sensor model, same firmware 4.2.1, same gateway group - ⚙Firmware history: 4.2.1 rolled out yesterday 02:00; no deviation on the previous version - ⌕Known issues: 4.2.1 calibration offset bug (KI-0788), workaround: rollback - ✓Root-cause confidence 0.91: firmware offset; physical cooling failure ruled out - ⇢Incident INC-40917 opened, priority P2, 284 devices attached, root-cause note added - ⏸Bulk rollback command > 100 devices → fleet operations lead approved - ➤Team summary: cluster, root cause, approved rollback plan and monitoring window - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Anomaly alarm or threshold-breach event from the IoT platform Outcome One incident record, affected device list and recommended action; critical commands approved ## Agents SUPERVISOR ### Fleet Health Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Signal Agent Normalises and clusters telemetry get_telemetrycluster_anomaliesget_device_metadata aginies-tulpar-4 · on-prem ### Diagnosis Agent Evaluates root cause with firmware, location and event history get_firmware_historysearch_known_issuescorrelate_events aginies-talay-2 · on-prem ### Operations Agent Opens the incident, prepares commands, informs the team create_incidentprepare_device_commandsend_slack ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Fleet Health Supervisor | Alarm: temperature deviation from 312 devices in 20 minutes, region: Marmara warehouses | — | 200 | — | | 02 | tool | Signal Agent | Last 6 hours of telemetry pulled; 312 devices split into 3 clusters, largest 284 | Time-series DB | 520 | — | | 03 | reason | Signal Agent | Main cluster: same sensor model, same firmware 4.2.1, same gateway group | — | 640 | 700 | | 04 | tool | Diagnosis Agent | Firmware history: 4.2.1 rolled out yesterday 02:00; no deviation on the previous version | IoT Platform (MQTT / Digital Twin) | 440 | — | | 05 | retrieve | Diagnosis Agent | Known issues: 4.2.1 calibration offset bug (KI-0788), workaround: rollback | Known Issues (vector) | 480 | 520 | | 06 | verify | Diagnosis Agent | Root-cause confidence 0.91: firmware offset; physical cooling failure ruled out | — | 560 | 480 | | 07 | write | Operations Agent | Incident INC-40917 opened, priority P2, 284 devices attached, root-cause note added | ITSM / Incident | 480 | — | | 08 | approval | Fleet Health Supervisor | Bulk rollback command > 100 devices → fleet operations lead approved | — | 2,600 | — | | 09 | notify | Operations Agent | Team summary: cluster, root cause, approved rollback plan and monitoring window | Slack | 240 | — | Simulation metrics (estimated) 18,000runs / month 96.9%success 6.2 sp50 $0.0019est. cost / run Approval rule Bulk device command > 100 devices → fleet operations lead approval Step profile ingest · 1retrieve · 1reason · 1tool · 2verify · 1approval · 1write · 1notify · 1 SYSTEMS - IoT Platform (MQTT / Digital Twin)core - Time-series DBdb - Known Issues (vector)db - ITSM / Incidentticketing - Slackchannel What you need to run it - Read/write API or MCP server and a service account for the core system - Read access to the database or vector store; a data owner - API access and assignment rules for the ticketing or work-order system - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors AIOps-style alarm clustering and root-cause flows of industrial IoT platforms Packages that include this scenario IoT device fleet operations → Sector: IoT & Connected Devices → Related agents IoT & Connected DevicesHITL ### Field device support assistant: telemetry, remote diagnosis, OTA and technician The customer says "the device is dead"; telemetry read, remote diagnosis made, update pushed, technician dispatched with the right part if needed. 37,000 runs / month 94.1% success 5 agents IoT & Connected DevicesHITL ### Staged OTA update rollout Firmware rolls out in 1% → 10% → 100% stages; a health gate at every stage, human approval for critical fleets. 320 runs / month 98.1% success 4 agents IoT & Connected DevicesHITL ### Cold-chain excursion response A temperature excursion is assessed instantly, product risk computed, a field crew dispatched, the quality decision recorded. 2,700 runs / month 97.4% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Dialogue banking: voice and text transaction assistant URL: https://www.aginies.com/en/agents/dialog-banking-assistant Language: en Summary: The customer talks or types; bill, limit, dispute and transfer are completed in core banking in one session. Multi-transaction request over the mobile app chat, WhatsApp or the call centre → Three transactions executed, each with its own reference; the customer informed with one summary Bankingappvoicewhatsapp # Dialogue banking: voice and text transaction assistant The customer talks or types; bill, limit, dispute and transfer are completed in core banking in one session. In short Dialogue banking: voice and text transaction assistant is an agent scenario in the Banking sector running on the app, voice, whatsapp channels. Trigger: Multi-transaction request over the mobile app chat, WhatsApp or the call centre. The Dialogue Banking Supervisor coordinates 4 specialist agents (Dialogue Agent, Identity Agent, Transaction Agent, Risk Agent) across 5 systems (Mobile App Chat, Core Banking, Card Management, Fraud Engine, SMS Gateway); of 11 steps, 3 write to systems and 2 are checks or approvals. Outcome: Three transactions executed, each with its own reference; the customer informed with one summary. Approval rule: Limit increase > 60%, dispute amount > 5,000 or transfer to a new payee → operations approval. Banking ### Dialogue banking: voice and text transaction assistant [Dialogue banking: voice and text transaction assistant] run #41,289,470elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Limit increase > 60%, dispute amount > 5,000 or transfer to a new payee → operations approval Agents in this run Dialogue Banking Supervisorsupervisorplans, routes, approves Dialogue Agentaginies-talay-2 · on-premRuns the multi-turn conversation, separates several intents, produces spoken and written replies Identity Agentaginies-sunkar-1 · on-premVerifies the session, biometric state and per-transaction OTP thresholds Transaction Agentaginies-tulpar-4 · on-premExecutes bill, limit, transfer and card operations with core-banking tools Risk Agentaginies-sunkar-1 · on-premApplies fraud score, policy thresholds and approval rules Audit trail00:00.00 - ◦App chat: "pay my electricity bill, raise my limit to 40k, I don't recognise yesterday's 1,250" - ∿Three intents separated: bill payment · limit increase · transaction dispute; order and dependencies set - ✓Session biometrically verified, device enrolled; OTP mandatory above 10,000 - ⌘Plan: query bill → pay → limit score → policy → dispute record → one summary - ⚙Electricity bill: 842.60, due in 3 days, no standing order - ⇢Bill paid from the current account, ref #BL-2048816; receipt attached to the chat - ✓Limit 25,000 → 40,000: +60%, score 742, income verified → inside the auto-approval band - ⇢Card limit updated to 40,000, ref #CL-77310; effective immediately - ⚙Disputed transaction: cross-border e-commerce, score 81/100; card temporarily blocked, replacement offered - ⇢Dispute file #DS-10944 opened, card blocked, replacement card ordered - ➤One summary: three transactions with references; read out on voice, SMS and push sent - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Multi-transaction request over the mobile app chat, WhatsApp or the call centre Outcome Three transactions executed, each with its own reference; the customer informed with one summary ## Agents SUPERVISOR ### Dialogue Banking Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Dialogue Agent Runs the multi-turn conversation, separates several intents, produces spoken and written replies detect_intentscompose_replytts_speak aginies-sunkar-1 · on-prem ### Identity Agent Verifies the session, biometric state and per-transaction OTP thresholds check_sessionvoiceprint_matchsend_otpverify_otp aginies-tulpar-4 · on-prem ### Transaction Agent Executes bill, limit, transfer and card operations with core-banking tools query_billpay_billget_card_limitupdate_card_limitopen_disputeblock_card aginies-sunkar-1 · on-prem ### Risk Agent Applies fraud score, policy thresholds and approval rules fraud_scorepolicy_checkroute_approval ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Dialogue Banking Supervisor | App chat: "pay my electricity bill, raise my limit to 40k, I don't recognise yesterday's 1,250" | Mobile App Chat | 180 | — | | 02 | reason | Dialogue Agent | Three intents separated: bill payment · limit increase · transaction dispute; order and dependencies set | — | 620 | 410 | | 03 | verify | Identity Agent | Session biometrically verified, device enrolled; OTP mandatory above 10,000 | — | 340 | 120 | | 04 | plan | Dialogue Banking Supervisor | Plan: query bill → pay → limit score → policy → dispute record → one summary | — | 260 | 240 | | 05 | tool | Transaction Agent | Electricity bill: 842.60, due in 3 days, no standing order | Core Banking | 420 | — | | 06 | write | Transaction Agent | Bill paid from the current account, ref #BL-2048816; receipt attached to the chat | Core Banking | 560 | — | | 07 | verify | Risk Agent | Limit 25,000 → 40,000: +60%, score 742, income verified → inside the auto-approval band | Card Management | 700 | 330 | | 08 | write | Transaction Agent | Card limit updated to 40,000, ref #CL-77310; effective immediately | Card Management | 480 | — | | 09 | tool | Risk Agent | Disputed transaction: cross-border e-commerce, score 81/100; card temporarily blocked, replacement offered | Fraud Engine | 640 | 280 | | 10 | write | Transaction Agent | Dispute file #DS-10944 opened, card blocked, replacement card ordered | Card Management | 520 | — | | 11 | notify | Dialogue Agent | One summary: three transactions with references; read out on voice, SMS and push sent | SMS Gateway | 460 | 220 | Simulation metrics (estimated) 186,000runs / month 94.7%success 7.8 sp50 $0.0037est. cost / run Approval rule Limit increase > 60%, dispute amount > 5,000 or transfer to a new payee → operations approval Step profile ingest · 1plan · 1reason · 1tool · 2verify · 2write · 3notify · 1 SYSTEMS - Mobile App Chatchannel - Core Bankingcore - Card Managementapi - Fraud Engineapi - SMS Gatewaychannel What you need to run it - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - Read/write API or MCP server and a service account for the core system - Credentials and a test environment for the enterprise APIs involved - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern The transacting version of large banks' chat and voice assistants; finishes several transactions in one session Packages that include this scenario Voice AI for claims intake →Conversational banking →Voice analysis & quality →Fraud and dispute operations → Sector: Banking → Related agents BankingHITL ### Credit card limit increase The customer types in the app or talks to the call centre; the decision takes seconds, and anything above threshold goes to a manager. 18,400 runs / month 96.4% success 4 agents BankingHITL ### Digital account opening (KYC) ID, liveness and sanctions checks in one flow; clean applications open an account in minutes, no branch visit. 42,000 runs / month 93.8% success 5 agents BankingHITL ### Card fraud alert triage The card is blocked in seconds and the customer asked on WhatsApp; the reply releases or cancels the card. 65,000 runs / month 97.1% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Email classification and attachment check URL: https://www.aginies.com/en/agents/email-classification-attachment-check Language: en Summary: Every e-mail in the shared mailbox is classified, attachments checked against a list, gaps asked for at once. New e-mail in the operations@ shared mailbox → Request in the right queue, attachments verified and archived, gaps asked of the sender Enterprise Functionsemail # Email classification and attachment check Every e-mail in the shared mailbox is classified, attachments checked against a list, gaps asked for at once. In short Email classification and attachment check is an agent scenario in the Enterprise Functions sector running on the email channel. Trigger: New e-mail in the operations@ shared mailbox. The Mailroom Supervisor coordinates 3 specialist agents (Classification Agent, Attachment Agent, Routing Agent) across 4 systems (Exchange / M365 Mail, Mailbox Rules & Checklists Vector Store, ServiceNow, Document Store (S3)); of 8 steps, 2 write to systems and 1 are checks or approvals. Outcome: Request in the right queue, attachments verified and archived, gaps asked of the sender. Enterprise Functions ### Email classification and attachment check [Email classification and attachment check] run #41,287,321elapsed 0 mstokens 0est. cost $0.000status — Agents in this run Mailroom Supervisorsupervisorplans, routes, approves Classification Agentaginies-sunkar-1 · on-premClassifies the e-mail by topic and language Attachment Agentaginies-konrul-1 · on-premReads attachments, checks type and validity Routing Agentaginies-talay-2 · on-premOpens the ticket and replies to the sender Audit trail00:00.00 - ◦New e-mail in operations@ with 2 attachments (PDF, XLSX) - ∿Class: "supplier onboarding request" · confidence 0.94 · language TR - ⌕Checklist "supplier onboarding" v4: tax certificate, signature circular, IBAN letter - ∿PDF = tax certificate (2025, expired); XLSX = bank form; signature circular missing - ✓Result: 1 missing, 1 expired → cannot proceed; auto-reply permitted by policy - ⇢Ticket SN-INC-771230 opened as "waiting for sender", SLA paused - ⇢Attachments archived under the ticket, hashes recorded - ➤Sender replied with the 2 missing/expired items, e-mail moved to "Pending" - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger New e-mail in the operations@ shared mailbox Outcome Request in the right queue, attachments verified and archived, gaps asked of the sender ## Agents SUPERVISOR ### Mailroom Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Classification Agent Classifies the e-mail by topic and language classify_emailextract_entities aginies-konrul-1 · on-prem ### Attachment Agent Reads attachments, checks type and validity parse_attachmentdetect_document_typesearch_checklistscheck_required_attachments aginies-talay-2 · on-prem ### Routing Agent Opens the ticket and replies to the sender create_ticketarchive_attachmentsreply_emailmove_to_folder ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Mailroom Supervisor | New e-mail in operations@ with 2 attachments (PDF, XLSX) | — | 180 | — | | 02 | reason | Classification Agent | Class: "supplier onboarding request" · confidence 0.94 · language TR | — | 520 | 430 | | 03 | retrieve | Attachment Agent | Checklist "supplier onboarding" v4: tax certificate, signature circular, IBAN letter | Mailbox Rules & Checklists Vector Store | 340 | 280 | | 04 | reason | Attachment Agent | PDF = tax certificate (2025, expired); XLSX = bank form; signature circular missing | — | 880 | 720 | | 05 | verify | Attachment Agent | Result: 1 missing, 1 expired → cannot proceed; auto-reply permitted by policy | — | 300 | 200 | | 06 | write | Routing Agent | Ticket SN-INC-771230 opened as "waiting for sender", SLA paused | ServiceNow | 480 | — | | 07 | write | Routing Agent | Attachments archived under the ticket, hashes recorded | Document Store (S3) | 260 | — | | 08 | notify | Routing Agent | Sender replied with the 2 missing/expired items, e-mail moved to "Pending" | Exchange / M365 Mail | 240 | — | Simulation metrics (estimated) 86,000runs / month 97.6%success 3.2 sp50 $0.0006est. cost / run Step profile ingest · 1retrieve · 1reason · 2verify · 1write · 2notify · 1 SYSTEMS - Exchange / M365 Mailemail - Mailbox Rules & Checklists Vector Storedb - ServiceNowticketing - Document Store (S3)storage What you need to run it - Mailbox access (Microsoft 365 or Google Workspace) - Read access to the database or vector store; a data owner - API access and assignment rules for the ticketing or work-order system - Read access to the document store and the document types - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors intelligent mailroom automation at insurers and shared-service centres Packages that include this scenario Email classification & attachment verification →HR & employee experience →KVKK / GDPR personal data protection → Related agents Enterprise FunctionsHITL ### Procure-to-pay desk: request, quotes, contract and invoice in one place The request arrives by email; quotes compared, contract clauses checked, PO opened, invoice three-way matched and payment scheduled. 14,000 runs / month 92.7% success 5 agents Enterprise FunctionsHITL ### Call centre quality monitoring Every call scored against the rubric, not a sample; compliance breaches and coaching flags ready by morning. 90,000 runs / month 98.8% success 4 agents Enterprise FunctionsHITL ### Contract clause extraction Clauses are extracted and compared with the legal playbook; deviations reach counsel with redline suggestions. 1,400 runs / month 95.1% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Employee onboarding URL: https://www.aginies.com/en/agents/employee-onboarding Language: en Summary: Accounts, kit and training plan ready the moment HR creates the hire; off-template access goes to the manager. A new hire record is approved in SuccessFactors → Accounts created, equipment ordered, training plan assigned, welcome e-mail sent Enterprise Functionsapiemailapp # Employee onboarding Accounts, kit and training plan ready the moment HR creates the hire; off-template access goes to the manager. In short Employee onboarding is an agent scenario in the Enterprise Functions sector running on the api, email, app channels. Trigger: A new hire record is approved in SuccessFactors. The Onboarding Supervisor coordinates 4 specialist agents (HR Agent, Identity Agent, IT Agent, Training Agent) across 5 systems (SAP SuccessFactors, Active Directory / Entra ID, ServiceNow, LMS, Exchange / M365 Mail); of 9 steps, 3 write to systems and 2 are checks or approvals. Outcome: Accounts created, equipment ordered, training plan assigned, welcome e-mail sent. Approval rule: Access outside the role template or admin rights → line manager approval. Enterprise Functions ### Employee onboarding [Employee onboarding] run #41,287,124elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Access outside the role template or admin rights → line manager approval Agents in this run Onboarding Supervisorsupervisorplans, routes, approves HR Agentaginies-sunkar-1 · on-premReads the hire record and role profile Identity Agentaginies-talay-2 · on-premCreates account, groups and mailbox IT Agentaginies-sunkar-1 · on-premOrders equipment from stock Training Agentaginies-tulpar-4 · on-premBuilds the training plan and sends the welcome pack Audit trail00:00.00 - ◦SuccessFactors event: new hire E-40718, Data Analyst, start 15 Sep, Istanbul - ⚙Role profile: cost centre 4410, manager M-2210, access template "Analytics-L2" - ⌘Plan: identity → equipment → training → welcome (parallel where possible) - ✓SoD check: "Analytics-L2" + finance read → no conflict; MFA required - ⇢Account created, 6 groups assigned, mailbox provisioned, MFA enrolment link issued - ⇢Laptop (14", 32 GB) ordered from stock, delivery to Istanbul office 12 Sep - ⏸Off-template "finance read" access → line manager approved - ⇢Training plan: 6 courses (KVKK, security, SQL basics), due within 30 days - ➤Welcome e-mail to the hire, checklist to manager and buddy, day-1 agenda attached - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger A new hire record is approved in SuccessFactors Outcome Accounts created, equipment ordered, training plan assigned, welcome e-mail sent ## Agents SUPERVISOR ### Onboarding Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### HR Agent Reads the hire record and role profile get_new_hireget_role_profile aginies-talay-2 · on-prem ### Identity Agent Creates account, groups and mailbox check_sod_rulescreate_userassign_groupscreate_mailbox aginies-sunkar-1 · on-prem ### IT Agent Orders equipment from stock check_stockorder_equipmentcreate_ticket aginies-tulpar-4 · on-prem ### Training Agent Builds the training plan and sends the welcome pack build_training_planenrol_coursessend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Onboarding Supervisor | SuccessFactors event: new hire E-40718, Data Analyst, start 15 Sep, Istanbul | — | 200 | — | | 02 | tool | HR Agent | Role profile: cost centre 4410, manager M-2210, access template "Analytics-L2" | SAP SuccessFactors | 420 | — | | 03 | plan | Onboarding Supervisor | Plan: identity → equipment → training → welcome (parallel where possible) | — | 280 | 200 | | 04 | verify | Identity Agent | SoD check: "Analytics-L2" + finance read → no conflict; MFA required | — | 360 | 240 | | 05 | write | Identity Agent | Account created, 6 groups assigned, mailbox provisioned, MFA enrolment link issued | Active Directory / Entra ID | 620 | — | | 06 | write | IT Agent | Laptop (14", 32 GB) ordered from stock, delivery to Istanbul office 12 Sep | ServiceNow | 480 | — | | 07 | approval | Onboarding Supervisor | Off-template "finance read" access → line manager approved | — | 2,400 | — | | 08 | write | Training Agent | Training plan: 6 courses (KVKK, security, SQL basics), due within 30 days | LMS | 440 | — | | 09 | notify | Training Agent | Welcome e-mail to the hire, checklist to manager and buddy, day-1 agenda attached | Exchange / M365 Mail | 260 | — | Simulation metrics (estimated) 1,900runs / month 97.2%success 5.5 sp50 $0.0028est. cost / run Approval rule Access outside the role template or admin rights → line manager approval Step profile ingest · 1plan · 1tool · 1verify · 1approval · 1write · 3notify · 1 SYSTEMS - SAP SuccessFactorshr - Active Directory / Entra IDcore - ServiceNowticketing - LMShr - Exchange / M365 Mailemail What you need to run it - HR system API access and personal-data boundaries - Read/write API or MCP server and a service account for the core system - API access and assignment rules for the ticketing or work-order system - Mailbox access (Microsoft 365 or Google Workspace) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors zero-touch joiner / mover / leaver automation at large enterprises Packages that include this scenario HR & employee experience → Related agents Enterprise FunctionsHITL ### Procure-to-pay desk: request, quotes, contract and invoice in one place The request arrives by email; quotes compared, contract clauses checked, PO opened, invoice three-way matched and payment scheduled. 14,000 runs / month 92.7% success 5 agents Enterprise Functions ### Email classification and attachment check Every e-mail in the shared mailbox is classified, attachments checked against a list, gaps asked for at once. 86,000 runs / month 97.6% success 4 agents Enterprise FunctionsHITL ### Call centre quality monitoring Every call scored against the rubric, not a sample; compliance breaches and coaching flags ready by morning. 90,000 runs / month 98.8% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Export control and end-user screening URL: https://www.aginies.com/en/agents/export-control-screening Language: en Summary: Part, buyer and end use are classified in one flow; licence need is detected, the decision always stays with the compliance officer. New export order or quotation request in the ERP → Control class, party screening and licence need documented; order released or held by a compliance decision Defence & Aerospaceapiemailbatch # Export control and end-user screening Part, buyer and end use are classified in one flow; licence need is detected, the decision always stays with the compliance officer. In short Export control and end-user screening is an agent scenario in the Defence & Aerospace sector running on the api, email, batch channels. Trigger: New export order or quotation request in the ERP. The Export Compliance Supervisor coordinates 3 specialist agents (Classification Agent, Party Screening Agent, Licence Agent) across 4 systems (ERP (SAP GTS), Control & Sanctions Lists (vector), Trade Compliance Case System, E-mail); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: Control class, party screening and licence need documented; order released or held by a compliance decision. Approval rule: Every controlled item → export compliance officer approval; the agent cannot make licence decisions. Defence & Aerospace ### Export control and end-user screening [Export control and end-user screening] run #41,287,471elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Every controlled item → export compliance officer approval; the agent cannot make licence decisions Agents in this run Export Compliance Supervisorsupervisorplans, routes, approves Classification Agentaginies-tulpar-4 · on-premMaps the part to control lists Party Screening Agentaginies-talay-2 · on-premScreens buyer, intermediary and end user against sanctions lists Licence Agentaginies-tulpar-4 · on-premDetermines licence need and prepares the file Audit trail00:00.00 - ◦Order SO-88120: 40 RF modules, buyer an overseas systems integrator - ⚙Item master pulled: frequency range, output power, technical data sheet - ⌕Control lists searched: dual-use category 5A001 candidate, national list annex B - ∿Parameters above threshold → controlled; classification rationale and source clauses written - ⚙Buyer, end user and forwarder screened: 0 hits; ownership chain resolved 3 levels - ✓End-use statement present and signed; stated use consistent with the classification - ∿Licence required: individual export licence; application draft and annex list prepared - ⏸Controlled item → export compliance officer reviewed and approved the application - ⇢Compliance case TC-2026-0419 opened; order set to “awaiting licence” - ➤Sales and logistics notified: shipment blocked until licence, estimated duration attached - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger New export order or quotation request in the ERP Outcome Control class, party screening and licence need documented; order released or held by a compliance decision ## Agents SUPERVISOR ### Export Compliance Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-tulpar-4 · on-prem ### Classification Agent Maps the part to control lists get_item_mastersearch_control_listsclassify_dual_use aginies-talay-2 · on-prem ### Party Screening Agent Screens buyer, intermediary and end user against sanctions lists screen_partyresolve_ownershipcheck_end_use_statement aginies-tulpar-4 · on-prem ### Licence Agent Determines licence need and prepares the file search_regulationdraft_licence_applicationcreate_compliance_case ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Export Compliance Supervisor | Order SO-88120: 40 RF modules, buyer an overseas systems integrator | — | 200 | — | | 02 | tool | Classification Agent | Item master pulled: frequency range, output power, technical data sheet | ERP (SAP GTS) | 420 | — | | 03 | retrieve | Classification Agent | Control lists searched: dual-use category 5A001 candidate, national list annex B | Control & Sanctions Lists (vector) | 640 | 900 | | 04 | reason | Classification Agent | Parameters above threshold → controlled; classification rationale and source clauses written | — | 760 | 1,100 | | 05 | tool | Party Screening Agent | Buyer, end user and forwarder screened: 0 hits; ownership chain resolved 3 levels | Control & Sanctions Lists (vector) | 520 | — | | 06 | verify | Party Screening Agent | End-use statement present and signed; stated use consistent with the classification | — | 480 | 420 | | 07 | reason | Licence Agent | Licence required: individual export licence; application draft and annex list prepared | — | 820 | 1,300 | | 08 | approval | Export Compliance Supervisor | Controlled item → export compliance officer reviewed and approved the application | — | 3,000 | — | | 09 | write | Licence Agent | Compliance case TC-2026-0419 opened; order set to “awaiting licence” | Trade Compliance Case System | 480 | — | | 10 | notify | Licence Agent | Sales and logistics notified: shipment blocked until licence, estimated duration attached | E-mail | 240 | — | Simulation metrics (estimated) 2,600runs / month 96.1%success 7.6 sp50 $0.0105est. cost / run Approval rule Every controlled item → export compliance officer approval; the agent cannot make licence decisions Step profile ingest · 1retrieve · 1reason · 2tool · 2verify · 1approval · 1write · 1notify · 1 SYSTEMS - ERP (SAP GTS)erp - Control & Sanctions Lists (vector)db - Trade Compliance Case Systemticketing - E-mailemail What you need to run it - Posting and query API access to ERP/MES - Read access to the database or vector store; a data owner - API access and assignment rules for the ticketing or work-order system - Mailbox access (Microsoft 365 or Google Workspace) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors export-compliance flows of European and US defence suppliers: automated classification, human decision Packages that include this scenario Defence export control & supply-chain security → Sector: Defence & Aerospace → Related agents Defence & AerospaceHITL ### Technical publication assistant: maintenance manual, parts, procedure and revision control The technician asks in the hangar; the right-revision procedure, part number and safety note arrive with clearance checks, every access traced. 9,800 runs / month 95.7% success 5 agents Defence & AerospaceHITL ### Platform maintenance (MRO) work order preparation The fault report is matched to technical publications, task card and parts list are produced, the work order opens with planner approval. 4,100 runs / month 95.7% success 5 agents Defence & AerospaceHITL ### Supplier security and compliance assessment Supplier questionnaires and documents are read, compared against facility security requirements, gaps listed; the qualification decision belongs to the security manager. 900 runs / month 95.2% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Feature delivery pipeline URL: https://www.aginies.com/en/agents/feature-delivery-pipeline Language: en Summary: A Jira epic turns into criteria, design, code and tests via agents; humans only review and approve the PR. Epic moved to the "Ready" column in Jira (webhook) → PR merged to main with human approval, tests green, Jira and Confluence up to date Software Deliveryapiweb # Feature delivery pipeline A Jira epic turns into criteria, design, code and tests via agents; humans only review and approve the PR. In short Feature delivery pipeline is an agent scenario in the Software Delivery sector running on the api, web channels. Trigger: Epic moved to the "Ready" column in Jira (webhook). The Delivery Supervisor coordinates 4 specialist agents (Analyst Agent, Architect Agent, Developer Agent, Tester Agent) across 4 systems (Jira, GitHub, CI Runner (GitHub Actions), Confluence); of 10 steps, 3 write to systems and 2 are checks or approvals. Outcome: PR merged to main with human approval, tests green, Jira and Confluence up to date. Approval rule: Every PR → at least 1 human code review approval; agents never push to main directly. Software Delivery ### Feature delivery pipeline [Feature delivery pipeline] run #41,288,756elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Every PR → at least 1 human code review approval; agents never push to main directly Agents in this run Delivery Supervisorsupervisorplans, routes, approves Analyst Agentaginies-tulpar-4 · on-premBreaks the epic into user stories and acceptance criteria Architect Agentaginies-tulpar-4 · on-premWrites the technical design, identifies affected services and the API contract Developer Agentaginies-tulpar-4 · on-premCreates the branch, writes the code, builds, opens the PR Tester Agentaginies-sunkar-1 · on-premWrites tests from acceptance criteria, runs them, reports coverage Audit trail00:00.00 - ◦Epic PROJ-4821 "Invoice PDF export" moved to Ready, 3 sub-tasks - ⌕Epic description, comments and 2 linked design files read - ∿3 user stories and 11 acceptance criteria written (Gherkin), PROJ-4821 updated - ⌘Plan: design → branch + code → tests → PR → human review → merge - ⇢Design note published: billing-service + new /invoices/{id}/pdf endpoint - ⇢Branch feature/PROJ-4821 created, 6 files changed (+412 / −38), build green - ⚙CI: 14 unit + 4 integration tests written and run, 18/18 passed, coverage 87% - ✓Acceptance criteria 11/11 met, coverage above the 80% threshold → PR #2291 ready - ⏸PR #2291 → senior engineer reviewed, 2 comments addressed, approved - ⇢PR #2291 squash-merged to main; PROJ-4821 moved to "Done" - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Epic moved to the "Ready" column in Jira (webhook) Outcome PR merged to main with human approval, tests green, Jira and Confluence up to date ## Agents SUPERVISOR ### Delivery Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-tulpar-4 · on-prem ### Analyst Agent Breaks the epic into user stories and acceptance criteria read_jira_issuewrite_acceptance_criteriaupdate_jira_issue aginies-tulpar-4 · on-prem ### Architect Agent Writes the technical design, identifies affected services and the API contract search_codebasedraft_design_docpublish_confluence_page aginies-tulpar-4 · on-prem ### Developer Agent Creates the branch, writes the code, builds, opens the PR create_branchedit_filesrun_buildopen_pull_request aginies-sunkar-1 · on-prem ### Tester Agent Writes tests from acceptance criteria, runs them, reports coverage generate_testsrun_testsreport_coverage ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Delivery Supervisor | Epic PROJ-4821 "Invoice PDF export" moved to Ready, 3 sub-tasks | — | 200 | — | | 02 | retrieve | Analyst Agent | Epic description, comments and 2 linked design files read | Jira | 420 | — | | 03 | reason | Analyst Agent | 3 user stories and 11 acceptance criteria written (Gherkin), PROJ-4821 updated | — | 1,600 | 2,400 | | 04 | plan | Delivery Supervisor | Plan: design → branch + code → tests → PR → human review → merge | — | 300 | 220 | | 05 | write | Architect Agent | Design note published: billing-service + new /invoices/{id}/pdf endpoint | Confluence | 1,900 | 3,100 | | 06 | write | Developer Agent | Branch feature/PROJ-4821 created, 6 files changed (+412 / −38), build green | GitHub | 3,600 | 9,800 | | 07 | tool | Tester Agent | CI: 14 unit + 4 integration tests written and run, 18/18 passed, coverage 87% | CI Runner (GitHub Actions) | 2,800 | 4,200 | | 08 | verify | Tester Agent | Acceptance criteria 11/11 met, coverage above the 80% threshold → PR #2291 ready | — | 640 | 520 | | 09 | approval | Delivery Supervisor | PR #2291 → senior engineer reviewed, 2 comments addressed, approved | — | 4,000 | — | | 10 | write | Developer Agent | PR #2291 squash-merged to main; PROJ-4821 moved to "Done" | GitHub | 520 | — | Simulation metrics (estimated) 1,200runs / month 91.2%success 15.5 sp50 $0.0173est. cost / run Approval rule Every PR → at least 1 human code review approval; agents never push to main directly Step profile ingest · 1plan · 1retrieve · 1reason · 1tool · 1verify · 1approval · 1write · 3 SYSTEMS - Jiraticketing - GitHubapi - CI Runner (GitHub Actions)api - Confluencestorage What you need to run it - API access and assignment rules for the ticketing or work-order system - Credentials and a test environment for the enterprise APIs involved - Read access to the document store and the document types - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors agentic SDLC pipelines emerging at large software organisations Packages that include this scenario SDLC agent team → Sector: Software Delivery → Related agents Software DeliveryHITL ### Legacy module modernisation: analysis, tests, rewrite, PR Starts from a Jira epic; the old module is read, its behaviour pinned with tests, rewritten and opened as a PR with evidence. 1,900 runs / month 90.4% success 5 agents Software DeliveryHITL ### CI/CD release pipeline Build, tests, scans and staging run unattended; production ships only on release-manager approval, via ArgoCD. 950 runs / month 94.7% success 5 agents Software DeliveryHITL ### Incident response and hotfix Alert fires, bug reproduced, patch and regression run ready; on-call approves, post-mortem already drafted. 320 runs / month 90.5% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Field device support assistant: telemetry, remote diagnosis, OTA and technician URL: https://www.aginies.com/en/agents/field-device-support-assistant Language: en Summary: The customer says "the device is dead"; telemetry read, remote diagnosis made, update pushed, technician dispatched with the right part if needed. Device fault report over WhatsApp or the support portal → Fault fixed remotely or technician assigned; device record and knowledge base updated IoT & Connected Deviceswhatsappwebapi # Field device support assistant: telemetry, remote diagnosis, OTA and technician The customer says "the device is dead"; telemetry read, remote diagnosis made, update pushed, technician dispatched with the right part if needed. In short Field device support assistant: telemetry, remote diagnosis, OTA and technician is an agent scenario in the IoT & Connected Devices sector running on the whatsapp, web, api channels. Trigger: Device fault report over WhatsApp or the support portal. The Device Support Supervisor coordinates 4 specialist agents (Dialogue Agent, Telemetry Agent, Diagnosis Agent, Dispatch Agent) across 5 systems (WhatsApp Business, Device Cloud (MQTT), Fleet Registry, OTA Service, Field Service); of 10 steps, 1 write to systems and 3 are checks or approvals. Outcome: Fault fixed remotely or technician assigned; device record and knowledge base updated. Approval rule: Bulk OTA (>100 devices) or out-of-warranty device replacement → product-owner approval. IoT & Connected Devices ### Field device support assistant: telemetry, remote diagnosis, OTA and technician [Field device support assistant: telemetry, remote diagnosis, OTA and technician] run #41,287,986elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Bulk OTA (>100 devices) or out-of-warranty device replacement → product-owner approval Agents in this run Device Support Supervisorsupervisorplans, routes, approves Dialogue Agentaginies-talay-2 · on-premTalks with the customer, pins down the device and the symptom Telemetry Agentaginies-sunkar-1 · on-premReads recent telemetry, error codes and connectivity history Diagnosis Agentaginies-tulpar-4 · on-premMatches symptom and data with known faults, decides remote fix or field visit Dispatch Agentaginies-sunkar-1 · on-premAssigns technician and part, books the visit Audit trail00:00.00 - ◦WhatsApp: "boiler controller shows E07 since morning, no heating" + serial number photo - ∿Device SN 8C-220914 found in the fleet registry, under warranty, firmware 2.3.1 - ⚙Telemetry: 14 restarts in 6 hours, temperature sensor reads −40 °C, connectivity stable - ⌘Plan: known-issue match → remote command → OTA → verify → field visit if needed - ∿Match: E07 + −40 °C reading = known sensor driver bug in 2.3.1 (bulletin KB-412), fixed in 2.3.4 - ⇢OTA 2.3.4 pushed to the single device, installed in 3 minutes, device restarted - ✓Verification: sensor 21.5 °C, error code clear, no restart for 15 minutes - ✓No field visit needed; staged OTA proposal created for the 1,240 devices on the same firmware - ⏸Bulk OTA (>100 devices) submitted for product-owner approval; to start with a 5% pilot wave - ➤To the customer: fixed, what was done in plain language; device record and knowledge base updated - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Device fault report over WhatsApp or the support portal Outcome Fault fixed remotely or technician assigned; device record and knowledge base updated ## Agents SUPERVISOR ### Device Support Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Dialogue Agent Talks with the customer, pins down the device and the symptom identify_devicedetect_intentcompose_reply aginies-sunkar-1 · on-prem ### Telemetry Agent Reads recent telemetry, error codes and connectivity history get_telemetryget_error_logget_connectivity aginies-tulpar-4 · on-prem ### Diagnosis Agent Matches symptom and data with known faults, decides remote fix or field visit match_known_issuesremote_commandpush_ota aginies-sunkar-1 · on-prem ### Dispatch Agent Assigns technician and part, books the visit find_technicianreserve_partbook_visit ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Device Support Supervisor | WhatsApp: "boiler controller shows E07 since morning, no heating" + serial number photo | WhatsApp Business | 180 | — | | 02 | reason | Dialogue Agent | Device SN 8C-220914 found in the fleet registry, under warranty, firmware 2.3.1 | Fleet Registry | 700 | 280 | | 03 | tool | Telemetry Agent | Telemetry: 14 restarts in 6 hours, temperature sensor reads −40 °C, connectivity stable | Device Cloud (MQTT) | 640 | — | | 04 | plan | Device Support Supervisor | Plan: known-issue match → remote command → OTA → verify → field visit if needed | — | 240 | 220 | | 05 | reason | Diagnosis Agent | Match: E07 + −40 °C reading = known sensor driver bug in 2.3.1 (bulletin KB-412), fixed in 2.3.4 | — | 1,300 | 760 | | 06 | write | Diagnosis Agent | OTA 2.3.4 pushed to the single device, installed in 3 minutes, device restarted | OTA Service | 3,200 | — | | 07 | verify | Telemetry Agent | Verification: sensor 21.5 °C, error code clear, no restart for 15 minutes | Device Cloud (MQTT) | 900 | — | | 08 | verify | Dispatch Agent | No field visit needed; staged OTA proposal created for the 1,240 devices on the same firmware | Fleet Registry | 520 | 240 | | 09 | approval | Dispatch Agent | Bulk OTA (>100 devices) submitted for product-owner approval; to start with a 5% pilot wave | OTA Service | 400 | — | | 10 | notify | Dialogue Agent | To the customer: fixed, what was done in plain language; device record and knowledge base updated | WhatsApp Business | 440 | 240 | Simulation metrics (estimated) 37,000runs / month 94.1%success 8.5 sp50 $0.0027est. cost / run Approval rule Bulk OTA (>100 devices) or out-of-warranty device replacement → product-owner approval Step profile ingest · 1plan · 1reason · 2tool · 1verify · 2approval · 1write · 1notify · 1 SYSTEMS - WhatsApp Businesschannel - Device Cloud (MQTT)api - Fleet Registrydb - OTA Serviceapi - Field Serviceticketing What you need to run it - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - Credentials and a test environment for the enterprise APIs involved - Read access to the database or vector store; a data owner - API access and assignment rules for the ticketing or work-order system - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Connected-device makers' "fix remotely first" support pattern, combined with fleet-wide OTA Packages that include this scenario IoT device fleet operations → Sector: IoT & Connected Devices → Related agents IoT & Connected DevicesHITL ### Device fleet anomaly triage Telemetry anomalies from thousands of devices are clustered, root-cause candidates listed, one incident reaches the right team. 18,000 runs / month 96.9% success 4 agents IoT & Connected DevicesHITL ### Staged OTA update rollout Firmware rolls out in 1% → 10% → 100% stages; a health gate at every stage, human approval for critical fleets. 320 runs / month 98.1% success 4 agents IoT & Connected DevicesHITL ### Cold-chain excursion response A temperature excursion is assessed instantly, product risk computed, a field crew dispatched, the quality decision recorded. 2,700 runs / month 97.4% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Card fraud alert triage URL: https://www.aginies.com/en/agents/fraud-alert-triage Language: en Summary: The card is blocked in seconds and the customer asked on WhatsApp; the reply releases or cancels the card. Fraud engine alert event with score ≥ 85 → Card status settled, case opened, customer informed, loss contained Bankingwhatsappsmsvoiceapi # Card fraud alert triage The card is blocked in seconds and the customer asked on WhatsApp; the reply releases or cancels the card. In short Card fraud alert triage is an agent scenario in the Banking sector running on the whatsapp, sms, voice, api channels. Trigger: Fraud engine alert event with score ≥ 85. The Fraud Supervisor coordinates 3 specialist agents (Triage Agent, Outreach Agent, Card Agent) across 5 systems (Fraud Engine, Card Management, WhatsApp Business API, Salesforce CRM, SMS Gateway); of 10 steps, 2 write to systems and 2 are checks or approvals. Outcome: Card status settled, case opened, customer informed, loss contained. Approval rule: Confirmed fraud loss > 2,500 → analyst approval before permanent cancellation. Banking ### Card fraud alert triage [Card fraud alert triage] run #41,288,375elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Confirmed fraud loss > 2,500 → analyst approval before permanent cancellation Agents in this run Fraud Supervisorsupervisorplans, routes, approves Triage Agentaginies-tulpar-4 · on-premAssesses the alert against transaction history and device data Outreach Agentaginies-sunkar-1 · on-premAsks the customer on WhatsApp/SMS and interprets the reply Card Agentaginies-talay-2 · on-premBlocks, releases or cancels the card; opens the case Audit trail00:00.00 - ◦Alert #FR-88213: 3 transactions in 3 minutes on ****7710, risk score 91 - ⚙Last 30 days of transactions and device fingerprint pulled - ∿Pattern: cross-border e-commerce, 4,850, unusual hour → likely fraud - ⌘Plan: temporary block → ask customer → permanent cancel or release by reply - ⇢Card temporarily blocked, 2 pending transactions declined - ➤WhatsApp: "Did you make the 4,850 purchase? YES / NO" - ✓Reply from the registered number, "NO" in 48 seconds → fraud confirmed - ⏸Loss > 2,500 → fraud analyst approved permanent cancellation - ⇢Card cancelled, replacement ordered, case #CS-40917 opened - ➤Summary SMS to customer: new card delivered within 3 working days - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Fraud engine alert event with score ≥ 85 Outcome Card status settled, case opened, customer informed, loss contained ## Agents SUPERVISOR ### Fraud Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-tulpar-4 · on-prem ### Triage Agent Assesses the alert against transaction history and device data get_alertget_recent_transactionsget_device_fingerprintscore_pattern aginies-sunkar-1 · on-prem ### Outreach Agent Asks the customer on WhatsApp/SMS and interprets the reply send_whatsappsend_smsparse_reply aginies-talay-2 · on-prem ### Card Agent Blocks, releases or cancels the card; opens the case block_cardunblock_cardcancel_cardorder_replacementcreate_case ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Fraud Supervisor | Alert #FR-88213: 3 transactions in 3 minutes on ****7710, risk score 91 | — | 160 | — | | 02 | tool | Triage Agent | Last 30 days of transactions and device fingerprint pulled | Fraud Engine | 380 | — | | 03 | reason | Triage Agent | Pattern: cross-border e-commerce, 4,850, unusual hour → likely fraud | — | 720 | 880 | | 04 | plan | Fraud Supervisor | Plan: temporary block → ask customer → permanent cancel or release by reply | — | 280 | 190 | | 05 | write | Card Agent | Card temporarily blocked, 2 pending transactions declined | Card Management | 430 | — | | 06 | notify | Outreach Agent | WhatsApp: "Did you make the 4,850 purchase? YES / NO" | WhatsApp Business API | 260 | — | | 07 | verify | Outreach Agent | Reply from the registered number, "NO" in 48 seconds → fraud confirmed | — | 340 | 150 | | 08 | approval | Fraud Supervisor | Loss > 2,500 → fraud analyst approved permanent cancellation | — | 2,600 | — | | 09 | write | Card Agent | Card cancelled, replacement ordered, case #CS-40917 opened | Salesforce CRM | 520 | — | | 10 | notify | Outreach Agent | Summary SMS to customer: new card delivered within 3 working days | SMS Gateway | 220 | — | Simulation metrics (estimated) 65,000runs / month 97.1%success 5.9 sp50 $0.0016est. cost / run Approval rule Confirmed fraud loss > 2,500 → analyst approval before permanent cancellation Step profile ingest · 1plan · 1reason · 1tool · 1verify · 1approval · 1write · 2notify · 2 SYSTEMS - Fraud Engineapi - Card Managementcore - WhatsApp Business APIchannel - Salesforce CRMcrm - SMS Gatewaychannel What you need to run it - Credentials and a test environment for the enterprise APIs involved - Read/write API or MCP server and a service account for the core system - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - CRM API access and a customer matching key - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the two-way SMS/WhatsApp fraud confirmation pattern of large card issuers, extended to take action Packages that include this scenario Conversational banking →Fraud and dispute operations → Sector: Banking → Related agents BankingHITL ### Dialogue banking: voice and text transaction assistant The customer talks or types; bill, limit, dispute and transfer are completed in core banking in one session. 186,000 runs / month 94.7% success 5 agents BankingHITL ### Credit card limit increase The customer types in the app or talks to the call centre; the decision takes seconds, and anything above threshold goes to a manager. 18,400 runs / month 96.4% success 4 agents BankingHITL ### Digital account opening (KYC) ID, liveness and sanctions checks in one flow; clean applications open an account in minutes, no branch visit. 42,000 runs / month 93.8% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Incident response and hotfix URL: https://www.aginies.com/en/agents/incident-hotfix Language: en Summary: Alert fires, bug reproduced, patch and regression run ready; on-call approves, post-mortem already drafted. Grafana alert: 5xx rate > 2% on the payments API (PagerDuty P1) → Fix in production, error rate back to normal, incident record and post-mortem draft ready Software Deliveryapiweb # Incident response and hotfix Alert fires, bug reproduced, patch and regression run ready; on-call approves, post-mortem already drafted. In short Incident response and hotfix is an agent scenario in the Software Delivery sector running on the api, web channels. Trigger: Grafana alert: 5xx rate > 2% on the payments API (PagerDuty P1). The Incident Supervisor coordinates 4 specialist agents (Developer Agent, Tester Agent, Release Agent, Post-mortem Agent) across 5 systems (Observability (Grafana / Datadog), GitHub, Kubernetes / ArgoCD, Confluence, Slack); of 10 steps, 3 write to systems and 2 are checks or approvals. Outcome: Fix in production, error rate back to normal, incident record and post-mortem draft ready. Approval rule: Every emergency production deploy → on-call engineer approval; the hotfix PR still gets a human review. Software Delivery ### Incident response and hotfix [Incident response and hotfix] run #41,287,907elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Every emergency production deploy → on-call engineer approval; the hotfix PR still gets a human review Agents in this run Incident Supervisorsupervisorplans, routes, approves Developer Agentaginies-tulpar-4 · on-premReads logs and traces, reproduces the bug, writes the patch Tester Agentaginies-tulpar-4 · on-premRuns the regression suite, checks the patch for side effects Release Agentaginies-talay-2 · on-premDeploys the hotfix via ArgoCD, watches health Post-mortem Agentaginies-sunkar-1 · on-premWrites the timeline and root-cause draft Audit trail00:00.00 - ◦P1 alert: payments-api 5xx rate 6.8% (threshold 2%) since 14:02; INC-3312 opened - ⚙Last 30 min of logs and traces pulled: NullPointer starts with the 14:01 deploy v2.14.1 - ∿Reproduced: null currency field → PaymentMapper line 118, 3-line patch - ⌘Plan: patch branch → regression → on-call approval → deploy → watch → post-mortem - ⇢Branch hotfix/INC-3312 created, PR #2304: patch + regression test (+21 / −2) - ✓Regression: 1,284 tests passed, new test catches the bug, no side effects - ⏸On-call engineer approved PR #2304 and the emergency deploy from Slack (4 min) - ⇢v2.14.2 deployed to production; 5xx rate down to 0.1% within 6 min - ⇢Post-mortem draft written to Confluence: detect 3 min, resolve 31 min, 5 actions - ➤Resolution summary posted to #incidents, INC-3312 set to "resolved" - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Grafana alert: 5xx rate > 2% on the payments API (PagerDuty P1) Outcome Fix in production, error rate back to normal, incident record and post-mortem draft ready ## Agents SUPERVISOR ### Incident Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-tulpar-4 · on-prem ### Developer Agent Reads logs and traces, reproduces the bug, writes the patch query_logsget_tracesreproduce_bugcreate_branchopen_pull_request aginies-tulpar-4 · on-prem ### Tester Agent Runs the regression suite, checks the patch for side effects run_testsrun_regressioncompare_baseline aginies-talay-2 · on-prem ### Release Agent Deploys the hotfix via ArgoCD, watches health deploy_argocdcheck_healthrollback aginies-sunkar-1 · on-prem ### Post-mortem Agent Writes the timeline and root-cause draft build_timelinedraft_postmortempublish_confluence_pagepost_slack ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Incident Supervisor | P1 alert: payments-api 5xx rate 6.8% (threshold 2%) since 14:02; INC-3312 opened | — | 160 | — | | 02 | tool | Developer Agent | Last 30 min of logs and traces pulled: NullPointer starts with the 14:01 deploy v2.14.1 | Observability (Grafana / Datadog) | 620 | — | | 03 | reason | Developer Agent | Reproduced: null currency field → PaymentMapper line 118, 3-line patch | — | 2,400 | 5,200 | | 04 | plan | Incident Supervisor | Plan: patch branch → regression → on-call approval → deploy → watch → post-mortem | — | 260 | 200 | | 05 | write | Developer Agent | Branch hotfix/INC-3312 created, PR #2304: patch + regression test (+21 / −2) | GitHub | 1,400 | 2,600 | | 06 | verify | Tester Agent | Regression: 1,284 tests passed, new test catches the bug, no side effects | — | 3,000 | 800 | | 07 | approval | Incident Supervisor | On-call engineer approved PR #2304 and the emergency deploy from Slack (4 min) | — | 3,600 | — | | 08 | write | Release Agent | v2.14.2 deployed to production; 5xx rate down to 0.1% within 6 min | Kubernetes / ArgoCD | 2,800 | — | | 09 | write | Post-mortem Agent | Post-mortem draft written to Confluence: detect 3 min, resolve 31 min, 5 actions | Confluence | 1,600 | 2,200 | | 10 | notify | Post-mortem Agent | Resolution summary posted to #incidents, INC-3312 set to "resolved" | Slack | 300 | — | Simulation metrics (estimated) 320runs / month 90.5%success 15.8 sp50 $0.0159est. cost / run Approval rule Every emergency production deploy → on-call engineer approval; the hotfix PR still gets a human review Step profile ingest · 1plan · 1reason · 1tool · 1verify · 1approval · 1write · 3notify · 1 SYSTEMS - Observability (Grafana / Datadog)api - GitHubapi - Kubernetes / ArgoCDapi - Confluencestorage - Slackchannel What you need to run it - Credentials and a test environment for the enterprise APIs involved - Read access to the document store and the document types - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors agent-assisted incident response and blameless post-mortems at large SRE-driven engineering teams Packages that include this scenario SDLC agent team → Sector: Software Delivery → Related agents Software DeliveryHITL ### Legacy module modernisation: analysis, tests, rewrite, PR Starts from a Jira epic; the old module is read, its behaviour pinned with tests, rewritten and opened as a PR with evidence. 1,900 runs / month 90.4% success 5 agents Software DeliveryHITL ### Feature delivery pipeline A Jira epic turns into criteria, design, code and tests via agents; humans only review and approve the PR. 1,200 runs / month 91.2% success 5 agents Software DeliveryHITL ### CI/CD release pipeline Build, tests, scans and staging run unattended; production ships only on release-manager approval, via ArgoCD. 950 runs / month 94.7% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # iOS feature implementation (hired developer agent) URL: https://www.aginies.com/en/agents/ios-feature-implementation Language: en Summary: The Jira story is taken, built in SwiftUI, tested and opened as a PR; review and release decisions stay with the team. An iOS story reaching “ready” in the sprint → Tested PR opened, CI green, review comments applied; output and hour-equivalent written to the work log Software Deliveryapiapp # iOS feature implementation (hired developer agent) The Jira story is taken, built in SwiftUI, tested and opened as a PR; review and release decisions stay with the team. In short iOS feature implementation (hired developer agent) is an agent scenario in the Software Delivery sector running on the api, app channels. Trigger: An iOS story reaching “ready” in the sprint. The iOS Developer Supervisor coordinates 3 specialist agents (Analysis Agent, Development Agent, Review Agent) across 5 systems (Jira, GitHub, Xcode Cloud, Figma, Workforce Work Log); of 9 steps, 2 write to systems and 2 are checks or approvals. Outcome: Tested PR opened, CI green, review comments applied; output and hour-equivalent written to the work log. Approval rule: Every PR is reviewed by a person; production deployment under the team lead’s approval. Software Delivery ### iOS feature implementation (hired developer agent) [iOS feature implementation (hired developer agent)] run #41,287,991elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Every PR is reviewed by a person; production deployment under the team lead’s approval Agents in this run iOS Developer Supervisorsupervisorplans, routes, approves Analysis Agentaginies-sunkar-1 · Code AginiesReads the story, the design and the existing code, drafts the plan Development Agentaginies-sunkar-1 · Code AginiesWrites the SwiftUI code and tests Review Agentaginies-talay-2 · on-premSelf-reviews against the rules, applies review comments Audit trail00:00.00 - ◦PAY-118 “Payment screen accessibility” ready: 5 points, 6 acceptance criteria - ⚙Figma frame and PaymentView.swift read; 3 components, 2 reusable modifiers - ∿Plan: VoiceOver labels, Dynamic Type, contrast; 4 unit + 2 UI tests - ⇢5 files changed, 214 lines; 6 tests written and passing locally - ⚙PR #4821 opened; Xcode Cloud: build and tests green (4 min 12 s) - ✓Lint 0 warnings; accessibility audit 6/6; code-ownership rule: mobile team reviews - ⏸Human review: 2 comments (naming, one more test) → applied, approved - ⇢Work log: PAY-118 · PR #4821 · 5 points · hour-equivalent 6.5 · quality 4.7 · cost written - ➤Jira story moved to “reviewed”; release-note draft attached - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger An iOS story reaching “ready” in the sprint Outcome Tested PR opened, CI green, review comments applied; output and hour-equivalent written to the work log ## Agents SUPERVISOR ### iOS Developer Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · Code Aginies ### Analysis Agent Reads the story, the design and the existing code, drafts the plan get_storyget_figma_framesearch_codebasedraft_plan aginies-sunkar-1 · Code Aginies ### Development Agent Writes the SwiftUI code and tests edit_filesrun_unit_testsrun_ui_testsopen_pr aginies-talay-2 · on-prem ### Review Agent Self-reviews against the rules, applies review comments lintcheck_accessibilityapply_review_commentslog_work ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | iOS Developer Supervisor | PAY-118 “Payment screen accessibility” ready: 5 points, 6 acceptance criteria | Jira | 200 | — | | 02 | tool | Analysis Agent | Figma frame and PaymentView.swift read; 3 components, 2 reusable modifiers | Figma | 520 | — | | 03 | reason | Analysis Agent | Plan: VoiceOver labels, Dynamic Type, contrast; 4 unit + 2 UI tests | — | 900 | 1,400 | | 04 | write | Development Agent | 5 files changed, 214 lines; 6 tests written and passing locally | GitHub | 1,600 | 3,200 | | 05 | tool | Development Agent | PR #4821 opened; Xcode Cloud: build and tests green (4 min 12 s) | Xcode Cloud | 700 | — | | 06 | verify | Review Agent | Lint 0 warnings; accessibility audit 6/6; code-ownership rule: mobile team reviews | — | 600 | 520 | | 07 | approval | iOS Developer Supervisor | Human review: 2 comments (naming, one more test) → applied, approved | — | 3,000 | — | | 08 | write | Review Agent | Work log: PAY-118 · PR #4821 · 5 points · hour-equivalent 6.5 · quality 4.7 · cost written | Workforce Work Log | 300 | — | | 09 | notify | Review Agent | Jira story moved to “reviewed”; release-note draft attached | Jira | 240 | — | Simulation metrics (estimated) 380runs / month 94.4%success 8.1 sp50 $0.0207est. cost / run Approval rule Every PR is reviewed by a person; production deployment under the team lead’s approval Step profile ingest · 1reason · 1tool · 2verify · 1approval · 1write · 2notify · 1 SYSTEMS - Jiraticketing - GitHubapi - Xcode Cloudapi - Figmaapi - Workforce Work Logdb What you need to run it - API access and assignment rules for the ticketing or work-order system - Credentials and a test environment for the enterprise APIs involved - Read access to the database or vector store; a data owner - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern The hired developer agent model: a PR-producing, work-logging agent instead of an outsourced developer Packages that include this scenario Agent outsourcing (agent workforce) → Sector: Software Delivery → Related agents Software DeliveryHITL ### Legacy module modernisation: analysis, tests, rewrite, PR Starts from a Jira epic; the old module is read, its behaviour pinned with tests, rewritten and opened as a PR with evidence. 1,900 runs / month 90.4% success 5 agents Software DeliveryHITL ### Feature delivery pipeline A Jira epic turns into criteria, design, code and tests via agents; humans only review and approve the PR. 1,200 runs / month 91.2% success 5 agents Software DeliveryHITL ### CI/CD release pipeline Build, tests, scans and staging run unattended; production ships only on release-manager approval, via ArgoCD. 950 runs / month 94.7% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # IT helpdesk L1 URL: https://www.aginies.com/en/agents/it-helpdesk-l1 Language: en Summary: Password resets and access requests solved in chat; access to finance groups asks the owner via a Teams card. IT request in Teams chat or on the portal → Password reset, access granted with approval, ticket opened and resolved, user confirmed Enterprise Functionsappvoiceweb # IT helpdesk L1 Password resets and access requests solved in chat; access to finance groups asks the owner via a Teams card. In short IT helpdesk L1 is an agent scenario in the Enterprise Functions sector running on the app, voice, web channels. Trigger: IT request in Teams chat or on the portal. The Helpdesk Supervisor coordinates 3 specialist agents (Intent Agent, Access Agent, Ticket Agent) across 3 systems (Active Directory / Entra ID, ServiceNow, Microsoft Teams); of 9 steps, 3 write to systems and 2 are checks or approvals. Outcome: Password reset, access granted with approval, ticket opened and resolved, user confirmed. Approval rule: Access to a finance or production group → group owner approval. Enterprise Functions ### IT helpdesk L1 [IT helpdesk L1] run #41,289,778elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Access to a finance or production group → group owner approval Agents in this run Helpdesk Supervisorsupervisorplans, routes, approves Intent Agentaginies-sunkar-1 · on-premUnderstands the request and verifies identity Access Agentaginies-talay-2 · on-premResets passwords and manages group membership Ticket Agentaginies-sunkar-1 · on-premOpens, resolves and notifies on tickets Audit trail00:00.00 - ◦Teams chat: "I can't log in and I need access to the SAP FI-CO report folder" - ∿2 intents: password reset + access request (folder "FI-CO Reports") - ✓Identity: MFA push confirmed, device managed, no active security flag - ⇢Password reset, temporary password sent by SMS, change forced at next login - ⚙Group "FI-CO-Reports-Read" owner: Finance Controller; no SoD conflict - ⏸Access to a finance group → group owner approved via Teams card in 9 min - ⇢User added to "FI-CO-Reports-Read", 180-day access review scheduled - ⇢Ticket SN-REQ-503311 created and resolved with both actions logged - ➤User confirmed in Teams, CSAT survey sent - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger IT request in Teams chat or on the portal Outcome Password reset, access granted with approval, ticket opened and resolved, user confirmed ## Agents SUPERVISOR ### Helpdesk Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Intent Agent Understands the request and verifies identity detect_intentverify_identity aginies-talay-2 · on-prem ### Access Agent Resets passwords and manages group membership reset_passwordget_group_ownercheck_sod_rulesadd_to_group aginies-sunkar-1 · on-prem ### Ticket Agent Opens, resolves and notifies on tickets create_ticketresolve_ticketsend_teams_message ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Helpdesk Supervisor | Teams chat: "I can't log in and I need access to the SAP FI-CO report folder" | — | 180 | — | | 02 | reason | Intent Agent | 2 intents: password reset + access request (folder "FI-CO Reports") | — | 520 | 380 | | 03 | verify | Intent Agent | Identity: MFA push confirmed, device managed, no active security flag | — | 420 | — | | 04 | write | Access Agent | Password reset, temporary password sent by SMS, change forced at next login | Active Directory / Entra ID | 480 | — | | 05 | tool | Access Agent | Group "FI-CO-Reports-Read" owner: Finance Controller; no SoD conflict | Active Directory / Entra ID | 380 | — | | 06 | approval | Helpdesk Supervisor | Access to a finance group → group owner approved via Teams card in 9 min | — | 2,600 | — | | 07 | write | Access Agent | User added to "FI-CO-Reports-Read", 180-day access review scheduled | Active Directory / Entra ID | 460 | — | | 08 | write | Ticket Agent | Ticket SN-REQ-503311 created and resolved with both actions logged | ServiceNow | 420 | — | | 09 | notify | Ticket Agent | User confirmed in Teams, CSAT survey sent | Microsoft Teams | 220 | — | Simulation metrics (estimated) 47,000runs / month 98.5%success 5.7 sp50 $0.0009est. cost / run Approval rule Access to a finance or production group → group owner approval Step profile ingest · 1reason · 1tool · 1verify · 1approval · 1write · 3notify · 1 SYSTEMS - Active Directory / Entra IDcore - ServiceNowticketing - Microsoft Teamschannel What you need to run it - Read/write API or MCP server and a service account for the core system - API access and assignment rules for the ticketing or work-order system - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors L1 helpdesk deflection programmes with ITSM virtual agents at global enterprises Packages that include this scenario Email classification & attachment verification →Operational automation & demand management →HR & employee experience → Related agents Enterprise FunctionsHITL ### Procure-to-pay desk: request, quotes, contract and invoice in one place The request arrives by email; quotes compared, contract clauses checked, PO opened, invoice three-way matched and payment scheduled. 14,000 runs / month 92.7% success 5 agents Enterprise Functions ### Email classification and attachment check Every e-mail in the shared mailbox is classified, attachments checked against a list, gaps asked for at once. 86,000 runs / month 97.6% success 4 agents Enterprise FunctionsHITL ### Call centre quality monitoring Every call scored against the rubric, not a sample; compliance breaches and coaching flags ready by morning. 90,000 runs / month 98.8% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Digital account opening (KYC) URL: https://www.aginies.com/en/agents/kyc-onboarding Language: en Summary: ID, liveness and sanctions checks in one flow; clean applications open an account in minutes, no branch visit. "Open an account" application with document upload in the app → Customer and account created, documents archived, risk tier and decision trail stored Bankingappweb # Digital account opening (KYC) ID, liveness and sanctions checks in one flow; clean applications open an account in minutes, no branch visit. In short Digital account opening (KYC) is an agent scenario in the Banking sector running on the app, web channels. Trigger: "Open an account" application with document upload in the app. The Onboarding Supervisor coordinates 4 specialist agents (Document Agent, Liveness Agent, Screening Agent, Account Agent) across 5 systems (KYC / Customer 360, Sanctions Screening (Dow Jones), Core Banking, Document Store (S3), SMS Gateway); of 10 steps, 2 write to systems and 2 are checks or approvals. Outcome: Customer and account created, documents archived, risk tier and decision trail stored. Approval rule: Sanctions/PEP fuzzy match score > 0.65 → compliance analyst approval. Banking ### Digital account opening (KYC) [Digital account opening (KYC)] run #41,287,707elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Sanctions/PEP fuzzy match score > 0.65 → compliance analyst approval Agents in this run Onboarding Supervisorsupervisorplans, routes, approves Document Agentaginies-konrul-1 · on-premReads the ID document, checks field consistency and tampering Liveness Agentaginies-sunkar-1 · on-premAssesses liveness and face match on the selfie video Screening Agentaginies-tulpar-4 · on-premScreens sanctions, PEP and adverse media, assigns a risk tier Account Agentaginies-talay-2 · on-premCreates the customer and opens the account in core banking Audit trail00:00.00 - ◦Application received: ID front/back and selfie video uploaded - ∿ID number and MRZ read, field consistency 99.2%, no tampering detected - ⌘Plan: liveness → sanctions → risk tier → account opening - ✓Liveness score 0.97, face match 96% → passed - ⚙Sanctions, PEP and adverse media screened: 1 fuzzy name match (0.71) - ⚙No duplicate in Customer 360; risk tier low, expected monthly volume 40,000 - ⏸Match > 0.65 → compliance analyst reviewed, closed as false positive - ⇢Customer created and current account opened: TR33 0006 …4471 - ⇢Documents and decision trail archived with a 10-year retention tag - ➤Welcome SMS with IBAN sent - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger "Open an account" application with document upload in the app Outcome Customer and account created, documents archived, risk tier and decision trail stored ## Agents SUPERVISOR ### Onboarding Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Document Agent Reads the ID document, checks field consistency and tampering extract_id_fieldscheck_mrzdetect_tampering aginies-sunkar-1 · on-prem ### Liveness Agent Assesses liveness and face match on the selfie video run_liveness_checkmatch_face aginies-tulpar-4 · on-prem ### Screening Agent Screens sanctions, PEP and adverse media, assigns a risk tier screen_sanctionsscreen_pepadverse_media_searchcheck_duplicate_customerassign_risk_tier aginies-talay-2 · on-prem ### Account Agent Creates the customer and opens the account in core banking create_customeropen_accountarchive_documentssend_sms ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Onboarding Supervisor | Application received: ID front/back and selfie video uploaded | — | 220 | — | | 02 | reason | Document Agent | ID number and MRZ read, field consistency 99.2%, no tampering detected | — | 920 | 1,300 | | 03 | plan | Onboarding Supervisor | Plan: liveness → sanctions → risk tier → account opening | — | 300 | 210 | | 04 | verify | Liveness Agent | Liveness score 0.97, face match 96% → passed | — | 860 | 640 | | 05 | tool | Screening Agent | Sanctions, PEP and adverse media screened: 1 fuzzy name match (0.71) | Sanctions Screening (Dow Jones) | 540 | — | | 06 | tool | Screening Agent | No duplicate in Customer 360; risk tier low, expected monthly volume 40,000 | KYC / Customer 360 | 610 | — | | 07 | approval | Onboarding Supervisor | Match > 0.65 → compliance analyst reviewed, closed as false positive | — | 3,200 | — | | 08 | write | Account Agent | Customer created and current account opened: TR33 0006 …4471 | Core Banking | 560 | — | | 09 | write | Account Agent | Documents and decision trail archived with a 10-year retention tag | Document Store (S3) | 340 | — | | 10 | notify | Account Agent | Welcome SMS with IBAN sent | SMS Gateway | 240 | — | Simulation metrics (estimated) 42,000runs / month 93.8%success 7.8 sp50 $0.0062est. cost / run Approval rule Sanctions/PEP fuzzy match score > 0.65 → compliance analyst approval Step profile ingest · 1plan · 1reason · 1tool · 2verify · 1approval · 1write · 2notify · 1 SYSTEMS - KYC / Customer 360db - Sanctions Screening (Dow Jones)api - Core Bankingcore - Document Store (S3)storage - SMS Gatewaychannel What you need to run it - Read access to the database or vector store; a data owner - Credentials and a test environment for the enterprise APIs involved - Read/write API or MCP server and a service account for the core system - Read access to the document store and the document types - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the minutes-long remote onboarding of digital banks, adapted to local remote-KYC regulation Packages that include this scenario Credit and onboarding operations → Sector: Banking → Related agents BankingHITL ### Dialogue banking: voice and text transaction assistant The customer talks or types; bill, limit, dispute and transfer are completed in core banking in one session. 186,000 runs / month 94.7% success 5 agents BankingHITL ### Credit card limit increase The customer types in the app or talks to the call centre; the decision takes seconds, and anything above threshold goes to a manager. 18,400 runs / month 96.4% success 4 agents BankingHITL ### Card fraud alert triage The card is blocked in seconds and the customer asked on WhatsApp; the reply releases or cancels the card. 65,000 runs / month 97.1% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Legacy module modernisation: analysis, tests, rewrite, PR URL: https://www.aginies.com/en/agents/legacy-module-modernization Language: en Summary: Starts from a Jira epic; the old module is read, its behaviour pinned with tests, rewritten and opened as a PR with evidence. A "modernisation" epic pulled into the sprint in Jira → Characterisation tests green, new module passes the same tests, PR under review Software Deliveryapibatch # Legacy module modernisation: analysis, tests, rewrite, PR Starts from a Jira epic; the old module is read, its behaviour pinned with tests, rewritten and opened as a PR with evidence. In short Legacy module modernisation: analysis, tests, rewrite, PR is an agent scenario in the Software Delivery sector running on the api, batch channels. Trigger: A "modernisation" epic pulled into the sprint in Jira. The Modernisation Supervisor coordinates 4 specialist agents (Analysis Agent, Test Agent, Developer Agent, Review Agent) across 4 systems (Jira, Git (self-hosted), CI Pipeline, Static Analysis); of 10 steps, 3 write to systems and 2 are checks or approvals. Outcome: Characterisation tests green, new module passes the same tests, PR under review. Approval rule: Production schema change or out-of-scope file → architect approval; PR merge always human. Software Delivery ### Legacy module modernisation: analysis, tests, rewrite, PR [Legacy module modernisation: analysis, tests, rewrite, PR] run #41,290,187elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Production schema change or out-of-scope file → architect approval; PR merge always human Agents in this run Modernisation Supervisorsupervisorplans, routes, approves Analysis Agentaginies-tulpar-4 · on-premMaps the old module's dependencies and behaviour, writes the risks Test Agentaginies-sunkar-1 · Code AginiesPins current behaviour with characterisation tests Developer Agentaginies-sunkar-1 · Code AginiesRewrites the module in the target architecture and makes the tests pass Review Agentaginies-tulpar-4 · on-premChecks security, performance and out-of-scope changes Audit trail00:00.00 - ◦Epic SRV-2210: "move the collections calculation module from Java 8 to a service", 14 files, 6,800 lines - ⌕Call graph extracted: 3 external dependencies, 2 shared tables, shared use with the nightly batch - ∿Design note: behaviour boundaries, 4 risky calculation branches, out-of-scope files listed - ⌘Plan: characterisation tests → green → rewrite → same tests → static analysis → PR - ⇢212 characterisation tests generated; 91% coverage on the old module, 3 ambiguous branches flagged - ⏸Expected behaviour of the 3 ambiguous branches asked of the product owner; answer in a Jira comment - ⇢Module rewritten as a service; 212/212 tests pass, p95 latency 180 ms → 62 ms - ✓Static analysis clean; 1 out-of-scope file change reverted; no secret leakage - ⇢PR #4471 opened: design note, test report, performance comparison and rollback plan attached - ➤Jira epic set to "in review"; architect and team lead assigned as PR reviewers - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger A "modernisation" epic pulled into the sprint in Jira Outcome Characterisation tests green, new module passes the same tests, PR under review ## Agents SUPERVISOR ### Modernisation Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-tulpar-4 · on-prem ### Analysis Agent Maps the old module's dependencies and behaviour, writes the risks read_repocall_graphwrite_design_note aginies-sunkar-1 · Code Aginies ### Test Agent Pins current behaviour with characterisation tests generate_testsrun_testscoverage_report aginies-sunkar-1 · Code Aginies ### Developer Agent Rewrites the module in the target architecture and makes the tests pass edit_filesrun_testsopen_pr aginies-tulpar-4 · on-prem ### Review Agent Checks security, performance and out-of-scope changes static_analysisdiff_reviewrequest_changes ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Modernisation Supervisor | Epic SRV-2210: "move the collections calculation module from Java 8 to a service", 14 files, 6,800 lines | Jira | 200 | — | | 02 | retrieve | Analysis Agent | Call graph extracted: 3 external dependencies, 2 shared tables, shared use with the nightly batch | Git (self-hosted) | 2,800 | 2,400 | | 03 | reason | Analysis Agent | Design note: behaviour boundaries, 4 risky calculation branches, out-of-scope files listed | — | 2,400 | 1,800 | | 04 | plan | Modernisation Supervisor | Plan: characterisation tests → green → rewrite → same tests → static analysis → PR | — | 260 | 240 | | 05 | write | Test Agent | 212 characterisation tests generated; 91% coverage on the old module, 3 ambiguous branches flagged | CI Pipeline | 3,600 | 2,900 | | 06 | approval | Modernisation Supervisor | Expected behaviour of the 3 ambiguous branches asked of the product owner; answer in a Jira comment | Jira | 400 | — | | 07 | write | Developer Agent | Module rewritten as a service; 212/212 tests pass, p95 latency 180 ms → 62 ms | Git (self-hosted) | 3,900 | 3,600 | | 08 | verify | Review Agent | Static analysis clean; 1 out-of-scope file change reverted; no secret leakage | Static Analysis | 1,500 | 900 | | 09 | write | Developer Agent | PR #4471 opened: design note, test report, performance comparison and rollback plan attached | Git (self-hosted) | 600 | — | | 10 | notify | Modernisation Supervisor | Jira epic set to "in review"; architect and team lead assigned as PR reviewers | Jira | 300 | — | Simulation metrics (estimated) 1,900runs / month 90.4%success 16 sp50 $0.0189est. cost / run Approval rule Production schema change or out-of-scope file → architect approval; PR merge always human Step profile ingest · 1plan · 1retrieve · 1reason · 1verify · 1approval · 1write · 3notify · 1 SYSTEMS - Jiraticketing - Git (self-hosted)api - CI Pipelineapi - Static Analysisapi What you need to run it - API access and assignment rules for the ticketing or work-order system - Credentials and a test environment for the enterprise APIs involved - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern The agent-assisted rewrite pattern in large banks' and telcos' core modernisation programmes Packages that include this scenario SDLC agent team → Sector: Software Delivery → Related agents Software DeliveryHITL ### Feature delivery pipeline A Jira epic turns into criteria, design, code and tests via agents; humans only review and approve the PR. 1,200 runs / month 91.2% success 5 agents Software DeliveryHITL ### CI/CD release pipeline Build, tests, scans and staging run unattended; production ships only on release-manager approval, via ArgoCD. 950 runs / month 94.7% success 5 agents Software DeliveryHITL ### Incident response and hotfix Alert fires, bug reproduced, patch and regression run ready; on-call approves, post-mortem already drafted. 320 runs / month 90.5% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Line stoppage response: from alarm to work order and shift report URL: https://www.aginies.com/en/agents/line-stoppage-response Language: en Summary: An MES alarm lands; root cause found from history, spare part checked, work order and plan update ready within minutes. Unplanned stoppage event from the MES (line 3, station 7) → Work order opened, part reserved, plan shifted, shift report written Manufacturing & Automotiveapiapp # Line stoppage response: from alarm to work order and shift report An MES alarm lands; root cause found from history, spare part checked, work order and plan update ready within minutes. In short Line stoppage response: from alarm to work order and shift report is an agent scenario in the Manufacturing & Automotive sector running on the api, app channels. Trigger: Unplanned stoppage event from the MES (line 3, station 7). The Production Supervisor coordinates 4 specialist agents (Diagnosis Agent, Parts Agent, Planning Agent, Report Agent) across 5 systems (MES, SCADA Historian, CMMS, ERP Spare Parts, Shift Channel (Teams)); of 10 steps, 2 write to systems and 1 are checks or approvals. Outcome: Work order opened, part reserved, plan shifted, shift report written. Approval rule: Schedule change > 2 hours or urgent purchase above 50,000 → shift lead approval. Manufacturing & Automotive ### Line stoppage response: from alarm to work order and shift report [Line stoppage response: from alarm to work order and shift report] run #41,289,190elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Schedule change > 2 hours or urgent purchase above 50,000 → shift lead approval Agents in this run Production Supervisorsupervisorplans, routes, approves Diagnosis Agentaginies-tulpar-4 · on-premBuilds a root-cause hypothesis from sensor history and maintenance records Parts Agentaginies-sunkar-1 · on-premChecks spare-part stock and lead time, reserves the part Planning Agentaginies-talay-2 · on-premRedistributes the production plan and shift load Report Agentaginies-sunkar-1 · on-premOpens the work order, informs the team, writes the shift report Audit trail00:00.00 - ◦MES event: line 3 / station 7 stopped, fault E-214 (servo torque overload), 14:02 - ⌕Historian: torque up 18% over 6 hours, vibration above threshold for 3 days - ∿Root-cause ranking: bearing wear 72%, belt tension 19%, servo drive 9% - ⌘Plan: part stock → work order → schedule simulation → team notice → shift report - ⚙Bearing 6208-2RS: 4 in stock, shelf B-12; no spare servo drive, lead time 9 days - ⇢2 bearings reserved to the work order; purchase request opened for the servo drive - ⇢CMMS work order #WO-18834: high priority, 2 technicians, estimated 2.5 h, procedure attached - ∿Simulation: 2.5 h stoppage costs 310 units; 180 units can shift to line 2 - ⏸Schedule change presented to the shift lead; approved, line 2 recipe updated - ➤Summary to the team channel; shift report written with root cause, actions and lost units - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Unplanned stoppage event from the MES (line 3, station 7) Outcome Work order opened, part reserved, plan shifted, shift report written ## Agents SUPERVISOR ### Production Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-tulpar-4 · on-prem ### Diagnosis Agent Builds a root-cause hypothesis from sensor history and maintenance records query_historianget_maintenance_logrank_causes aginies-sunkar-1 · on-prem ### Parts Agent Checks spare-part stock and lead time, reserves the part check_stockreserve_partcreate_purchase_req aginies-talay-2 · on-prem ### Planning Agent Redistributes the production plan and shift load get_schedulesimulate_shiftupdate_schedule aginies-sunkar-1 · on-prem ### Report Agent Opens the work order, informs the team, writes the shift report open_work_ordernotify_teamwrite_shift_report ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Production Supervisor | MES event: line 3 / station 7 stopped, fault E-214 (servo torque overload), 14:02 | MES | 160 | — | | 02 | retrieve | Diagnosis Agent | Historian: torque up 18% over 6 hours, vibration above threshold for 3 days | SCADA Historian | 1,400 | 520 | | 03 | reason | Diagnosis Agent | Root-cause ranking: bearing wear 72%, belt tension 19%, servo drive 9% | — | 1,800 | 1,100 | | 04 | plan | Production Supervisor | Plan: part stock → work order → schedule simulation → team notice → shift report | — | 240 | 220 | | 05 | tool | Parts Agent | Bearing 6208-2RS: 4 in stock, shelf B-12; no spare servo drive, lead time 9 days | ERP Spare Parts | 480 | — | | 06 | write | Parts Agent | 2 bearings reserved to the work order; purchase request opened for the servo drive | ERP Spare Parts | 520 | — | | 07 | write | Report Agent | CMMS work order #WO-18834: high priority, 2 technicians, estimated 2.5 h, procedure attached | CMMS | 600 | — | | 08 | reason | Planning Agent | Simulation: 2.5 h stoppage costs 310 units; 180 units can shift to line 2 | MES | 2,200 | 900 | | 09 | approval | Planning Agent | Schedule change presented to the shift lead; approved, line 2 recipe updated | MES | 900 | — | | 10 | notify | Report Agent | Summary to the team channel; shift report written with root cause, actions and lost units | Shift Channel (Teams) | 460 | 260 | Simulation metrics (estimated) 3,100runs / month 92.2%success 8.8 sp50 $0.0048est. cost / run Approval rule Schedule change > 2 hours or urgent purchase above 50,000 → shift lead approval Step profile ingest · 1plan · 1retrieve · 1reason · 2tool · 1approval · 1write · 2notify · 1 SYSTEMS - MESerp - SCADA Historiandb - CMMSticketing - ERP Spare Partserp - Shift Channel (Teams)channel What you need to run it - Posting and query API access to ERP/MES - Read access to the database or vector store; a data owner - API access and assignment rules for the ticketing or work-order system - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Automotive suppliers' predictive maintenance and rapid response pattern, combined with schedule simulation Packages that include this scenario Operational automation & demand management →Production and maintenance operations → Sector: Manufacturing & Automotive → Related agents Manufacturing & AutomotiveHITL ### Supplier invoice three-way match Invoice, PO and goods receipt matched in SAP automatically; only out-of-tolerance lines reach a buyer. 21,000 runs / month 97.1% success 4 agents Manufacturing & AutomotiveHITL ### Maintenance ticket triage Sensor alarm and technician voice note become one work order, spare parts and a safe stop window included. 3,400 runs / month 95.2% success 5 agents Manufacturing & AutomotiveHITL ### Warranty claim processing Dealer claims validated for coverage, parts and labour in minutes; big amounts go to the regional manager. 8,600 runs / month 96.3% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Maintenance ticket triage URL: https://www.aginies.com/en/agents/maintenance-ticket-triage Language: en Summary: Sensor alarm and technician voice note become one work order, spare parts and a safe stop window included. Vibration threshold alarm from SCADA → Prioritised work order opened in the CMMS, part transfer requested, technician assigned Manufacturing & Automotiveapiappvoice # Maintenance ticket triage Sensor alarm and technician voice note become one work order, spare parts and a safe stop window included. In short Maintenance ticket triage is an agent scenario in the Manufacturing & Automotive sector running on the api, app, voice channels. Trigger: Vibration threshold alarm from SCADA. The Maintenance Supervisor coordinates 4 specialist agents (Signal Agent, Field Note Agent, Diagnosis Agent, Planning Agent) across 4 systems (SCADA / Historian, CMMS (Maximo), SAP S/4HANA MM, Manuals & Work Order Vector Store); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: Prioritised work order opened in the CMMS, part transfer requested, technician assigned. Approval rule: Planned stop > 4 hours → shift supervisor approval. Manufacturing & Automotive ### Maintenance ticket triage [Maintenance ticket triage] run #41,287,213elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Planned stop > 4 hours → shift supervisor approval Agents in this run Maintenance Supervisorsupervisorplans, routes, approves Signal Agentaginies-sunkar-1 · on-premReads trend and anomaly from historian data Field Note Agentaginies-voice-tr-v3 · on-premTranscribes the technician's voice note Diagnosis Agentaginies-tulpar-4 · on-premClassifies the fault using manuals and past work orders Planning Agentaginies-talay-2 · on-premPlans the work order with parts, safety and shift window Audit trail00:00.00 - ◦Alarm: bearing vibration 7.8 mm/s on press line P-03 (threshold 6.3) - ⚙Last 48 h of vibration and temperature trend pulled from the historian - ∿Trend: +38% in 6 days, temperature +9 °C → developing bearing fault - ∿Technician note transcribed: "grinding noise on the drive side since Monday" - ⌕Manual §7.4 and 3 similar work orders (2024–2026) retrieved - ∿Fault class: drive-side bearing 6312-2RS · priority P2 · 5–7 days to failure - ⚙Spare part 6312-2RS: 2 in central store, 0 at plant → transfer needed - ✓Safety: LOTO procedure and 2-person rule attached, Sat 06:00 window available - ⏸Planned stop > 4 h → shift supervisor approved the Saturday window - ⇢Work order WO-88213 created, part transfer requested, technician assigned - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Vibration threshold alarm from SCADA Outcome Prioritised work order opened in the CMMS, part transfer requested, technician assigned ## Agents SUPERVISOR ### Maintenance Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Signal Agent Reads trend and anomaly from historian data get_sensor_windowdetect_anomaly aginies-voice-tr-v3 · on-prem ### Field Note Agent Transcribes the technician's voice note transcribe_audioextract_observations aginies-tulpar-4 · on-prem ### Diagnosis Agent Classifies the fault using manuals and past work orders search_manualssearch_work_order_historyclassify_failure aginies-talay-2 · on-prem ### Planning Agent Plans the work order with parts, safety and shift window check_spare_partscreate_work_orderassign_technician ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Maintenance Supervisor | Alarm: bearing vibration 7.8 mm/s on press line P-03 (threshold 6.3) | — | 180 | — | | 02 | tool | Signal Agent | Last 48 h of vibration and temperature trend pulled from the historian | SCADA / Historian | 420 | — | | 03 | reason | Signal Agent | Trend: +38% in 6 days, temperature +9 °C → developing bearing fault | — | 640 | 520 | | 04 | reason | Field Note Agent | Technician note transcribed: "grinding noise on the drive side since Monday" | — | 820 | 340 | | 05 | retrieve | Diagnosis Agent | Manual §7.4 and 3 similar work orders (2024–2026) retrieved | Manuals & Work Order Vector Store | 480 | 760 | | 06 | reason | Diagnosis Agent | Fault class: drive-side bearing 6312-2RS · priority P2 · 5–7 days to failure | — | 780 | 980 | | 07 | tool | Planning Agent | Spare part 6312-2RS: 2 in central store, 0 at plant → transfer needed | SAP S/4HANA MM | 430 | — | | 08 | verify | Planning Agent | Safety: LOTO procedure and 2-person rule attached, Sat 06:00 window available | — | 360 | 280 | | 09 | approval | Maintenance Supervisor | Planned stop > 4 h → shift supervisor approved the Saturday window | — | 2,400 | — | | 10 | write | Planning Agent | Work order WO-88213 created, part transfer requested, technician assigned | CMMS (Maximo) | 560 | — | Simulation metrics (estimated) 3,400runs / month 95.2%success 7.1 sp50 $0.0073est. cost / run Approval rule Planned stop > 4 hours → shift supervisor approval Step profile ingest · 1retrieve · 1reason · 3tool · 2verify · 1approval · 1write · 1 SYSTEMS - SCADA / Historiancore - CMMS (Maximo)ticketing - SAP S/4HANA MMerp - Manuals & Work Order Vector Storedb What you need to run it - Read/write API or MCP server and a service account for the core system - API access and assignment rules for the ticketing or work-order system - Posting and query API access to ERP/MES - Read access to the database or vector store; a data owner - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors predictive maintenance programmes at automotive OEMs that combine historian data with the CMMS Packages that include this scenario Operational automation & demand management →Field service & dispatch automation →Production and maintenance operations → Sector: Manufacturing & Automotive → Related agents Manufacturing & AutomotiveHITL ### Line stoppage response: from alarm to work order and shift report An MES alarm lands; root cause found from history, spare part checked, work order and plan update ready within minutes. 3,100 runs / month 92.2% success 5 agents Manufacturing & AutomotiveHITL ### Supplier invoice three-way match Invoice, PO and goods receipt matched in SAP automatically; only out-of-tolerance lines reach a buyer. 21,000 runs / month 97.1% success 4 agents Manufacturing & AutomotiveHITL ### Warranty claim processing Dealer claims validated for coverage, parts and labour in minutes; big amounts go to the regional manager. 8,600 runs / month 96.3% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Merchant KYB onboarding URL: https://www.aginies.com/en/agents/merchant-kyb-onboarding Language: en Summary: Registry, UBO and sanctions checks in one flow; low-risk merchants start accepting payments the same day. Application form and document upload on the Merchant Portal → Merchant account live, risk tier assigned, MSA e-signed, decision trail stored Fintech & Paymentswebemail # Merchant KYB onboarding Registry, UBO and sanctions checks in one flow; low-risk merchants start accepting payments the same day. In short Merchant KYB onboarding is an agent scenario in the Fintech & Payments sector running on the web, email channels. Trigger: Application form and document upload on the Merchant Portal. The Merchant Onboarding Supervisor coordinates 4 specialist agents (Business Identity Agent, UBO Agent, Risk Agent, Contract Agent) across 5 systems (Merchant Portal, Trade Registry (MERSİS), Sanctions Screening, e-Sign, Email); of 10 steps, 2 write to systems and 2 are checks or approvals. Outcome: Merchant account live, risk tier assigned, MSA e-signed, decision trail stored. Approval rule: Risk tier high or UBO fuzzy match > 0.65 → compliance manager approval. Fintech & Payments ### Merchant KYB onboarding [Merchant KYB onboarding] run #41,289,176elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Risk tier high or UBO fuzzy match > 0.65 → compliance manager approval Agents in this run Merchant Onboarding Supervisorsupervisorplans, routes, approves Business Identity Agentaginies-konrul-1 · on-premVerifies trade registry, tax number and signatory circular UBO Agentaginies-tulpar-4 · on-premResolves the ownership chain, identifies and screens UBOs at ≥ 25% Risk Agentaginies-talay-2 · on-premAssigns a risk tier from MCC, volume and sector data Contract Agentaginies-talay-2 · on-premSends the MSA for e-signature, activates the merchant account Audit trail00:00.00 - ◦Application #MB-3107: Demir Tekstil Ltd., MCC 5651, expected volume 850,000/month - ⚙Registry record pulled: incorporated 2016, 2 signatories; tax number and circular match - ⌘Plan: ownership chain → sanctions/PEP → risk tier → MSA → account - ∿Ownership: 2 UBOs (60% / 40%), chain depth 2, one intermediate legal entity - ⚙2 UBOs and the company screened for sanctions/PEP, 1 fuzzy match (0.58) - ✓Policy v2.1: MCC 5651 medium risk, volume above threshold → risk tier: high - ⏸Risk tier high → compliance manager reviewed and approved the file - ⇢MSA v4.2 sent to both signatories for e-signature, signed within 3 hours - ⇢Merchant account MID 88231045 created, payment acceptance live - ➤Welcome email: MID, API key and T+1 settlement schedule - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Application form and document upload on the Merchant Portal Outcome Merchant account live, risk tier assigned, MSA e-signed, decision trail stored ## Agents SUPERVISOR ### Merchant Onboarding Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Business Identity Agent Verifies trade registry, tax number and signatory circular extract_registry_fieldsverify_tax_numberparse_signatory_circular aginies-tulpar-4 · on-prem ### UBO Agent Resolves the ownership chain, identifies and screens UBOs at ≥ 25% resolve_ownership_chainidentify_uboscreen_sanctionsscreen_pep aginies-talay-2 · on-prem ### Risk Agent Assigns a risk tier from MCC, volume and sector data check_mcc_policyestimate_volumeassign_risk_tier aginies-talay-2 · on-prem ### Contract Agent Sends the MSA for e-signature, activates the merchant account send_esigncreate_merchantsend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Merchant Onboarding Supervisor | Application #MB-3107: Demir Tekstil Ltd., MCC 5651, expected volume 850,000/month | — | 220 | — | | 02 | tool | Business Identity Agent | Registry record pulled: incorporated 2016, 2 signatories; tax number and circular match | Trade Registry (MERSİS) | 480 | — | | 03 | plan | Merchant Onboarding Supervisor | Plan: ownership chain → sanctions/PEP → risk tier → MSA → account | — | 300 | 210 | | 04 | reason | UBO Agent | Ownership: 2 UBOs (60% / 40%), chain depth 2, one intermediate legal entity | — | 760 | 900 | | 05 | tool | UBO Agent | 2 UBOs and the company screened for sanctions/PEP, 1 fuzzy match (0.58) | Sanctions Screening | 540 | — | | 06 | verify | Risk Agent | Policy v2.1: MCC 5651 medium risk, volume above threshold → risk tier: high | — | 640 | 720 | | 07 | approval | Merchant Onboarding Supervisor | Risk tier high → compliance manager reviewed and approved the file | — | 3,400 | — | | 08 | write | Contract Agent | MSA v4.2 sent to both signatories for e-signature, signed within 3 hours | e-Sign | 620 | — | | 09 | write | Contract Agent | Merchant account MID 88231045 created, payment acceptance live | Merchant Portal | 520 | — | | 10 | notify | Contract Agent | Welcome email: MID, API key and T+1 settlement schedule | Email | 260 | — | Simulation metrics (estimated) 1,400runs / month 93.4%success 7.8 sp50 $0.0087est. cost / run Approval rule Risk tier high or UBO fuzzy match > 0.65 → compliance manager approval Step profile ingest · 1plan · 1reason · 1tool · 2verify · 1approval · 1write · 2notify · 1 SYSTEMS - Merchant Portalapi - Trade Registry (MERSİS)api - Sanctions Screeningapi - e-Signapi - Emailemail What you need to run it - Credentials and a test environment for the enterprise APIs involved - Mailbox access (Microsoft 365 or Google Workspace) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the self-serve, minutes-long merchant onboarding of global payment providers, adapted to local KYB rules Packages that include this scenario Merchant and payment operations → Sector: Fintech & Payments → Related agents Fintech & PaymentsHITL ### Merchant operations assistant: POS, settlement, refund and limit The merchant writes on WhatsApp; POS fault, settlement query, refund and limit increase are resolved in one dialogue. 64,000 runs / month 93.8% success 5 agents Fintech & PaymentsHITL ### AML transaction monitoring alert triage Alerts arrive enriched with a draft narrative; the analyst always decides and the filing goes out in minutes. 8,500 runs / month 95.8% success 5 agents Fintech & PaymentsHITL ### BNPL instant instalment decision Checkout instalment decision in seconds from bureau and open-banking data; large tickets go to underwriting. 54,000 runs / month 97.3% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Merchant operations assistant: POS, settlement, refund and limit URL: https://www.aginies.com/en/agents/merchant-operations-assistant Language: en Summary: The merchant writes on WhatsApp; POS fault, settlement query, refund and limit increase are resolved in one dialogue. Operations request over WhatsApp Business or the merchant portal → POS reactivated, settlement explained, refund issued, limit request with risk Fintech & Paymentswhatsappwebvoice # Merchant operations assistant: POS, settlement, refund and limit The merchant writes on WhatsApp; POS fault, settlement query, refund and limit increase are resolved in one dialogue. In short Merchant operations assistant: POS, settlement, refund and limit is an agent scenario in the Fintech & Payments sector running on the whatsapp, web, voice channels. Trigger: Operations request over WhatsApp Business or the merchant portal. The Merchant Supervisor coordinates 4 specialist agents (Dialogue Agent, POS Agent, Operations Agent, Risk Agent) across 5 systems (WhatsApp Business, Terminal Management, Settlement Ledger, Refund Service, Ticketing); of 10 steps, 2 write to systems and 3 are checks or approvals. Outcome: POS reactivated, settlement explained, refund issued, limit request with risk. Approval rule: Refund > 25,000 or limit increase > 50% → risk team approval. Fintech & Payments ### Merchant operations assistant: POS, settlement, refund and limit [Merchant operations assistant: POS, settlement, refund and limit] run #41,288,027elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Refund > 25,000 or limit increase > 50% → risk team approval Agents in this run Merchant Supervisorsupervisorplans, routes, approves Dialogue Agentaginies-talay-2 · on-premVerifies the merchant contact, separates requests, writes replies POS Agentaginies-sunkar-1 · on-premReads terminal state, runs remote diagnostics and reactivation Operations Agentaginies-tulpar-4 · on-premHandles settlement, refund and limit operations with ledger and payment services Risk Agentaginies-sunkar-1 · on-premAssesses refund and limit requests against the risk policy Audit trail00:00.00 - ◦WhatsApp: "POS has been down 2 days, yesterday's settlement is short, I need to refund a customer" - ✓Contact number matched to the merchant record, one-time code verified - ⌘Plan: terminal diagnosis → settlement breakdown → refund eligibility → risk → one reply - ⚙Terminal TID 48213: last transaction 2 days ago, SIM data plan exhausted, device healthy - ⇢Data plan renewed, terminal restarted remotely, test transaction successful - ⚙Settlement 48,610: 2 instalment sales settle next day, 1 chargeback on hold; breakdown prepared - ✓Refund 1,890, original sale 6 days ago, merchant risk low → inside the auto-refund band - ⇢Refund sent to the card, ref #RF-55120; customer SMS triggered - ⏸Limit 150,000 → 300,000: +100% → routed to risk team approval - ➤One reply: POS active, settlement breakdown, refund reference, limit request in 1 business day - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Operations request over WhatsApp Business or the merchant portal Outcome POS reactivated, settlement explained, refund issued, limit request with risk ## Agents SUPERVISOR ### Merchant Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Dialogue Agent Verifies the merchant contact, separates requests, writes replies verify_contactdetect_intentscompose_reply aginies-sunkar-1 · on-prem ### POS Agent Reads terminal state, runs remote diagnostics and reactivation get_terminalremote_diagnosereactivate_terminal aginies-tulpar-4 · on-prem ### Operations Agent Handles settlement, refund and limit operations with ledger and payment services get_settlementexplain_feesissue_refundrequest_limit aginies-sunkar-1 · on-prem ### Risk Agent Assesses refund and limit requests against the risk policy merchant_risk_scorepolicy_checkroute_approval ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Merchant Supervisor | WhatsApp: "POS has been down 2 days, yesterday's settlement is short, I need to refund a customer" | WhatsApp Business | 180 | — | | 02 | verify | Dialogue Agent | Contact number matched to the merchant record, one-time code verified | — | 520 | 140 | | 03 | plan | Merchant Supervisor | Plan: terminal diagnosis → settlement breakdown → refund eligibility → risk → one reply | — | 240 | 220 | | 04 | tool | POS Agent | Terminal TID 48213: last transaction 2 days ago, SIM data plan exhausted, device healthy | Terminal Management | 460 | — | | 05 | write | POS Agent | Data plan renewed, terminal restarted remotely, test transaction successful | Terminal Management | 1,400 | — | | 06 | tool | Operations Agent | Settlement 48,610: 2 instalment sales settle next day, 1 chargeback on hold; breakdown prepared | Settlement Ledger | 540 | 260 | | 07 | verify | Risk Agent | Refund 1,890, original sale 6 days ago, merchant risk low → inside the auto-refund band | — | 480 | 200 | | 08 | write | Operations Agent | Refund sent to the card, ref #RF-55120; customer SMS triggered | Refund Service | 620 | — | | 09 | approval | Risk Agent | Limit 150,000 → 300,000: +100% → routed to risk team approval | Ticketing | 300 | — | | 10 | notify | Dialogue Agent | One reply: POS active, settlement breakdown, refund reference, limit request in 1 business day | WhatsApp Business | 420 | 210 | Simulation metrics (estimated) 64,000runs / month 93.8%success 7.2 sp50 $0.0034est. cost / run Approval rule Refund > 25,000 or limit increase > 50% → risk team approval Step profile ingest · 1plan · 1tool · 2verify · 2approval · 1write · 2notify · 1 SYSTEMS - WhatsApp Businesschannel - Terminal Managementapi - Settlement Ledgerpayments - Refund Servicepayments - Ticketingticketing What you need to run it - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - Credentials and a test environment for the enterprise APIs involved - Authorised API access and limits for payment services - API access and assignment rules for the ticketing or work-order system - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern The assistant pattern that folds payment institutions' merchant support lines into one dialogue Packages that include this scenario Financial analysis & operational intelligence →Merchant and payment operations → Sector: Fintech & Payments → Related agents Fintech & PaymentsHITL ### Merchant KYB onboarding Registry, UBO and sanctions checks in one flow; low-risk merchants start accepting payments the same day. 1,400 runs / month 93.4% success 5 agents Fintech & PaymentsHITL ### AML transaction monitoring alert triage Alerts arrive enriched with a draft narrative; the analyst always decides and the filing goes out in minutes. 8,500 runs / month 95.8% success 5 agents Fintech & PaymentsHITL ### BNPL instant instalment decision Checkout instalment decision in seconds from bureau and open-banking data; large tickets go to underwriting. 54,000 runs / month 97.3% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Meter reading dispute URL: https://www.aginies.com/en/agents/meter-reading-dispute Language: en Summary: Customer sends a meter photo; the index is read by vision, the estimated bill fixed, big credits approved. Bill dispute with a meter photo via WhatsApp or the app → Wrong invoice cancelled, rebill issued, customer informed with an explanation Energy & Utilitieswhatsappappweb # Meter reading dispute Customer sends a meter photo; the index is read by vision, the estimated bill fixed, big credits approved. In short Meter reading dispute is an agent scenario in the Energy & Utilities sector running on the whatsapp, app, web channels. Trigger: Bill dispute with a meter photo via WhatsApp or the app. The Billing Dispute Supervisor coordinates 3 specialist agents (Vision Agent, Billing Agent, Operations Agent) across 4 systems (Billing / CIS (SAP IS-U), Meter Data Management, Regulation Vector Store (EPDK), WhatsApp Business API); of 9 steps, 1 write to systems and 2 are checks or approvals. Outcome: Wrong invoice cancelled, rebill issued, customer informed with an explanation. Approval rule: Credit / correction > 1,000 TL → billing team lead approval. Energy & Utilities ### Meter reading dispute [Meter reading dispute] run #41,289,131elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Credit / correction > 1,000 TL → billing team lead approval Agents in this run Billing Dispute Supervisorsupervisorplans, routes, approves Vision Agentaginies-konrul-1 · on-premReads serial and index from the meter photo Billing Agentaginies-tulpar-4 · on-premComputes the difference against history and regulation Operations Agentaginies-talay-2 · on-premCancels, rebills and informs the customer Audit trail00:00.00 - ◦WhatsApp: "my bill is 2,840 TL, the meter shows less" + photo; customer verified - ∿Meter serial 3812-0091, index 48,213 kWh; serial matches, no tampering signs - ⚙Last 12 readings pulled; billed reading 49,760 kWh was an estimate (code E) - ∿Difference 1,547 kWh · monthly average 310 kWh → estimate ran ~5 months ahead - ⌕Regulation: estimated bills correctable within 12 months on a verified reading - ✓Rebill: 2,840 TL → 1,318 TL; difference 1,522 TL > 1,000 TL threshold - ⏸Credit > 1,000 TL → billing team lead approved, note attached - ⇢Invoice cancelled, rebill of 1,318 TL issued, field reading scheduled - ➤Customer sent the new bill and explanation, case closed with an audit trail - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Bill dispute with a meter photo via WhatsApp or the app Outcome Wrong invoice cancelled, rebill issued, customer informed with an explanation ## Agents SUPERVISOR ### Billing Dispute Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Vision Agent Reads serial and index from the meter photo read_meter_photocheck_photo_integrity aginies-tulpar-4 · on-prem ### Billing Agent Computes the difference against history and regulation get_reading_historyestimate_consumptionsearch_billing_rules aginies-talay-2 · on-prem ### Operations Agent Cancels, rebills and informs the customer cancel_invoicecreate_rebillschedule_field_readingsend_whatsapp ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Billing Dispute Supervisor | WhatsApp: "my bill is 2,840 TL, the meter shows less" + photo; customer verified | — | 200 | — | | 02 | reason | Vision Agent | Meter serial 3812-0091, index 48,213 kWh; serial matches, no tampering signs | — | 920 | 610 | | 03 | tool | Billing Agent | Last 12 readings pulled; billed reading 49,760 kWh was an estimate (code E) | Meter Data Management | 420 | — | | 04 | reason | Billing Agent | Difference 1,547 kWh · monthly average 310 kWh → estimate ran ~5 months ahead | — | 700 | 820 | | 05 | retrieve | Billing Agent | Regulation: estimated bills correctable within 12 months on a verified reading | Regulation Vector Store (EPDK) | 380 | 360 | | 06 | verify | Billing Agent | Rebill: 2,840 TL → 1,318 TL; difference 1,522 TL > 1,000 TL threshold | — | 460 | 300 | | 07 | approval | Billing Dispute Supervisor | Credit > 1,000 TL → billing team lead approved, note attached | — | 2,600 | — | | 08 | write | Operations Agent | Invoice cancelled, rebill of 1,318 TL issued, field reading scheduled | Billing / CIS (SAP IS-U) | 560 | — | | 09 | notify | Operations Agent | Customer sent the new bill and explanation, case closed with an audit trail | WhatsApp Business API | 240 | — | Simulation metrics (estimated) 9,200runs / month 94.9%success 6.5 sp50 $0.003est. cost / run Approval rule Credit / correction > 1,000 TL → billing team lead approval Step profile ingest · 1retrieve · 1reason · 2tool · 1verify · 1approval · 1write · 1notify · 1 SYSTEMS - Billing / CIS (SAP IS-U)erp - Meter Data Managementdb - Regulation Vector Store (EPDK)db - WhatsApp Business APIchannel What you need to run it - Posting and query API access to ERP/MES - Read access to the database or vector store; a data owner - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors in-app meter photo and dispute flows of European utilities Packages that include this scenario Utility customer operations → Sector: Energy & Utilities → Related agents Energy & UtilitiesHITL ### Utility customer assistant: bill, outage, consumption and instalments in one dialogue The customer calls or writes; the high bill is explained, outage ETA given, meter data shown, an instalment plan set up. 520,000 runs / month 94.6% success 5 agents Energy & UtilitiesHITL ### Outage report handling Reports are clustered by feeder, one OSS ticket is raised and everyone affected hears before the crew leaves. 38,000 runs / month 97.7% success 5 agents Energy & UtilitiesHITL ### Rooftop solar connection application Application documents are read, transformer capacity and regulation checked, gaps reported the same day; the connection opinion belongs to the engineer. 5,400 runs / month 95.3% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Month-end close and reconciliation (hired accountant agent) URL: https://www.aginies.com/en/agents/month-end-close-bookkeeping Language: en Summary: Invoices are processed, bank and vendor reconciliations done, the close checklist run; entries are posted with the accountant’s approval. The month-end close calendar opening → Close tasks completed, differences explained, approved entries posted; work log and invoice line ready Enterprise Functionsbatchemailapi # Month-end close and reconciliation (hired accountant agent) Invoices are processed, bank and vendor reconciliations done, the close checklist run; entries are posted with the accountant’s approval. In short Month-end close and reconciliation (hired accountant agent) is an agent scenario in the Enterprise Functions sector running on the batch, email, api channels. Trigger: The month-end close calendar opening. The Accounting Supervisor coordinates 3 specialist agents (Document Agent, Reconciliation Agent, Close Agent) across 4 systems (ERP (SAP / Logo), Bank API, e-Invoice Portal, Workforce Work Log); of 9 steps, 2 write to systems and 2 are checks or approvals. Outcome: Close tasks completed, differences explained, approved entries posted; work log and invoice line ready. Approval rule: Every posted entry and every filing with the accountant’s approval. Enterprise Functions ### Month-end close and reconciliation (hired accountant agent) [Month-end close and reconciliation (hired accountant agent)] run #41,289,999elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Every posted entry and every filing with the accountant’s approval Agents in this run Accounting Supervisorsupervisorplans, routes, approves Document Agentaginies-konrul-1 · on-premReads invoices and statements, proposes entries Reconciliation Agentaginies-tulpar-4 · on-premMatches bank, vendor and customer accounts Close Agentaginies-talay-2 · on-premRuns the checklist, posts entries, reports Audit trail00:00.00 - ◦September close opened: 41 tasks, 238 pending invoices, 3 bank accounts - ⚙238 e-invoices and 3 statements pulled; fields extracted, 2 duplicate candidates flagged - ∿236 entry proposals: account, cost centre, VAT rate; 4 marked low-confidence - ⚙Bank movements matched: 1,184 / 1,191; 7 open items - ∿7 breaks explained: 5 timing, 1 bank fee, 1 possible double vendor payment - ✓Checklist 38/41; remaining 3 tasks await approval (accrual, provision, double payment) - ⏸Accountant approved 236 entries and 2 adjustments, marked the double payment for vendor refund - ⇢Entries posted; VAT summary and close report attached in the ERP - ⇢Work log: 236 entries · 3 reconciliations · 41 tasks · hour-equivalent 27 · quality 4.9 - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger The month-end close calendar opening Outcome Close tasks completed, differences explained, approved entries posted; work log and invoice line ready ## Agents SUPERVISOR ### Accounting Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Document Agent Reads invoices and statements, proposes entries extract_invoicefetch_bank_statementpropose_journal_entry aginies-tulpar-4 · on-prem ### Reconciliation Agent Matches bank, vendor and customer accounts match_transactionsexplain_breaksget_vendor_ledger aginies-talay-2 · on-prem ### Close Agent Runs the checklist, posts entries, reports run_checklistpost_entriesprepare_vat_summarylog_work ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Accounting Supervisor | September close opened: 41 tasks, 238 pending invoices, 3 bank accounts | ERP (SAP / Logo) | 200 | — | | 02 | tool | Document Agent | 238 e-invoices and 3 statements pulled; fields extracted, 2 duplicate candidates flagged | e-Invoice Portal | 1,400 | — | | 03 | reason | Document Agent | 236 entry proposals: account, cost centre, VAT rate; 4 marked low-confidence | — | 1,200 | 2,400 | | 04 | tool | Reconciliation Agent | Bank movements matched: 1,184 / 1,191; 7 open items | Bank API | 900 | — | | 05 | reason | Reconciliation Agent | 7 breaks explained: 5 timing, 1 bank fee, 1 possible double vendor payment | — | 800 | 900 | | 06 | verify | Close Agent | Checklist 38/41; remaining 3 tasks await approval (accrual, provision, double payment) | — | 520 | 380 | | 07 | approval | Accounting Supervisor | Accountant approved 236 entries and 2 adjustments, marked the double payment for vendor refund | — | 3,200 | — | | 08 | write | Close Agent | Entries posted; VAT summary and close report attached in the ERP | ERP (SAP / Logo) | 700 | — | | 09 | write | Close Agent | Work log: 236 entries · 3 reconciliations · 41 tasks · hour-equivalent 27 · quality 4.9 | Workforce Work Log | 300 | — | Simulation metrics (estimated) 60runs / month 96.2%success 9.2 sp50 $0.0213est. cost / run Approval rule Every posted entry and every filing with the accountant’s approval Step profile ingest · 1reason · 2tool · 2verify · 1approval · 1write · 2 SYSTEMS - ERP (SAP / Logo)erp - Bank APIpayments - e-Invoice Portalapi - Workforce Work Logdb What you need to run it - Posting and query API access to ERP/MES - Authorised API access and limits for payment services - Credentials and a test environment for the enterprise APIs involved - Read access to the database or vector store; a data owner - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern The hired accountant agent model: an agent that runs the close and proposes entries instead of an outsourced accountant Packages that include this scenario Agent outsourcing (agent workforce) → Related agents Enterprise FunctionsHITL ### Procure-to-pay desk: request, quotes, contract and invoice in one place The request arrives by email; quotes compared, contract clauses checked, PO opened, invoice three-way matched and payment scheduled. 14,000 runs / month 92.7% success 5 agents Enterprise Functions ### Email classification and attachment check Every e-mail in the shared mailbox is classified, attachments checked against a list, gaps asked for at once. 86,000 runs / month 97.6% success 4 agents Enterprise FunctionsHITL ### Call centre quality monitoring Every call scored against the rubric, not a sample; compliance breaches and coaching flags ready by morning. 90,000 runs / month 98.8% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Mortgage document review URL: https://www.aginies.com/en/agents/mortgage-document-review Language: en Summary: Payslips, social security records and appraisal read automatically; income, LTV and missing documents checked. Documents uploaded to the loan file (web, email or branch) → File pre-approved, missing-document list with the customer, income and LTV evidence on file Bankingwebemailbranch # Mortgage document review Payslips, social security records and appraisal read automatically; income, LTV and missing documents checked. In short Mortgage document review is an agent scenario in the Banking sector running on the web, email, branch channels. Trigger: Documents uploaded to the loan file (web, email or branch). The Loan File Supervisor coordinates 3 specialist agents (Extraction Agent, Income Agent, Valuation Agent) across 4 systems (Loan Origination System, Document Store (S3), Credit Bureau (KKB), Email (Exchange)); of 10 steps, 1 write to systems and 3 are checks or approvals. Outcome: File pre-approved, missing-document list with the customer, income and LTV evidence on file. Approval rule: Loan > 3,000,000 or DTI > 45% → underwriting officer approval. Banking ### Mortgage document review [Mortgage document review] run #41,290,257elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Loan > 3,000,000 or DTI > 45% → underwriting officer approval Agents in this run Loan File Supervisorsupervisorplans, routes, approves Extraction Agentaginies-konrul-1 · on-premClassifies documents, extracts fields and tables Income Agentaginies-talay-2 · on-premComputes net income, cross-checks against social security and bureau data Valuation Agentaginies-tulpar-4 · on-premReads the appraisal, determines LTV and missing documents Audit trail00:00.00 - ◦File #ML-7731: 9 documents uploaded (payslips, social security record, appraisal, deed) - ⌕Documents fetched from S3 and classified; fields extracted from 3 payslips and the record - ⌘Plan: income check → appraisal/LTV → missing docs → update file - ⚙Credit bureau report: score 1,612, existing instalments 9,800/month - ✓Net income 68,500, within 2% of the social security record, DTI 41% → eligible - ∿Appraisal 6,400,000, requested 4,800,000 → LTV 75%, limit 80% - ✓Missing: compulsory earthquake policy and last 3 months' statements → 2 documents - ⏸Loan > 3,000,000 → underwriting officer pre-approved the file - ⇢LOS file set to "pre-approved – awaiting documents", checklist attached - ➤Missing-document email sent to the customer and branch advisor - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Documents uploaded to the loan file (web, email or branch) Outcome File pre-approved, missing-document list with the customer, income and LTV evidence on file ## Agents SUPERVISOR ### Loan File Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Extraction Agent Classifies documents, extracts fields and tables classify_documentextract_fieldsocr_page aginies-talay-2 · on-prem ### Income Agent Computes net income, cross-checks against social security and bureau data compute_net_incomecross_check_sgkget_credit_bureaucompute_dti aginies-tulpar-4 · on-prem ### Valuation Agent Reads the appraisal, determines LTV and missing documents parse_appraisalcheck_ltvlist_missing_docsupdate_loan_filesend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Loan File Supervisor | File #ML-7731: 9 documents uploaded (payslips, social security record, appraisal, deed) | — | 240 | — | | 02 | retrieve | Extraction Agent | Documents fetched from S3 and classified; fields extracted from 3 payslips and the record | Document Store (S3) | 1,400 | 2,600 | | 03 | plan | Loan File Supervisor | Plan: income check → appraisal/LTV → missing docs → update file | — | 300 | 200 | | 04 | tool | Income Agent | Credit bureau report: score 1,612, existing instalments 9,800/month | Credit Bureau (KKB) | 480 | — | | 05 | verify | Income Agent | Net income 68,500, within 2% of the social security record, DTI 41% → eligible | — | 760 | 820 | | 06 | reason | Valuation Agent | Appraisal 6,400,000, requested 4,800,000 → LTV 75%, limit 80% | — | 720 | 690 | | 07 | verify | Valuation Agent | Missing: compulsory earthquake policy and last 3 months' statements → 2 documents | — | 520 | 430 | | 08 | approval | Loan File Supervisor | Loan > 3,000,000 → underwriting officer pre-approved the file | — | 3,100 | — | | 09 | write | Valuation Agent | LOS file set to "pre-approved – awaiting documents", checklist attached | Loan Origination System | 460 | — | | 10 | notify | Valuation Agent | Missing-document email sent to the customer and branch advisor | Email (Exchange) | 260 | — | Simulation metrics (estimated) 6,200runs / month 94.7%success 8.3 sp50 $0.0111est. cost / run Approval rule Loan > 3,000,000 or DTI > 45% → underwriting officer approval Step profile ingest · 1plan · 1retrieve · 1reason · 1tool · 1verify · 2approval · 1write · 1notify · 1 SYSTEMS - Loan Origination Systemcore - Document Store (S3)storage - Credit Bureau (KKB)api - Email (Exchange)email What you need to run it - Read/write API or MCP server and a service account for the core system - Read access to the document store and the document types - Credentials and a test environment for the enterprise APIs involved - Mailbox access (Microsoft 365 or Google Workspace) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the document classification and income verification automation used in US/UK mortgage origination Packages that include this scenario KVKK / GDPR personal data protection →Credit and onboarding operations → Sector: Banking → Related agents BankingHITL ### Dialogue banking: voice and text transaction assistant The customer talks or types; bill, limit, dispute and transfer are completed in core banking in one session. 186,000 runs / month 94.7% success 5 agents BankingHITL ### Credit card limit increase The customer types in the app or talks to the call centre; the decision takes seconds, and anything above threshold goes to a manager. 18,400 runs / month 96.4% success 4 agents BankingHITL ### Digital account opening (KYC) ID, liveness and sanctions checks in one flow; clean applications open an account in minutes, no branch visit. 42,000 runs / month 93.8% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # End-to-end motor claim: from notice to payment URL: https://www.aginies.com/en/agents/motor-claim-end-to-end Language: en Summary: Voice notice, photo assessment, repair-shop assignment and payment order in one flow; the adjuster steps in only above the threshold. Claim notice to the call centre or a photo report from the app → File opened, damage amount set, repair shop assigned, payment order issued Insurancevoiceapp # End-to-end motor claim: from notice to payment Voice notice, photo assessment, repair-shop assignment and payment order in one flow; the adjuster steps in only above the threshold. In short End-to-end motor claim: from notice to payment is an agent scenario in the Insurance sector running on the voice, app channels. Trigger: Claim notice to the call centre or a photo report from the app. The Claims Supervisor coordinates 4 specialist agents (Intake Agent, Photo Assessment Agent, Adjuster Agent, Communication Agent) across 5 systems (Call Centre SIP, Policy Admin, Claims System, Repair Network, Payment Hub); of 10 steps, 2 write to systems and 2 are checks or approvals. Outcome: File opened, damage amount set, repair shop assigned, payment order issued. Approval rule: Estimate > 150,000, unclear fault share or overlap with prior damage → adjuster approval. Insurance ### End-to-end motor claim: from notice to payment [End-to-end motor claim: from notice to payment] run #41,286,996elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Estimate > 150,000, unclear fault share or overlap with prior damage → adjuster approval Agents in this run Claims Supervisorsupervisorplans, routes, approves Intake Agentaginies-voice-tr-v3 · on-premTranscribes the voice notice, collects incident facts, matches the policy Photo Assessment Agentaginies-konrul-1 · on-premEstimates parts and labour from damage photos Adjuster Agentaginies-tulpar-4 · on-premEvaluates cover, deductible and fault share, assigns the shop, prepares payment Communication Agentaginies-sunkar-1 · on-premKeeps the customer and the shop informed at every step Audit trail00:00.00 - ◦Call: "rear-ended in a car park, the other driver admitted it, photos uploaded to the app" - ∿Transcript and incident form: date, place, other plate, no injuries; policy 34 ABC 123 active - ⌘Plan: photo assessment → cover and deductible → fault → shop → payment → subrogation - ∿6 photos: rear bumper replace, boot lid repair, 2 parking sensors; estimate 38,400 - ✓Compared with prior damage history: no overlapping area, photo metadata consistent - ✓Full comprehensive cover, zero deductible, other party 100% at fault (report) → no missing document - ⚙Network shop: 4 km away, capacity in 2 days, parts in stock; appointment proposed - ⇢File #CL-230517: 38,400 approved, shop assigned, subrogation file opened - ⇢Payment order to the shop on completion; parts advance 14,200 today - ➤Spoken summary and WhatsApp to the customer: appointment, courtesy car, file tracking link - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Claim notice to the call centre or a photo report from the app Outcome File opened, damage amount set, repair shop assigned, payment order issued ## Agents SUPERVISOR ### Claims Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-voice-tr-v3 · on-prem ### Intake Agent Transcribes the voice notice, collects incident facts, matches the policy transcribe_streamcollect_fnolmatch_policy aginies-konrul-1 · on-prem ### Photo Assessment Agent Estimates parts and labour from damage photos detect_damageestimate_repaircheck_prior_damage aginies-tulpar-4 · on-prem ### Adjuster Agent Evaluates cover, deductible and fault share, assigns the shop, prepares payment check_coverageassign_repair_shopprepare_paymentopen_subrogation aginies-sunkar-1 · on-prem ### Communication Agent Keeps the customer and the shop informed at every step send_smssend_whatsappnotify_shop ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Claims Supervisor | Call: "rear-ended in a car park, the other driver admitted it, photos uploaded to the app" | Call Centre SIP | 220 | — | | 02 | reason | Intake Agent | Transcript and incident form: date, place, other plate, no injuries; policy 34 ABC 123 active | Policy Admin | 1,400 | 520 | | 03 | plan | Claims Supervisor | Plan: photo assessment → cover and deductible → fault → shop → payment → subrogation | — | 260 | 240 | | 04 | reason | Photo Assessment Agent | 6 photos: rear bumper replace, boot lid repair, 2 parking sensors; estimate 38,400 | — | 2,600 | 1,900 | | 05 | verify | Photo Assessment Agent | Compared with prior damage history: no overlapping area, photo metadata consistent | Claims System | 700 | 380 | | 06 | verify | Adjuster Agent | Full comprehensive cover, zero deductible, other party 100% at fault (report) → no missing document | Policy Admin | 820 | 460 | | 07 | tool | Adjuster Agent | Network shop: 4 km away, capacity in 2 days, parts in stock; appointment proposed | Repair Network | 540 | — | | 08 | write | Adjuster Agent | File #CL-230517: 38,400 approved, shop assigned, subrogation file opened | Claims System | 620 | — | | 09 | write | Adjuster Agent | Payment order to the shop on completion; parts advance 14,200 today | Payment Hub | 480 | — | | 10 | notify | Communication Agent | Spoken summary and WhatsApp to the customer: appointment, courtesy car, file tracking link | Call Centre SIP | 440 | 200 | Simulation metrics (estimated) 21,000runs / month 91.2%success 9.6 sp50 $0.0062est. cost / run Approval rule Estimate > 150,000, unclear fault share or overlap with prior damage → adjuster approval Step profile ingest · 1plan · 1reason · 2tool · 1verify · 2write · 2notify · 1 SYSTEMS - Call Centre SIPvoice - Policy Admincore - Claims Systemcore - Repair Networkapi - Payment Hubpayments What you need to run it - SIP integration with the switch and recording access - Read/write API or MCP server and a service account for the core system - Credentials and a test environment for the enterprise APIs involved - Authorised API access and limits for payment services - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Digital insurers' "claim in minutes" pattern, combined with photo assessment and the repair network Packages that include this scenario Voice AI for claims intake →Claims operations end to end → Sector: Insurance → Related agents InsuranceHITL ### Voice claims intake (FNOL) The customer calls and describes it; policy found by plate, cover checked, claim opened and adjuster assigned. 34,000 runs / month 95.8% success 5 agents InsuranceHITL ### Policy renewal offer Policies re-priced, sharp increases go to approval; customer accepts with one tap and the policy is issued. 22,000 runs / month 94.4% success 4 agents InsuranceHITL ### Underwriting document extraction Medical reports and labs read and assessed against the manual; underwriters only see files above threshold. 4,800 runs / month 94.9% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Motor claim photo assessment URL: https://www.aginies.com/en/agents/motor-claim-photo-assessment Language: en Summary: Damage detected from photos, estimate built from parts and labour, adjuster only approves; repair slot booked. Photos uploaded to an open claim via WhatsApp or the app → Approved estimate and reserve on the claim, repair slot booked, customer informed Insuranceappwhatsapp # Motor claim photo assessment Damage detected from photos, estimate built from parts and labour, adjuster only approves; repair slot booked. In short Motor claim photo assessment is an agent scenario in the Insurance sector running on the app, whatsapp channels. Trigger: Photos uploaded to an open claim via WhatsApp or the app. The Claims Supervisor coordinates 3 specialist agents (Vision Agent, Estimate Agent, Repair Routing Agent) across 4 systems (Guidewire ClaimCenter, Parts & Labour Catalogue (Audatex), Repair Network Portal, WhatsApp Business API); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: Approved estimate and reserve on the claim, repair slot booked, customer informed. Approval rule: Estimate > 15,000 or model confidence < 80% → adjuster approval. Insurance ### Motor claim photo assessment [Motor claim photo assessment] run #41,287,258elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Estimate > 15,000 or model confidence < 80% → adjuster approval Agents in this run Claims Supervisorsupervisorplans, routes, approves Vision Agentaginies-konrul-1 · on-premAssesses photo quality and damage, checks for reuse or manipulation Estimate Agentaginies-tulpar-4 · on-premBuilds the estimate from parts and labour prices Repair Routing Agentaginies-talay-2 · on-premFinds a partner shop, books the slot, updates the claim Audit trail00:00.00 - ◦6 photos received on WhatsApp for claim #CL-2026-118342 - ✓Image quality sufficient, EXIF date matches the incident, no reused images - ∿Damage: front-right wing + bumper, medium depth, headlamp intact - ⌘Plan: parts/labour → estimate → threshold check → repair routing - ⚙Prices: wing 4,200, bumper 6,800, paint 3.5 hours - ∿Estimated cost 19,600 (± 12%), not a total loss, excess 2,000 - ⏸Estimate > 15,000 → adjuster approved in 3 minutes, one line item corrected - ⚙3 partner shops within 5 km, earliest slot tomorrow 14:00 - ⇢Estimate and repair booking written to the claim, reserve updated to 19,600 - ➤WhatsApp to customer: estimate summary, shop address and appointment - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Photos uploaded to an open claim via WhatsApp or the app Outcome Approved estimate and reserve on the claim, repair slot booked, customer informed ## Agents SUPERVISOR ### Claims Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Vision Agent Assesses photo quality and damage, checks for reuse or manipulation check_image_qualitydetect_damageestimate_severitydetect_reuse aginies-tulpar-4 · on-prem ### Estimate Agent Builds the estimate from parts and labour prices get_parts_pricesget_labour_ratesbuild_estimate aginies-talay-2 · on-prem ### Repair Routing Agent Finds a partner shop, books the slot, updates the claim find_repair_shopsbook_slotupdate_claimsend_whatsapp ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Claims Supervisor | 6 photos received on WhatsApp for claim #CL-2026-118342 | — | 240 | — | | 02 | verify | Vision Agent | Image quality sufficient, EXIF date matches the incident, no reused images | — | 620 | 540 | | 03 | reason | Vision Agent | Damage: front-right wing + bumper, medium depth, headlamp intact | — | 1,200 | 1,650 | | 04 | plan | Claims Supervisor | Plan: parts/labour → estimate → threshold check → repair routing | — | 280 | 190 | | 05 | tool | Estimate Agent | Prices: wing 4,200, bumper 6,800, paint 3.5 hours | Parts & Labour Catalogue (Audatex) | 460 | — | | 06 | reason | Estimate Agent | Estimated cost 19,600 (± 12%), not a total loss, excess 2,000 | — | 780 | 720 | | 07 | approval | Claims Supervisor | Estimate > 15,000 → adjuster approved in 3 minutes, one line item corrected | — | 2,900 | — | | 08 | tool | Repair Routing Agent | 3 partner shops within 5 km, earliest slot tomorrow 14:00 | Repair Network Portal | 380 | — | | 09 | write | Repair Routing Agent | Estimate and repair booking written to the claim, reserve updated to 19,600 | Guidewire ClaimCenter | 520 | — | | 10 | notify | Repair Routing Agent | WhatsApp to customer: estimate summary, shop address and appointment | WhatsApp Business API | 260 | — | Simulation metrics (estimated) 15,500runs / month 93.3%success 7.6 sp50 $0.0053est. cost / run Approval rule Estimate > 15,000 or model confidence < 80% → adjuster approval Step profile ingest · 1plan · 1reason · 2tool · 2verify · 1approval · 1write · 1notify · 1 SYSTEMS - Guidewire ClaimCentercore - Parts & Labour Catalogue (Audatex)api - Repair Network Portalapi - WhatsApp Business APIchannel What you need to run it - Read/write API or MCP server and a service account for the core system - Credentials and a test environment for the enterprise APIs involved - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the computer-vision instant-estimate pattern deployed at US/EU motor insurers Packages that include this scenario Claims operations end to end → Sector: Insurance → Related agents InsuranceHITL ### End-to-end motor claim: from notice to payment Voice notice, photo assessment, repair-shop assignment and payment order in one flow; the adjuster steps in only above the threshold. 21,000 runs / month 91.2% success 5 agents InsuranceHITL ### Voice claims intake (FNOL) The customer calls and describes it; policy found by plate, cover checked, claim opened and adjuster assigned. 34,000 runs / month 95.8% success 5 agents InsuranceHITL ### Policy renewal offer Policies re-priced, sharp increases go to approval; customer accepts with one tap and the policy is issued. 22,000 runs / month 94.4% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Network outage communications URL: https://www.aginies.com/en/agents/network-outage-communications Language: en Summary: When an outage ticket opens, affected subscribers are found and the approved message sent before they call. P1/P2 network outage ticket in ServiceNow → Affected subscribers informed, ticket updated, "restored" message scheduled Telecomsmswhatsappapibatch # Network outage communications When an outage ticket opens, affected subscribers are found and the approved message sent before they call. In short Network outage communications is an agent scenario in the Telecom sector running on the sms, whatsapp, api, batch channels. Trigger: P1/P2 network outage ticket in ServiceNow. The Outage Communications Supervisor coordinates 3 specialist agents (Detection Agent, Impact Agent, Communications Agent) across 4 systems (ServiceNow, Network OSS, Salesforce CRM, SMS Gateway); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: Affected subscribers informed, ticket updated, "restored" message scheduled. Approval rule: Affected subscribers > 10,000 or P1 outage → network operations manager approval. Telecom ### Network outage communications [Network outage communications] run #41,288,173elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Affected subscribers > 10,000 or P1 outage → network operations manager approval Agents in this run Outage Communications Supervisorsupervisorplans, routes, approves Detection Agentaginies-talay-2 · on-premGets the cause, cells and restoration estimate of the outage Impact Agentaginies-sunkar-1 · on-premMaps cells to subscribers, derives segment and channel preference Communications Agentaginies-tulpar-4 · on-premDrafts the message, runs guardrails, sends in bulk Audit trail00:00.00 - ◦ServiceNow INC-204188: 14 cell sites unreachable in Bursa Nilüfer, priority P1 - ⚙OSS: fibre cut, 14 cells, estimated restoration 3 h 20 min - ⌘Plan: affected subscribers → segment → message → approval → send → update - ⚙38,400 affected subscribers; 2,150 business, 310 SLA lines - ∿Segments: business → account manager + SMS; consumer → SMS or WhatsApp by preference - ∿Message drafted: cause, affected area, ETA, apology; 148 characters - ✓Guardrail: no personal data, no SLA commitment implied, on-brand tone - ⏸> 10,000 subscribers → network operations manager approved in 4 minutes - ⇢INC-204188 updated to "customers informed", status page published - ➤36,250 SMS + 2,150 WhatsApp sent; "restored" message scheduled for after repair - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger P1/P2 network outage ticket in ServiceNow Outcome Affected subscribers informed, ticket updated, "restored" message scheduled ## Agents SUPERVISOR ### Outage Communications Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Detection Agent Gets the cause, cells and restoration estimate of the outage get_incidentget_affected_cellsestimate_restoration aginies-sunkar-1 · on-prem ### Impact Agent Maps cells to subscribers, derives segment and channel preference map_cells_to_subscriberssegment_customersget_contact_preferences aginies-tulpar-4 · on-prem ### Communications Agent Drafts the message, runs guardrails, sends in bulk draft_messagecheck_guardrailssend_sms_batchsend_whatsapp_batchupdate_incident ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Outage Communications Supervisor | ServiceNow INC-204188: 14 cell sites unreachable in Bursa Nilüfer, priority P1 | — | 220 | — | | 02 | tool | Detection Agent | OSS: fibre cut, 14 cells, estimated restoration 3 h 20 min | Network OSS | 460 | — | | 03 | plan | Outage Communications Supervisor | Plan: affected subscribers → segment → message → approval → send → update | — | 260 | 200 | | 04 | tool | Impact Agent | 38,400 affected subscribers; 2,150 business, 310 SLA lines | Salesforce CRM | 720 | — | | 05 | reason | Impact Agent | Segments: business → account manager + SMS; consumer → SMS or WhatsApp by preference | — | 540 | 460 | | 06 | reason | Communications Agent | Message drafted: cause, affected area, ETA, apology; 148 characters | — | 620 | 540 | | 07 | verify | Communications Agent | Guardrail: no personal data, no SLA commitment implied, on-brand tone | — | 380 | 300 | | 08 | approval | Outage Communications Supervisor | > 10,000 subscribers → network operations manager approved in 4 minutes | — | 2,800 | — | | 09 | write | Communications Agent | INC-204188 updated to "customers informed", status page published | ServiceNow | 420 | — | | 10 | notify | Communications Agent | 36,250 SMS + 2,150 WhatsApp sent; "restored" message scheduled for after repair | SMS Gateway | 900 | — | Simulation metrics (estimated) 1,400runs / month 98.5%success 7.3 sp50 $0.0075est. cost / run Approval rule Affected subscribers > 10,000 or P1 outage → network operations manager approval Step profile ingest · 1plan · 1reason · 2tool · 2verify · 1approval · 1write · 1notify · 1 SYSTEMS - ServiceNowticketing - Network OSSapi - Salesforce CRMcrm - SMS Gatewaychannel What you need to run it - API access and assignment rules for the ticketing or work-order system - Credentials and a test environment for the enterprise APIs involved - CRM API access and a customer matching key - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the proactive outage-notification pattern large operators use to cut inbound call volume by around 30% Packages that include this scenario Subscriber care and network support → Sector: Telecom → Related agents TelecomHITL ### Subscriber care assistant: bill, plan, fault and contract The subscriber calls or writes; bill dispute, plan change and line fault are resolved in BSS and network systems in one session. 410,000 runs / month 94.3% success 5 agents TelecomHITL ### Bill dispute resolution Invoice lines and usage explained in seconds; goodwill credit applied by policy, larger amounts need a lead. 88,000 runs / month 96.3% success 4 agents TelecomHITL ### SIM swap fraud check Every SIM swap is risk-scored before provisioning; suspicious requests are held and banks are signalled. 41,000 runs / month 97.7% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Where is my order? (WISMO) URL: https://www.aginies.com/en/agents/order-status-wismo Language: en Summary: Customer asks, order and carrier events join instantly; on delay a new ETA is given and a trace opened. "Where is my order?" message on WhatsApp, web or app → Customer knows the status and new ETA, carrier trace opened, re-check scheduled Retail & E-commercewhatsappwebappvoice # Where is my order? (WISMO) Customer asks, order and carrier events join instantly; on delay a new ETA is given and a trace opened. In short Where is my order? (WISMO) is an agent scenario in the Retail & E-commerce sector running on the whatsapp, web, app, voice channels. Trigger: "Where is my order?" message on WhatsApp, web or app. The Order Supervisor coordinates 3 specialist agents (Intent Agent, Tracking Agent, Reply Agent) across 4 systems (Shopify / OMS, Carrier API (Yurtiçi, DHL), WhatsApp Business API, Zendesk); of 9 steps, 1 write to systems and 1 are checks or approvals. Outcome: Customer knows the status and new ETA, carrier trace opened, re-check scheduled. Retail & E-commerce ### Where is my order? (WISMO) [Where is my order? (WISMO)] run #41,287,736elapsed 0 mstokens 0est. cost $0.000status — Agents in this run Order Supervisorsupervisorplans, routes, approves Intent Agentaginies-sunkar-1 · on-premUnderstands the request, verifies order number and customer Tracking Agentaginies-talay-2 · on-premJoins order and carrier events, estimates the revised ETA Reply Agentaginies-tulpar-4 · on-premApplies policy, composes the reply, opens a ticket when needed Audit trail00:00.00 - ◦WhatsApp: "my order hasn't arrived in 5 days, number 100482" - ∿Intent: WISMO · order #100482 · phone matched → verified - ⚙Order has 3 items, shipped 2 Sep, tracking YK-7731 8820 - ⚙Carrier events: held at the transfer hub since 4 Sep, no movement for 2 days - ⌘Plan: delay class → revised ETA → policy → carrier trace → reply - ∿Delay is carrier-side; revised ETA 10 Sep, confidence 82% - ✓Policy: delay > 7 days → 100 voucher; this case 5 days → information only - ⇢Zendesk ticket #ZD-51937 opened, trace request sent to the carrier - ➤WhatsApp: status, revised ETA and live tracking link; automatic re-check on 10 Sep - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger "Where is my order?" message on WhatsApp, web or app Outcome Customer knows the status and new ETA, carrier trace opened, re-check scheduled ## Agents SUPERVISOR ### Order Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Intent Agent Understands the request, verifies order number and customer detect_intentextract_order_refverify_customer aginies-talay-2 · on-prem ### Tracking Agent Joins order and carrier events, estimates the revised ETA get_orderget_shipmentget_carrier_eventsestimate_eta aginies-tulpar-4 · on-prem ### Reply Agent Applies policy, composes the reply, opens a ticket when needed check_delay_policycompose_replysend_whatsappcreate_ticket ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Order Supervisor | WhatsApp: "my order hasn't arrived in 5 days, number 100482" | — | 180 | — | | 02 | reason | Intent Agent | Intent: WISMO · order #100482 · phone matched → verified | — | 420 | 260 | | 03 | tool | Tracking Agent | Order has 3 items, shipped 2 Sep, tracking YK-7731 8820 | Shopify / OMS | 340 | — | | 04 | tool | Tracking Agent | Carrier events: held at the transfer hub since 4 Sep, no movement for 2 days | Carrier API (Yurtiçi, DHL) | 460 | — | | 05 | plan | Order Supervisor | Plan: delay class → revised ETA → policy → carrier trace → reply | — | 240 | 170 | | 06 | reason | Tracking Agent | Delay is carrier-side; revised ETA 10 Sep, confidence 82% | — | 520 | 430 | | 07 | verify | Reply Agent | Policy: delay > 7 days → 100 voucher; this case 5 days → information only | — | 380 | 290 | | 08 | write | Reply Agent | Zendesk ticket #ZD-51937 opened, trace request sent to the carrier | Zendesk | 400 | — | | 09 | notify | Reply Agent | WhatsApp: status, revised ETA and live tracking link; automatic re-check on 10 Sep | WhatsApp Business API | 300 | 240 | Simulation metrics (estimated) 110,000runs / month 98.1%success 3.3 sp50 $0.0006est. cost / run Step profile ingest · 1plan · 1reason · 2tool · 2verify · 1write · 1notify · 1 SYSTEMS - Shopify / OMSerp - Carrier API (Yurtiçi, DHL)api - WhatsApp Business APIchannel - Zendeskticketing What you need to run it - Posting and query API access to ERP/MES - Credentials and a test environment for the enterprise APIs involved - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - API access and assignment rules for the ticketing or work-order system - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the proactive-notification pattern large e-commerce players use to deflect over 60% of WISMO contacts Packages that include this scenario E-commerce support & action → Sector: Retail & E-commerce → Related agents Retail & E-commerceHITL ### Outfit stylist: a complete look by size, stock, budget and style The customer sends a photo and a sentence; the agent builds three looks that are in stock, fit and within budget, and adds the chosen one to the cart. 142,000 runs / month 96.1% success 5 agents Retail & E-commerceHITL ### Returns and refunds Policy and history checked in seconds, label issued; the refund executes when the carrier scans the parcel. 48,000 runs / month 96.7% success 4 agents Retail & E-commerce ### Product Q&A assistant Size, compatibility and spec questions answered from catalogue and return data; stock and price verified. 120,000 runs / month 98.7% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Staged OTA update rollout URL: https://www.aginies.com/en/agents/ota-update-staged-rollout Language: en Summary: Firmware rolls out in 1% → 10% → 100% stages; a health gate at every stage, human approval for critical fleets. Signed firmware release event from CI/CD → Release reached the fleet through health gates or was rolled back automatically; every step recorded IoT & Connected Devicesapibatch # Staged OTA update rollout Firmware rolls out in 1% → 10% → 100% stages; a health gate at every stage, human approval for critical fleets. In short Staged OTA update rollout is an agent scenario in the IoT & Connected Devices sector running on the api, batch channels. Trigger: Signed firmware release event from CI/CD. The Rollout Supervisor coordinates 3 specialist agents (Plan Agent, Health Agent, Deploy Agent) across 4 systems (CI/CD (signed artefacts), Device Management / OTA, Observability (metrics), Slack); of 9 steps, 2 write to systems and 2 are checks or approvals. Outcome: Release reached the fleet through health gates or was rolled back automatically; every step recorded. Approval rule: Rollout to a critical fleet segment → product owner approval; automatic rollback if a health gate fails. IoT & Connected Devices ### Staged OTA update rollout [Staged OTA update rollout] run #41,290,162elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Rollout to a critical fleet segment → product owner approval; automatic rollback if a health gate fails Agents in this run Rollout Supervisorsupervisorplans, routes, approves Plan Agentaginies-talay-2 · on-premChooses stages and the canary group Health Agentaginies-tulpar-4 · on-premEvaluates metrics after each stage Deploy Agentaginies-sunkar-1 · on-premStarts a stage, rolls back if needed Audit trail00:00.00 - ◦Release 5.0.3 signed: 46,200 target devices, 3 fleet segments - ⚙Segments pulled: warehouse sensors, vehicle telematics, critical cold chain - ∿Plan: 1% canary (462 devices, low risk) → 10% → 100%; critical segment last and approved - ⇢Stage 1 started: 462 devices, 30-minute observation window - ⚙Stage metrics: connectivity 99.6%, crashes 0, battery drain +1.2% (threshold 5%) - ✓Health gate passed; difference from baseline not statistically significant - ⇢Stage 2 started: 4,620 devices; gate passed, stage 3 ready - ⏸Critical cold-chain segment → product owner approved the 100% rollout - ➤Team report: 3 stages, 0 rollbacks, 46,200 devices on 5.0.3; record stored with the release - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Signed firmware release event from CI/CD Outcome Release reached the fleet through health gates or was rolled back automatically; every step recorded ## Agents SUPERVISOR ### Rollout Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Plan Agent Chooses stages and the canary group get_fleet_segmentsselect_canarycreate_rollout_plan aginies-tulpar-4 · on-prem ### Health Agent Evaluates metrics after each stage get_stage_metricscompare_baselinedecide_gate aginies-sunkar-1 · on-prem ### Deploy Agent Starts a stage, rolls back if needed start_stagepause_rolloutrollback_stagenotify_team ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Rollout Supervisor | Release 5.0.3 signed: 46,200 target devices, 3 fleet segments | CI/CD (signed artefacts) | 200 | — | | 02 | tool | Plan Agent | Segments pulled: warehouse sensors, vehicle telematics, critical cold chain | Device Management / OTA | 420 | — | | 03 | reason | Plan Agent | Plan: 1% canary (462 devices, low risk) → 10% → 100%; critical segment last and approved | — | 700 | 820 | | 04 | write | Deploy Agent | Stage 1 started: 462 devices, 30-minute observation window | Device Management / OTA | 480 | — | | 05 | tool | Health Agent | Stage metrics: connectivity 99.6%, crashes 0, battery drain +1.2% (threshold 5%) | Observability (metrics) | 520 | — | | 06 | verify | Health Agent | Health gate passed; difference from baseline not statistically significant | — | 560 | 520 | | 07 | write | Deploy Agent | Stage 2 started: 4,620 devices; gate passed, stage 3 ready | Device Management / OTA | 520 | — | | 08 | approval | Rollout Supervisor | Critical cold-chain segment → product owner approved the 100% rollout | — | 2,800 | — | | 09 | notify | Deploy Agent | Team report: 3 stages, 0 rollbacks, 46,200 devices on 5.0.3; record stored with the release | Slack | 240 | — | Simulation metrics (estimated) 320runs / month 98.1%success 6.9 sp50 $0.0083est. cost / run Approval rule Rollout to a critical fleet segment → product owner approval; automatic rollback if a health gate fails Step profile ingest · 1reason · 1tool · 2verify · 1approval · 1write · 2notify · 1 SYSTEMS - CI/CD (signed artefacts)api - Device Management / OTAcore - Observability (metrics)db - Slackchannel What you need to run it - Credentials and a test environment for the enterprise APIs involved - Read/write API or MCP server and a service account for the core system - Read access to the database or vector store; a data owner - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors canary and health-gated OTA rollout practices of large IoT operators Packages that include this scenario IoT device fleet operations → Sector: IoT & Connected Devices → Related agents IoT & Connected DevicesHITL ### Field device support assistant: telemetry, remote diagnosis, OTA and technician The customer says "the device is dead"; telemetry read, remote diagnosis made, update pushed, technician dispatched with the right part if needed. 37,000 runs / month 94.1% success 5 agents IoT & Connected DevicesHITL ### Device fleet anomaly triage Telemetry anomalies from thousands of devices are clustered, root-cause candidates listed, one incident reaches the right team. 18,000 runs / month 96.9% success 4 agents IoT & Connected DevicesHITL ### Cold-chain excursion response A temperature excursion is assessed instantly, product risk computed, a field crew dispatched, the quality decision recorded. 2,700 runs / month 97.4% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Outage report handling URL: https://www.aginies.com/en/agents/outage-report-handling Language: en Summary: Reports are clustered by feeder, one OSS ticket is raised and everyone affected hears before the crew leaves. Voice call to the 186 line or an outage report in the app → One OSS ticket raised, crew dispatched, affected customers got an ETA and a closure message Energy & Utilitiesvoiceappwhatsappweb # Outage report handling Reports are clustered by feeder, one OSS ticket is raised and everyone affected hears before the crew leaves. In short Outage report handling is an agent scenario in the Energy & Utilities sector running on the voice, app, whatsapp, web channels. Trigger: Voice call to the 186 line or an outage report in the app. The Outage Supervisor coordinates 4 specialist agents (Voice Agent, Report Agent, Clustering Agent, Comms Agent) across 5 systems (Customer CRM / Billing, SCADA / Historian, Network OSS, WhatsApp Business API, SMS Gateway); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: One OSS ticket raised, crew dispatched, affected customers got an ETA and a closure message. Approval rule: Mass notification > 1,000 customers → control room dispatcher approval. Energy & Utilities ### Outage report handling [Outage report handling] run #41,288,490elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Mass notification > 1,000 customers → control room dispatcher approval Agents in this run Outage Supervisorsupervisorplans, routes, approves Voice Agentaginies-voice-tr-v3 · on-premTranscribes the voice call Report Agentaginies-sunkar-1 · on-premExtracts address, meter and intent Clustering Agentaginies-tulpar-4 · on-premCorrelates reports with SCADA and feeder topology Comms Agentaginies-talay-2 · on-premRaises the OSS ticket and keeps customers informed Audit trail00:00.00 - ◦Call to the 186 line: "no power since 20:15", caller verified by phone number - ∿Transcribed (38 s); intent: outage report; address: Kadıköy, Moda Cad. - ⚙Meter 4020-77812 on feeder KD-17; 3 more reports from the same feeder in 12 min - ⚙SCADA: breaker on KD-17 tripped at 20:14, 1,840 customers affected - ∿Reports clustered into one incident; likely feeder fault, not a single premise - ✓Deduplication: no open OSS ticket for KD-17 → new ticket required - ⇢Ticket OSS-2026-51873 raised, priority P1, crew dispatch requested - ⏸Mass message > 1,000 customers → control room dispatcher approved - ➤Caller and 1,640 opted-in customers told: crew on the way, ETA 22:30 - ➤Power restored at 22:12 → closure SMS sent, 2 single-premise cases escalated - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Voice call to the 186 line or an outage report in the app Outcome One OSS ticket raised, crew dispatched, affected customers got an ETA and a closure message ## Agents SUPERVISOR ### Outage Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-voice-tr-v3 · on-prem ### Voice Agent Transcribes the voice call transcribe_audio aginies-sunkar-1 · on-prem ### Report Agent Extracts address, meter and intent detect_intentextract_locationget_customer_meter aginies-tulpar-4 · on-prem ### Clustering Agent Correlates reports with SCADA and feeder topology get_feeder_topologyget_scada_eventscluster_reportsfind_open_ticket aginies-talay-2 · on-prem ### Comms Agent Raises the OSS ticket and keeps customers informed create_oss_ticketupdate_oss_ticketsend_whatsappsend_sms ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Outage Supervisor | Call to the 186 line: "no power since 20:15", caller verified by phone number | — | 200 | — | | 02 | reason | Voice Agent | Transcribed (38 s); intent: outage report; address: Kadıköy, Moda Cad. | — | 780 | 340 | | 03 | tool | Report Agent | Meter 4020-77812 on feeder KD-17; 3 more reports from the same feeder in 12 min | Customer CRM / Billing | 420 | — | | 04 | tool | Clustering Agent | SCADA: breaker on KD-17 tripped at 20:14, 1,840 customers affected | SCADA / Historian | 380 | — | | 05 | reason | Clustering Agent | Reports clustered into one incident; likely feeder fault, not a single premise | — | 640 | 720 | | 06 | verify | Clustering Agent | Deduplication: no open OSS ticket for KD-17 → new ticket required | — | 300 | 180 | | 07 | write | Comms Agent | Ticket OSS-2026-51873 raised, priority P1, crew dispatch requested | Network OSS | 520 | — | | 08 | approval | Outage Supervisor | Mass message > 1,000 customers → control room dispatcher approved | — | 2,200 | — | | 09 | notify | Comms Agent | Caller and 1,640 opted-in customers told: crew on the way, ETA 22:30 | WhatsApp Business API | 300 | — | | 10 | notify | Comms Agent | Power restored at 22:12 → closure SMS sent, 2 single-premise cases escalated | SMS Gateway | 260 | — | Simulation metrics (estimated) 38,000runs / month 97.7%success 6 sp50 $0.0015est. cost / run Approval rule Mass notification > 1,000 customers → control room dispatcher approval Step profile ingest · 1reason · 2tool · 2verify · 1approval · 1write · 1notify · 2 SYSTEMS - Customer CRM / Billingcrm - SCADA / Historiancore - Network OSSticketing - WhatsApp Business APIchannel - SMS Gatewaychannel What you need to run it - CRM API access and a customer matching key - Read/write API or MCP server and a service account for the core system - API access and assignment rules for the ticketing or work-order system - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors outage communication practices of European DSOs with proactive customer updates Packages that include this scenario Utility customer operations → Sector: Energy & Utilities → Related agents Energy & UtilitiesHITL ### Utility customer assistant: bill, outage, consumption and instalments in one dialogue The customer calls or writes; the high bill is explained, outage ETA given, meter data shown, an instalment plan set up. 520,000 runs / month 94.6% success 5 agents Energy & UtilitiesHITL ### Meter reading dispute Customer sends a meter photo; the index is read by vision, the estimated bill fixed, big credits approved. 9,200 runs / month 94.9% success 4 agents Energy & UtilitiesHITL ### Rooftop solar connection application Application documents are read, transformer capacity and regulation checked, gaps reported the same day; the connection opinion belongs to the engineer. 5,400 runs / month 95.3% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Outfit stylist: a complete look by size, stock, budget and style URL: https://www.aginies.com/en/agents/outfit-stylist-agent Language: en Summary: The customer sends a photo and a sentence; the agent builds three looks that are in stock, fit and within budget, and adds the chosen one to the cart. "What do I wear with this" request over the app, web or WhatsApp → Three looks presented, one in the cart; size profile and preferences saved Retail & E-commerceappwebwhatsapp # Outfit stylist: a complete look by size, stock, budget and style The customer sends a photo and a sentence; the agent builds three looks that are in stock, fit and within budget, and adds the chosen one to the cart. In short Outfit stylist: a complete look by size, stock, budget and style is an agent scenario in the Retail & E-commerce sector running on the app, web, whatsapp channels. Trigger: "What do I wear with this" request over the app, web or WhatsApp. The Style Supervisor coordinates 4 specialist agents (Vision Agent, Stylist Agent, Catalogue Agent, Cart Agent) across 5 systems (WhatsApp Business, Product Catalogue (PIM), Inventory, Customer Profile, Cart & Checkout API); of 10 steps, 2 write to systems and 1 are checks or approvals. Outcome: Three looks presented, one in the cart; size profile and preferences saved. Approval rule: Discount outside campaign rules or a bespoke price request → customer service. Retail & E-commerce ### Outfit stylist: a complete look by size, stock, budget and style [Outfit stylist: a complete look by size, stock, budget and style] run #41,290,510elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Discount outside campaign rules or a bespoke price request → customer service Agents in this run Style Supervisorsupervisorplans, routes, approves Vision Agentaginies-konrul-1 · on-premExtracts type, colour, fit and fabric of the submitted garment Stylist Agentaginies-tulpar-4 · on-premBuilds and explains looks from style rules, size profile and budget Catalogue Agentaginies-sunkar-1 · on-premFetches candidate products filtered by stock, size and price Cart Agentaginies-sunkar-1 · on-premAdds the choice to the cart, states delivery and return terms Audit trail00:00.00 - ◦WhatsApp: jacket photo + "Friday dinner, smart casual, up to 3,000" - ∿Garment: navy wool-blend blazer, slim fit; matching catalogue item found - ⌕Size profile: 50 / 32-32 / 43; past returns show narrow-fit trouser deviation - ⌘Plan: fetch candidates → drop return-risk items → build 3 looks → confirm stock → cart - ⚙Shirt, trousers, shoes, accessories: 38 candidates after size and budget filter, all in stock - ✓5 items with return rate > 35% removed; 2 trousers with size-chart deviation dropped - ∿3 looks: 2,740 / 2,890 / 2,980; with colour and fit reasoning, visual collage prepared - ➤Looks sent on WhatsApp; the customer picked look 2 and asked for shoes in 44 - ⇢4 items added to the cart, shoes in 44 confirmed; 2-hour store pickup available - ⇢Size profile and style preferences updated; note stored for the next recommendation - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger "What do I wear with this" request over the app, web or WhatsApp Outcome Three looks presented, one in the cart; size profile and preferences saved ## Agents SUPERVISOR ### Style Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Vision Agent Extracts type, colour, fit and fabric of the submitted garment describe_garmentmatch_catalogue_item aginies-tulpar-4 · on-prem ### Stylist Agent Builds and explains looks from style rules, size profile and budget get_style_rulesget_size_profilecompose_outfits aginies-sunkar-1 · on-prem ### Catalogue Agent Fetches candidate products filtered by stock, size and price search_cataloguecheck_stockget_return_rate aginies-sunkar-1 · on-prem ### Cart Agent Adds the choice to the cart, states delivery and return terms add_to_cartget_delivery_optionssave_preferences ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Style Supervisor | WhatsApp: jacket photo + "Friday dinner, smart casual, up to 3,000" | WhatsApp Business | 180 | — | | 02 | reason | Vision Agent | Garment: navy wool-blend blazer, slim fit; matching catalogue item found | Product Catalogue (PIM) | 1,300 | 760 | | 03 | retrieve | Stylist Agent | Size profile: 50 / 32-32 / 43; past returns show narrow-fit trouser deviation | Customer Profile | 420 | 180 | | 04 | plan | Style Supervisor | Plan: fetch candidates → drop return-risk items → build 3 looks → confirm stock → cart | — | 240 | 220 | | 05 | tool | Catalogue Agent | Shirt, trousers, shoes, accessories: 38 candidates after size and budget filter, all in stock | Inventory | 560 | — | | 06 | verify | Catalogue Agent | 5 items with return rate > 35% removed; 2 trousers with size-chart deviation dropped | — | 380 | 150 | | 07 | reason | Stylist Agent | 3 looks: 2,740 / 2,890 / 2,980; with colour and fit reasoning, visual collage prepared | — | 2,200 | 1,500 | | 08 | notify | Cart Agent | Looks sent on WhatsApp; the customer picked look 2 and asked for shoes in 44 | WhatsApp Business | 1,800 | 240 | | 09 | write | Cart Agent | 4 items added to the cart, shoes in 44 confirmed; 2-hour store pickup available | Cart & Checkout API | 520 | — | | 10 | write | Cart Agent | Size profile and style preferences updated; note stored for the next recommendation | Customer Profile | 300 | — | Simulation metrics (estimated) 142,000runs / month 96.1%success 6.1 sp50 $0.0016est. cost / run Approval rule Discount outside campaign rules or a bespoke price request → customer service Step profile ingest · 1plan · 1retrieve · 1reason · 2tool · 1verify · 1write · 2notify · 1 SYSTEMS - WhatsApp Businesschannel - Product Catalogue (PIM)db - Inventoryerp - Customer Profilecrm - Cart & Checkout APIapi What you need to run it - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - Read access to the database or vector store; a data owner - Posting and query API access to ERP/MES - CRM API access and a customer matching key - Credentials and a test environment for the enterprise APIs involved - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Large fashion retailers' stylist assistant pattern, combined with stock, size and return data Packages that include this scenario E-commerce support & action → Sector: Retail & E-commerce → Related agents Retail & E-commerce ### Where is my order? (WISMO) Customer asks, order and carrier events join instantly; on delay a new ETA is given and a trace opened. 110,000 runs / month 98.1% success 4 agents Retail & E-commerceHITL ### Returns and refunds Policy and history checked in seconds, label issued; the refund executes when the carrier scans the parcel. 48,000 runs / month 96.7% success 4 agents Retail & E-commerce ### Product Q&A assistant Size, compatibility and spec questions answered from catalogue and return data; stock and price verified. 120,000 runs / month 98.7% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Patient intake and triage URL: https://www.aginies.com/en/agents/patient-intake-triage Language: en Summary: The patient types or talks; intake and urgency scoring take seconds, and a nurse always makes the final call. Voice or text symptom report in the app or on WhatsApp → Urgency confirmed by a nurse, appointment booked in the right clinic, patient guided Healthcareappvoicewhatsapp # Patient intake and triage The patient types or talks; intake and urgency scoring take seconds, and a nurse always makes the final call. In short Patient intake and triage is an agent scenario in the Healthcare sector running on the app, voice, whatsapp channels. Trigger: Voice or text symptom report in the app or on WhatsApp. The Triage Supervisor coordinates 4 specialist agents (Voice Agent, Symptom Agent, Urgency Agent, Routing Agent) across 4 systems (EHR (HL7/FHIR), HIS Scheduling, Triage Protocol Vector Store, WhatsApp Business API); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: Urgency confirmed by a nurse, appointment booked in the right clinic, patient guided. Approval rule: A triage nurse confirms the urgency level on every case; "emergency" cases are directed to 112 immediately. Healthcare ### Patient intake and triage [Patient intake and triage] run #41,289,972elapsed 0 mstokens 0est. cost $0.000status —Approval rule: A triage nurse confirms the urgency level on every case; "emergency" cases are directed to 112 immediately Agents in this run Triage Supervisorsupervisorplans, routes, approves Voice Agentaginies-voice-tr-v3 · on-premTranscribes the voice message and detects language Symptom Agentaginies-sunkar-1 · on-premAsks follow-up questions and structures the complaints Urgency Agentaginies-tulpar-4 · on-premScores urgency against protocol; never makes the final decision Routing Agentaginies-talay-2 · on-premFinds the right clinic and slot, books the appointment Audit trail00:00.00 - ◦Voice message received in the app, patient identity verified via OTP - ∿Transcribed (Turkish, 42 s): "chest tightness for two days, worse on the stairs" - ⚙History pulled: 58 y, hypertension, no prior cardiac events - ∿3 follow-up questions asked: no radiating pain, no shortness of breath at rest - ⌕Manchester Triage discriminators for chest pain retrieved - ∿Urgency score 71/100 → "urgent" (same-day cardiology), red flags: 1 of 6 - ✓Guardrail: no diagnosis given to the patient, emergency wording checked - ⏸Triage nurse confirmed the urgency level on the dashboard (mandatory for every case) - ⇢Cardiology slot booked today 15:40, Dr Kaya, room C-204 - ➤Patient sent the slot, directions and a "call 112 if worse" instruction on WhatsApp - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Voice or text symptom report in the app or on WhatsApp Outcome Urgency confirmed by a nurse, appointment booked in the right clinic, patient guided ## Agents SUPERVISOR ### Triage Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-voice-tr-v3 · on-prem ### Voice Agent Transcribes the voice message and detects language transcribe_audiodetect_language aginies-sunkar-1 · on-prem ### Symptom Agent Asks follow-up questions and structures the complaints ask_followupextract_symptomsget_patient_history aginies-tulpar-4 · on-prem ### Urgency Agent Scores urgency against protocol; never makes the final decision search_triage_protocolscore_urgencycheck_red_flags aginies-talay-2 · on-prem ### Routing Agent Finds the right clinic and slot, books the appointment find_clinic_slotbook_appointmentsend_whatsapp ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Triage Supervisor | Voice message received in the app, patient identity verified via OTP | — | 220 | — | | 02 | reason | Voice Agent | Transcribed (Turkish, 42 s): "chest tightness for two days, worse on the stairs" | — | 900 | 380 | | 03 | tool | Symptom Agent | History pulled: 58 y, hypertension, no prior cardiac events | EHR (HL7/FHIR) | 430 | — | | 04 | reason | Symptom Agent | 3 follow-up questions asked: no radiating pain, no shortness of breath at rest | — | 1,100 | 720 | | 05 | retrieve | Urgency Agent | Manchester Triage discriminators for chest pain retrieved | Triage Protocol Vector Store | 380 | 520 | | 06 | reason | Urgency Agent | Urgency score 71/100 → "urgent" (same-day cardiology), red flags: 1 of 6 | — | 760 | 940 | | 07 | verify | Urgency Agent | Guardrail: no diagnosis given to the patient, emergency wording checked | — | 340 | 260 | | 08 | approval | Triage Supervisor | Triage nurse confirmed the urgency level on the dashboard (mandatory for every case) | — | 2,800 | — | | 09 | write | Routing Agent | Cardiology slot booked today 15:40, Dr Kaya, room C-204 | HIS Scheduling | 480 | — | | 10 | notify | Routing Agent | Patient sent the slot, directions and a "call 112 if worse" instruction on WhatsApp | WhatsApp Business API | 260 | — | Simulation metrics (estimated) 24,500runs / month 95.8%success 7.7 sp50 $0.0034est. cost / run Approval rule A triage nurse confirms the urgency level on every case; "emergency" cases are directed to 112 immediately Step profile ingest · 1retrieve · 1reason · 3tool · 1verify · 1approval · 1write · 1notify · 1 SYSTEMS - EHR (HL7/FHIR)core - HIS Schedulingcore - Triage Protocol Vector Storedb - WhatsApp Business APIchannel What you need to run it - Read/write API or MCP server and a service account for the core system - Read access to the database or vector store; a data owner - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the digital front-door and symptom assessment patterns of NHS 111 and large hospital groups Packages that include this scenario Healthcare operations & clinical guidance → Sector: Healthcare → Related agents HealthcareHITL ### Patient journey coordinator: appointment, preparation, approval and results The patient writes or calls; appointment, preparation instructions, insurer pre-approval and result notice come together in one coordinator. 58,000 runs / month 95.1% success 5 agents HealthcareHITL ### Prior authorization Payer rules searched, clinical evidence gathered and the packet ready to submit within minutes of the order. 6,200 runs / month 94.1% success 4 agents Healthcare ### Appointment scheduling and no-show reduction High no-show-risk patients are reached on WhatsApp in advance; freed slots are refilled from the waiting list. 62,000 runs / month 98.2% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Patient journey coordinator: appointment, preparation, approval and results URL: https://www.aginies.com/en/agents/patient-journey-coordinator Language: en Summary: The patient writes or calls; appointment, preparation instructions, insurer pre-approval and result notice come together in one coordinator. Appointment request over WhatsApp or the call centre after a physician order → Appointment booked, preparation sent, pre-approval obtained, result reminder scheduled Healthcarewhatsappvoiceapp # Patient journey coordinator: appointment, preparation, approval and results The patient writes or calls; appointment, preparation instructions, insurer pre-approval and result notice come together in one coordinator. In short Patient journey coordinator: appointment, preparation, approval and results is an agent scenario in the Healthcare sector running on the whatsapp, voice, app channels. Trigger: Appointment request over WhatsApp or the call centre after a physician order. The Patient Journey Supervisor coordinates 4 specialist agents (Dialogue Agent, Scheduling Agent, Clinical Preparation Agent, Insurer Agent) across 5 systems (WhatsApp Business, Hospital Information System, Scheduling, Insurer Portal, Lab / Imaging); of 10 steps, 3 write to systems and 3 are checks or approvals. Outcome: Appointment booked, preparation sent, pre-approval obtained, result reminder scheduled. Approval rule: Drug interaction, suspected contraindication or extra test need → physician approval. Healthcare ### Patient journey coordinator: appointment, preparation, approval and results [Patient journey coordinator: appointment, preparation, approval and results] run #41,289,683elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Drug interaction, suspected contraindication or extra test need → physician approval Agents in this run Patient Journey Supervisorsupervisorplans, routes, approves Dialogue Agentaginies-talay-2 · on-premTalks with the patient, verifies identity, collects preferences Scheduling Agentaginies-sunkar-1 · on-premMatches device, physician and patient availability, creates the booking Clinical Preparation Agentaginies-tulpar-4 · on-premProduces preparation instructions from the protocol, checks medication and contraindications Insurer Agentaginies-sunkar-1 · on-premPrepares the pre-approval request and tracks the result Audit trail00:00.00 - ◦WhatsApp: "the doctor ordered an MRI, as soon as possible, mornings, I have private insurance" - ✓Patient verified by ID and birth date; contrast lumbar MRI order found in the HIS - ⌘Plan: clinical check → pre-approval → eligible slot → booking → preparation → reminders - ✓Medication list: metformin → 48-hour pause per contrast protocol; no kidney value within 3 months - ⏸Physician approval requested for a creatinine order; physician added it via mobile approval - ⇢Pre-approval submitted: ICD and procedure code, physician report attached - ⚙Insurer reply: approved, 10% co-payment, valid 30 days - ⇢MRI Thursday 08:40, blood draw Wednesday 08:00; both bookings created in the HIS - ➤Preparation: metformin pause, fasting, metal objects; co-payment and address sent - ⇢Reminders scheduled: 24 hours before, physician follow-up suggestion when results land - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Appointment request over WhatsApp or the call centre after a physician order Outcome Appointment booked, preparation sent, pre-approval obtained, result reminder scheduled ## Agents SUPERVISOR ### Patient Journey Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Dialogue Agent Talks with the patient, verifies identity, collects preferences verify_patientdetect_intentscompose_reply aginies-sunkar-1 · on-prem ### Scheduling Agent Matches device, physician and patient availability, creates the booking find_slotsbook_slotreschedule aginies-tulpar-4 · on-prem ### Clinical Preparation Agent Produces preparation instructions from the protocol, checks medication and contraindications get_protocolcheck_medicationsflag_physician aginies-sunkar-1 · on-prem ### Insurer Agent Prepares the pre-approval request and tracks the result prepare_preauthsubmit_preauthpoll_status ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Patient Journey Supervisor | WhatsApp: "the doctor ordered an MRI, as soon as possible, mornings, I have private insurance" | WhatsApp Business | 180 | — | | 02 | verify | Dialogue Agent | Patient verified by ID and birth date; contrast lumbar MRI order found in the HIS | Hospital Information System | 520 | 180 | | 03 | plan | Patient Journey Supervisor | Plan: clinical check → pre-approval → eligible slot → booking → preparation → reminders | — | 240 | 220 | | 04 | verify | Clinical Preparation Agent | Medication list: metformin → 48-hour pause per contrast protocol; no kidney value within 3 months | Hospital Information System | 1,100 | 620 | | 05 | approval | Clinical Preparation Agent | Physician approval requested for a creatinine order; physician added it via mobile approval | Hospital Information System | 900 | — | | 06 | write | Insurer Agent | Pre-approval submitted: ICD and procedure code, physician report attached | Insurer Portal | 700 | — | | 07 | tool | Insurer Agent | Insurer reply: approved, 10% co-payment, valid 30 days | Insurer Portal | 2,200 | — | | 08 | write | Scheduling Agent | MRI Thursday 08:40, blood draw Wednesday 08:00; both bookings created in the HIS | Scheduling | 640 | — | | 09 | notify | Dialogue Agent | Preparation: metformin pause, fasting, metal objects; co-payment and address sent | WhatsApp Business | 520 | 320 | | 10 | write | Scheduling Agent | Reminders scheduled: 24 hours before, physician follow-up suggestion when results land | Lab / Imaging | 300 | — | Simulation metrics (estimated) 58,000runs / month 95.1%success 7.3 sp50 $0.0026est. cost / run Approval rule Drug interaction, suspected contraindication or extra test need → physician approval Step profile ingest · 1plan · 1tool · 1verify · 2approval · 1write · 3notify · 1 SYSTEMS - WhatsApp Businesschannel - Hospital Information Systemcore - Schedulingapi - Insurer Portalapi - Lab / Imagingapi What you need to run it - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - Read/write API or MCP server and a service account for the core system - Credentials and a test environment for the enterprise APIs involved - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Large hospital groups' patient access centre pattern, combined with clinical protocol checks Packages that include this scenario Healthcare operations & clinical guidance → Sector: Healthcare → Related agents HealthcareHITL ### Prior authorization Payer rules searched, clinical evidence gathered and the packet ready to submit within minutes of the order. 6,200 runs / month 94.1% success 4 agents HealthcareHITL ### Patient intake and triage The patient types or talks; intake and urgency scoring take seconds, and a nurse always makes the final call. 24,500 runs / month 95.8% success 5 agents Healthcare ### Appointment scheduling and no-show reduction High no-show-risk patients are reached on WhatsApp in advance; freed slots are refilled from the waiting list. 62,000 runs / month 98.2% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Payment reconciliation exceptions URL: https://www.aginies.com/en/agents/payment-reconciliation-exceptions Language: en Summary: PSP settlement files matched to the ledger; small breaks fix themselves, large ones go to finance approval. Nightly batch: PSP/acquirer T+1 settlement file received → All exceptions matched or corrected, journals posted, open items with finance Fintech & Paymentsbatchapiemail # Payment reconciliation exceptions PSP settlement files matched to the ledger; small breaks fix themselves, large ones go to finance approval. In short Payment reconciliation exceptions is an agent scenario in the Fintech & Payments sector running on the batch, api, email channels. Trigger: Nightly batch: PSP/acquirer T+1 settlement file received. The Reconciliation Supervisor coordinates 3 specialist agents (Matching Agent, Investigation Agent, Correction Agent) across 5 systems (PSP / Acquirer API, Card Network, Core Ledger, Case Management, Email); of 10 steps, 2 write to systems and 2 are checks or approvals. Outcome: All exceptions matched or corrected, journals posted, open items with finance. Approval rule: Single break > 10,000 or total > 50,000 → finance manager approval before adjustment. Fintech & Payments ### Payment reconciliation exceptions [Payment reconciliation exceptions] run #41,289,704elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Single break > 10,000 or total > 50,000 → finance manager approval before adjustment Agents in this run Reconciliation Supervisorsupervisorplans, routes, approves Matching Agentaginies-talay-2 · on-premMatches settlement lines to ledger entries, isolates exceptions Investigation Agentaginies-tulpar-4 · on-premDetermines the root cause: fees, FX, refunds, timing Correction Agentaginies-sunkar-1 · on-premPosts in-tolerance correcting journals, routes the rest to finance Audit trail00:00.00 - ◦Settlement file received: 184,320 transactions, 41,870,500, T+1 - ⚙PSP detail pulled; 99.86% matched, 258 breaks: 143 fees, 61 refunds, 39 FX, 15 missing - ⌘Plan: root-cause classification → tolerance check → auto-fix → finance approval - ⚙Card network fee schedule pulled: 143 lines explained by a scheme fee change - ⌕Ledger entries retrieved: 61 refunds booked on T+2 → timing difference - ✓Tolerance: 243 exceptions ≤ 500 → auto-fix; 15 missing transactions 186,400 → finance - ⇢243 correcting journals posted (fees and timing), total 21,340 - ⏸Break > 10,000 → finance manager approved the adjustment for 15 missing transactions - ⇢Case #REC-1188 opened, missing-transaction query sent to the PSP - ➤Summary email to finance: 243 auto-corrected, 15 items in an open case - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Nightly batch: PSP/acquirer T+1 settlement file received Outcome All exceptions matched or corrected, journals posted, open items with finance ## Agents SUPERVISOR ### Reconciliation Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Matching Agent Matches settlement lines to ledger entries, isolates exceptions parse_settlement_filematch_transactionslist_exceptions aginies-tulpar-4 · on-prem ### Investigation Agent Determines the root cause: fees, FX, refunds, timing get_psp_transactionget_ledger_entryget_card_network_feesclassify_break aginies-sunkar-1 · on-prem ### Correction Agent Posts in-tolerance correcting journals, routes the rest to finance post_journalcreate_exception_casesend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Reconciliation Supervisor | Settlement file received: 184,320 transactions, 41,870,500, T+1 | — | 300 | — | | 02 | tool | Matching Agent | PSP detail pulled; 99.86% matched, 258 breaks: 143 fees, 61 refunds, 39 FX, 15 missing | PSP / Acquirer API | 1,400 | — | | 03 | plan | Reconciliation Supervisor | Plan: root-cause classification → tolerance check → auto-fix → finance approval | — | 280 | 200 | | 04 | tool | Investigation Agent | Card network fee schedule pulled: 143 lines explained by a scheme fee change | Card Network | 540 | — | | 05 | retrieve | Investigation Agent | Ledger entries retrieved: 61 refunds booked on T+2 → timing difference | Core Ledger | 480 | — | | 06 | verify | Investigation Agent | Tolerance: 243 exceptions ≤ 500 → auto-fix; 15 missing transactions 186,400 → finance | — | 760 | 820 | | 07 | write | Correction Agent | 243 correcting journals posted (fees and timing), total 21,340 | Core Ledger | 680 | — | | 08 | approval | Reconciliation Supervisor | Break > 10,000 → finance manager approved the adjustment for 15 missing transactions | — | 3,200 | — | | 09 | write | Correction Agent | Case #REC-1188 opened, missing-transaction query sent to the PSP | Case Management | 440 | — | | 10 | notify | Correction Agent | Summary email to finance: 243 auto-corrected, 15 items in an open case | Email | 260 | — | Simulation metrics (estimated) 620runs / month 95.1%success 8.1 sp50 $0.0133est. cost / run Approval rule Single break > 10,000 or total > 50,000 → finance manager approval before adjustment Step profile ingest · 1plan · 1retrieve · 1tool · 2verify · 1approval · 1write · 2notify · 1 SYSTEMS - PSP / Acquirer APIpayments - Card Networkapi - Core Ledgercore - Case Managementticketing - Emailemail What you need to run it - Authorised API access and limits for payment services - Credentials and a test environment for the enterprise APIs involved - Read/write API or MCP server and a service account for the core system - API access and assignment rules for the ticketing or work-order system - Mailbox access (Microsoft 365 or Google Workspace) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors multi-PSP automated reconciliation and exception management at large payment companies Packages that include this scenario Financial analysis & operational intelligence →Merchant and payment operations → Sector: Fintech & Payments → Related agents Fintech & PaymentsHITL ### Merchant operations assistant: POS, settlement, refund and limit The merchant writes on WhatsApp; POS fault, settlement query, refund and limit increase are resolved in one dialogue. 64,000 runs / month 93.8% success 5 agents Fintech & PaymentsHITL ### Merchant KYB onboarding Registry, UBO and sanctions checks in one flow; low-risk merchants start accepting payments the same day. 1,400 runs / month 93.4% success 5 agents Fintech & PaymentsHITL ### AML transaction monitoring alert triage Alerts arrive enriched with a draft narrative; the analyst always decides and the filing goes out in minutes. 8,500 runs / month 95.8% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Permit application check URL: https://www.aginies.com/en/agents/permit-application-check Language: en Summary: Missing documents and regulatory fit reported in minutes; the licensing officer always makes the decision. Workplace opening permit application via e-Government → Officer decided with a pre-review summary, applicant sent the deficiency list and deadline Government & Public Sectorwebemail # Permit application check Missing documents and regulatory fit reported in minutes; the licensing officer always makes the decision. In short Permit application check is an agent scenario in the Government & Public Sector sector running on the web, email channels. Trigger: Workplace opening permit application via e-Government. The Permit Supervisor coordinates 3 specialist agents (Document Agent, Regulation Agent, File Agent) across 5 systems (e-Devlet Gateway, Case Management, GIS / Zoning Plan, Regulation Vector Store, Exchange / M365 Mail); of 9 steps, 1 write to systems and 2 are checks or approvals. Outcome: Officer decided with a pre-review summary, applicant sent the deficiency list and deadline. Approval rule: Every permit decision is taken by the licensing officer; the system only pre-reviews. Government & Public Sector ### Permit application check [Permit application check] run #41,289,846elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Every permit decision is taken by the licensing officer; the system only pre-reviews Agents in this run Permit Supervisorsupervisorplans, routes, approves Document Agentaginies-konrul-1 · on-premChecks attachment type, signature and completeness Regulation Agentaginies-tulpar-4 · on-premAssesses fit against zoning and licensing regulation File Agentaginies-talay-2 · on-premPrepares the officer summary and updates the case Audit trail00:00.00 - ◦Workplace opening permit application BR-2026-7731 received with 9 attachments - ∿9 of 11 required documents present; missing: fire report; lease copy unsigned - ⚙Parcel 412/7: zoning "commercial", ground floor, 96 m² - ⌕Workplace licensing regulation arts. 5, 6 and annex 1 retrieved - ∿Café class 2: distance to school 210 m (> 100 m OK), fee 4,750 TL, 2 conditions - ✓Result: incomplete (2 items); automatic rejection not allowed by policy - ⇢Case updated with the checklist and a pre-review summary for the officer - ⏸Licensing officer reviewed and decided: "deficiency letter" (mandatory for every permit) - ➤Applicant e-mailed the deficiency list, 15-day completion period started - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Workplace opening permit application via e-Government Outcome Officer decided with a pre-review summary, applicant sent the deficiency list and deadline ## Agents SUPERVISOR ### Permit Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Document Agent Checks attachment type, signature and completeness extract_document_typecheck_signatureread_form_fields aginies-tulpar-4 · on-prem ### Regulation Agent Assesses fit against zoning and licensing regulation get_parcel_infosearch_regulationscheck_zoning_rulescompute_fee aginies-talay-2 · on-prem ### File Agent Prepares the officer summary and updates the case build_review_summaryupdate_casesend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Permit Supervisor | Workplace opening permit application BR-2026-7731 received with 9 attachments | — | 200 | — | | 02 | reason | Document Agent | 9 of 11 required documents present; missing: fire report; lease copy unsigned | — | 940 | 880 | | 03 | tool | Regulation Agent | Parcel 412/7: zoning "commercial", ground floor, 96 m² | GIS / Zoning Plan | 420 | — | | 04 | retrieve | Regulation Agent | Workplace licensing regulation arts. 5, 6 and annex 1 retrieved | Regulation Vector Store | 460 | 620 | | 05 | reason | Regulation Agent | Café class 2: distance to school 210 m (> 100 m OK), fee 4,750 TL, 2 conditions | — | 760 | 1,040 | | 06 | verify | File Agent | Result: incomplete (2 items); automatic rejection not allowed by policy | — | 320 | 240 | | 07 | write | File Agent | Case updated with the checklist and a pre-review summary for the officer | Case Management | 480 | — | | 08 | approval | Permit Supervisor | Licensing officer reviewed and decided: "deficiency letter" (mandatory for every permit) | — | 3,200 | — | | 09 | notify | File Agent | Applicant e-mailed the deficiency list, 15-day completion period started | Exchange / M365 Mail | 260 | — | Simulation metrics (estimated) 3,100runs / month 95.8%success 7 sp50 $0.0079est. cost / run Approval rule Every permit decision is taken by the licensing officer; the system only pre-reviews Step profile ingest · 1retrieve · 1reason · 2tool · 1verify · 1approval · 1write · 1notify · 1 SYSTEMS - e-Devlet Gatewayapi - Case Managementticketing - GIS / Zoning Plandb - Regulation Vector Storedb - Exchange / M365 Mailemail What you need to run it - Credentials and a test environment for the enterprise APIs involved - API access and assignment rules for the ticketing or work-order system - Read access to the database or vector store; a data owner - Mailbox access (Microsoft 365 or Google Workspace) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors permit pre-screening pilots in Singapore and Nordic municipalities Packages that include this scenario Information management & decision support →Citizen services & file pre-review → Sector: Government & Public Sector → Related agents Government & Public SectorHITL ### Citizen services assistant: application, documents, appointment and status The citizen writes or calls; eligibility checked, documents read, application opened, appointment given, every step traceable. 230,000 runs / month 93.8% success 5 agents Government & Public SectorHITL ### Citizen request routing e-Government requests classified in seconds, given a legal deadline and routed to the right unit with an SLA. 54,000 runs / month 98.1% success 4 agents Government & Public SectorHITL ### Public procurement bid file pre-check Bid files are compared with the specification, missing and non-conforming documents listed; the evaluation decision belongs to the commission. 380 runs / month 94.8% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Plan upgrade advisor URL: https://www.aginies.com/en/agents/plan-upgrade-advisor Language: en Summary: Usage analysed, the right plan offered with a comparison; on acceptance the change provisions instantly. "I keep running out of data" message or add-ons in 3 consecutive months → New plan provisioned, consent time-stamped, customer knows the effective date Telecomappwhatsappvoice # Plan upgrade advisor Usage analysed, the right plan offered with a comparison; on acceptance the change provisions instantly. In short Plan upgrade advisor is an agent scenario in the Telecom sector running on the app, whatsapp, voice channels. Trigger: "I keep running out of data" message or add-ons in 3 consecutive months. The Plan Supervisor coordinates 3 specialist agents (Usage Agent, Offer Agent, Provisioning Agent) across 5 systems (Billing (Amdocs), Product Catalogue / Offer Engine, HLR / Provisioning, WhatsApp Business API, SMS Gateway); of 10 steps, 1 write to systems and 1 are checks or approvals. Outcome: New plan provisioned, consent time-stamped, customer knows the effective date. Approval rule: Commitment-based or discounted bespoke offer → campaign manager approval. Telecom ### Plan upgrade advisor [Plan upgrade advisor] run #41,289,464elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Commitment-based or discounted bespoke offer → campaign manager approval Agents in this run Plan Supervisorsupervisorplans, routes, approves Usage Agentaginies-talay-2 · on-premPulls usage history and forecasts the coming period Offer Agentaginies-tulpar-4 · on-premCompares eligible plans, presents the offer, interprets the reply Provisioning Agentaginies-sunkar-1 · on-premRecords consent, changes the plan, informs the customer Audit trail00:00.00 - ◦WhatsApp: "I run out of data every month, what should I do?" - ⚙Last 6 months: 28 GB average usage, add-ons bought in 3 months (210 in total) - ∿Current plan 20 GB / 349; forecast rising to 32 GB → upgrade fits - ⌘Plan: eligible plans → comparison → guardrail → offer → consent → provisioning - ⚙Eligible plans: 40 GB / 419, 60 GB / 479; no commitment - ∿Recommendation: 40 GB / 419 → saves about 70 per month versus add-ons - ✓Guardrail: offer complies with campaign rules; price and commitment terms correct - ➤WhatsApp: comparison card with an "I accept" button; customer accepted - ⇢Consent recorded; plan change provisioned effective from the next billing cycle - ➤SMS: new plan summary and effective date - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger "I keep running out of data" message or add-ons in 3 consecutive months Outcome New plan provisioned, consent time-stamped, customer knows the effective date ## Agents SUPERVISOR ### Plan Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Usage Agent Pulls usage history and forecasts the coming period get_usage_historyget_current_planforecast_usage aginies-tulpar-4 · on-prem ### Offer Agent Compares eligible plans, presents the offer, interprets the reply get_eligible_planscompare_planscheck_campaign_rulessend_whatsappparse_reply aginies-sunkar-1 · on-prem ### Provisioning Agent Records consent, changes the plan, informs the customer record_consentchange_plansend_sms ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Plan Supervisor | WhatsApp: "I run out of data every month, what should I do?" | — | 180 | — | | 02 | tool | Usage Agent | Last 6 months: 28 GB average usage, add-ons bought in 3 months (210 in total) | Billing (Amdocs) | 460 | — | | 03 | reason | Usage Agent | Current plan 20 GB / 349; forecast rising to 32 GB → upgrade fits | — | 620 | 560 | | 04 | plan | Plan Supervisor | Plan: eligible plans → comparison → guardrail → offer → consent → provisioning | — | 240 | 180 | | 05 | tool | Offer Agent | Eligible plans: 40 GB / 419, 60 GB / 479; no commitment | Product Catalogue / Offer Engine | 380 | — | | 06 | reason | Offer Agent | Recommendation: 40 GB / 419 → saves about 70 per month versus add-ons | — | 680 | 640 | | 07 | verify | Offer Agent | Guardrail: offer complies with campaign rules; price and commitment terms correct | — | 360 | 280 | | 08 | notify | Offer Agent | WhatsApp: comparison card with an "I accept" button; customer accepted | WhatsApp Business API | 1,400 | 300 | | 09 | write | Provisioning Agent | Consent recorded; plan change provisioned effective from the next billing cycle | HLR / Provisioning | 520 | — | | 10 | notify | Provisioning Agent | SMS: new plan summary and effective date | SMS Gateway | 220 | — | Simulation metrics (estimated) 52,000runs / month 95.8%success 5.1 sp50 $0.001est. cost / run Approval rule Commitment-based or discounted bespoke offer → campaign manager approval Step profile ingest · 1plan · 1reason · 2tool · 2verify · 1write · 1notify · 2 SYSTEMS - Billing (Amdocs)core - Product Catalogue / Offer Engineapi - HLR / Provisioningcore - WhatsApp Business APIchannel - SMS Gatewaychannel What you need to run it - Read/write API or MCP server and a service account for the core system - Credentials and a test environment for the enterprise APIs involved - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the usage-driven "right plan" recommendation pattern at operators, extended to provision the change Packages that include this scenario Sales & lead management →Subscriber care and network support → Sector: Telecom → Related agents TelecomHITL ### Subscriber care assistant: bill, plan, fault and contract The subscriber calls or writes; bill dispute, plan change and line fault are resolved in BSS and network systems in one session. 410,000 runs / month 94.3% success 5 agents TelecomHITL ### Bill dispute resolution Invoice lines and usage explained in seconds; goodwill credit applied by policy, larger amounts need a lead. 88,000 runs / month 96.3% success 4 agents TelecomHITL ### SIM swap fraud check Every SIM swap is risk-scored before provisioning; suspicious requests are held and banks are signalled. 41,000 runs / month 97.7% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Platform maintenance (MRO) work order preparation URL: https://www.aginies.com/en/agents/platform-mro-work-order Language: en Summary: The fault report is matched to technical publications, task card and parts list are produced, the work order opens with planner approval. Post-flight fault entry or a scheduled-maintenance trigger from the MMS → Work order opened with the right task cards and reserved parts; assigned to the technician crew Defence & Aerospaceappapibatch # Platform maintenance (MRO) work order preparation The fault report is matched to technical publications, task card and parts list are produced, the work order opens with planner approval. In short Platform maintenance (MRO) work order preparation is an agent scenario in the Defence & Aerospace sector running on the app, api, batch channels. Trigger: Post-flight fault entry or a scheduled-maintenance trigger from the MMS. The Maintenance Supervisor coordinates 4 specialist agents (Fault Agent, Tech Pub Agent, Parts Agent, Planning Agent) across 4 systems (Maintenance Management System, Technical Publications (S1000D vector), Spares Inventory, Technician App); of 9 steps, 1 write to systems and 2 are checks or approvals. Outcome: Work order opened with the right task cards and reserved parts; assigned to the technician crew. Approval rule: Every work package affecting airworthiness → maintenance planner approval. Defence & Aerospace ### Platform maintenance (MRO) work order preparation [Platform maintenance (MRO) work order preparation] run #41,288,235elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Every work package affecting airworthiness → maintenance planner approval Agents in this run Maintenance Supervisorsupervisorplans, routes, approves Fault Agentaginies-talay-2 · on-premInterprets the fault entry and history Tech Pub Agentaginies-tulpar-4 · on-premFinds task cards in the maintenance manuals Parts Agentaginies-sunkar-1 · on-premChecks parts, tooling and stock availability Planning Agentaginies-talay-2 · on-premOpens the work order and assigns the crew Audit trail00:00.00 - ◦Fault entry F-30917: tail number 07, “hydraulic pressure warning, system 2” - ⚙Component history pulled: pump P2 had 2 similar warnings in the last 400 hours - ∿Fault class: hydraulic pump degradation; likely cause internal leakage; airworthiness priority - ⌕Task cards found: 29-11-05 pump test, 29-11-07 replacement; revision current - ⚙IPC parts list: pump P/N 4211-88, 2 seals; stock has 1 pump, 6 seals - ✓Tool list and calibration dates valid; estimated 6.5 man-hours - ⏸Airworthiness impact → maintenance planner approved the work package - ⇢Work order WO-71822 opened, parts reserved, crew assigned to shift B - ➤Task cards, parts list and safety notes pushed to the technician app - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Post-flight fault entry or a scheduled-maintenance trigger from the MMS Outcome Work order opened with the right task cards and reserved parts; assigned to the technician crew ## Agents SUPERVISOR ### Maintenance Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Fault Agent Interprets the fault entry and history get_fault_recordget_component_historyclassify_fault aginies-tulpar-4 · on-prem ### Tech Pub Agent Finds task cards in the maintenance manuals search_tech_pubsget_task_cardcheck_revision aginies-sunkar-1 · on-prem ### Parts Agent Checks parts, tooling and stock availability get_ipc_partscheck_stockreserve_parts aginies-talay-2 · on-prem ### Planning Agent Opens the work order and assigns the crew create_work_orderassign_crewnotify_crew ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Maintenance Supervisor | Fault entry F-30917: tail number 07, “hydraulic pressure warning, system 2” | — | 200 | — | | 02 | tool | Fault Agent | Component history pulled: pump P2 had 2 similar warnings in the last 400 hours | Maintenance Management System | 440 | — | | 03 | reason | Fault Agent | Fault class: hydraulic pump degradation; likely cause internal leakage; airworthiness priority | — | 780 | 980 | | 04 | retrieve | Tech Pub Agent | Task cards found: 29-11-05 pump test, 29-11-07 replacement; revision current | Technical Publications (S1000D vector) | 620 | 760 | | 05 | tool | Parts Agent | IPC parts list: pump P/N 4211-88, 2 seals; stock has 1 pump, 6 seals | Spares Inventory | 420 | — | | 06 | verify | Parts Agent | Tool list and calibration dates valid; estimated 6.5 man-hours | — | 360 | 240 | | 07 | approval | Maintenance Supervisor | Airworthiness impact → maintenance planner approved the work package | — | 2,800 | — | | 08 | write | Planning Agent | Work order WO-71822 opened, parts reserved, crew assigned to shift B | Maintenance Management System | 520 | — | | 09 | notify | Planning Agent | Task cards, parts list and safety notes pushed to the technician app | Technician App | 260 | — | Simulation metrics (estimated) 4,100runs / month 95.7%success 6.4 sp50 $0.007est. cost / run Approval rule Every work package affecting airworthiness → maintenance planner approval Step profile ingest · 1retrieve · 1reason · 1tool · 2verify · 1approval · 1write · 1notify · 1 SYSTEMS - Maintenance Management Systemerp - Technical Publications (S1000D vector)db - Spares Inventoryerp - Technician Appchannel What you need to run it - Posting and query API access to ERP/MES - Read access to the database or vector store; a data owner - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors S1000D-based task card and work order automation in aerospace and defence MRO organisations Packages that include this scenario Field service & dispatch automation →Defence export control & supply-chain security → Sector: Defence & Aerospace → Related agents Defence & AerospaceHITL ### Technical publication assistant: maintenance manual, parts, procedure and revision control The technician asks in the hangar; the right-revision procedure, part number and safety note arrive with clearance checks, every access traced. 9,800 runs / month 95.7% success 5 agents Defence & AerospaceHITL ### Export control and end-user screening Part, buyer and end use are classified in one flow; licence need is detected, the decision always stays with the compliance officer. 2,600 runs / month 96.1% success 4 agents Defence & AerospaceHITL ### Supplier security and compliance assessment Supplier questionnaires and documents are read, compared against facility security requirements, gaps listed; the qualification decision belongs to the security manager. 900 runs / month 95.2% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Policy renewal offer URL: https://www.aginies.com/en/agents/policy-renewal Language: en Summary: Policies re-priced, sharp increases go to approval; customer accepts with one tap and the policy is issued. Batch: policies expiring within 30 days → New policy issued, consent time-stamped, PDF and payment link with the customer Insurancebatchwhatsappemailapp # Policy renewal offer Policies re-priced, sharp increases go to approval; customer accepts with one tap and the policy is issued. In short Policy renewal offer is an agent scenario in the Insurance sector running on the batch, whatsapp, email, app channels. Trigger: Batch: policies expiring within 30 days. The Renewal Supervisor coordinates 3 specialist agents (Pricing Agent, Offer Agent, Issuance Agent) across 4 systems (Policy Admin System, Rating Engine, WhatsApp Business API, Email (Exchange)); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: New policy issued, consent time-stamped, PDF and payment link with the customer. Approval rule: Premium increase > 20% or discount > 10% → renewal team lead approval. Insurance ### Policy renewal offer [Policy renewal offer] run #41,290,352elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Premium increase > 20% or discount > 10% → renewal team lead approval Agents in this run Renewal Supervisorsupervisorplans, routes, approves Pricing Agentaginies-talay-2 · on-premComputes the new premium from claim history and the rating engine Offer Agentaginies-tulpar-4 · on-premDrafts and sends the offer, interprets reply and consent Issuance Agentaginies-sunkar-1 · on-premRecords consent, issues the policy, creates the payment link Audit trail00:00.00 - ◦Batch: 8,120 motor policies expiring within 30 days listed - ⚙Policy #KS-1187720: 2 claim-free years, 2021 vehicle, no-claims discount 30% - ⚙Rating engine: new premium 19,700 (last year 15,900, +23.9%) - ⌘Plan: increase check → approval → offer → consent → issue - ✓Increase 23.9% > 20% threshold → loyalty discount of 8% proposed, approval needed - ⏸Increase > 20% → renewal team lead approved the 8% discount → 18,124 - ➤Offer summary with an "I accept" button sent on WhatsApp - ∿Customer accepted; explicit consent and data-protection notice acknowledged - ⇢Consent stored with timestamp, policy #KS-1187720/2 issued - ➤Policy PDF and payment link emailed - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Batch: policies expiring within 30 days Outcome New policy issued, consent time-stamped, PDF and payment link with the customer ## Agents SUPERVISOR ### Renewal Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Pricing Agent Computes the new premium from claim history and the rating engine get_policyget_claim_historycalculate_premiumapply_discounts aginies-tulpar-4 · on-prem ### Offer Agent Drafts and sends the offer, interprets reply and consent draft_offersend_whatsappsend_emailparse_reply aginies-sunkar-1 · on-prem ### Issuance Agent Records consent, issues the policy, creates the payment link record_consentissue_policycreate_payment_link ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Renewal Supervisor | Batch: 8,120 motor policies expiring within 30 days listed | — | 320 | — | | 02 | tool | Pricing Agent | Policy #KS-1187720: 2 claim-free years, 2021 vehicle, no-claims discount 30% | Policy Admin System | 410 | — | | 03 | tool | Pricing Agent | Rating engine: new premium 19,700 (last year 15,900, +23.9%) | Rating Engine | 520 | — | | 04 | plan | Renewal Supervisor | Plan: increase check → approval → offer → consent → issue | — | 260 | 180 | | 05 | verify | Pricing Agent | Increase 23.9% > 20% threshold → loyalty discount of 8% proposed, approval needed | — | 580 | 520 | | 06 | approval | Renewal Supervisor | Increase > 20% → renewal team lead approved the 8% discount → 18,124 | — | 2,600 | — | | 07 | notify | Offer Agent | Offer summary with an "I accept" button sent on WhatsApp | WhatsApp Business API | 260 | 340 | | 08 | reason | Offer Agent | Customer accepted; explicit consent and data-protection notice acknowledged | — | 420 | 210 | | 09 | write | Issuance Agent | Consent stored with timestamp, policy #KS-1187720/2 issued | Policy Admin System | 540 | — | | 10 | notify | Issuance Agent | Policy PDF and payment link emailed | Email (Exchange) | 240 | — | Simulation metrics (estimated) 22,000runs / month 94.4%success 6.2 sp50 $0.0015est. cost / run Approval rule Premium increase > 20% or discount > 10% → renewal team lead approval Step profile ingest · 1plan · 1reason · 1tool · 2verify · 1approval · 1write · 1notify · 2 SYSTEMS - Policy Admin Systemcore - Rating Engineapi - WhatsApp Business APIchannel - Email (Exchange)email What you need to run it - Read/write API or MCP server and a service account for the core system - Credentials and a test environment for the enterprise APIs involved - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - Mailbox access (Microsoft 365 or Google Workspace) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors EU insurers' automated renewal flows built around price-transparency rules Packages that include this scenario Underwriting and renewal operations → Sector: Insurance → Related agents InsuranceHITL ### End-to-end motor claim: from notice to payment Voice notice, photo assessment, repair-shop assignment and payment order in one flow; the adjuster steps in only above the threshold. 21,000 runs / month 91.2% success 5 agents InsuranceHITL ### Voice claims intake (FNOL) The customer calls and describes it; policy found by plate, cover checked, claim opened and adjuster assigned. 34,000 runs / month 95.8% success 5 agents InsuranceHITL ### Underwriting document extraction Medical reports and labs read and assessed against the manual; underwriters only see files above threshold. 4,800 runs / month 94.9% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Prior authorization URL: https://www.aginies.com/en/agents/prior-authorization Language: en Summary: Payer rules searched, clinical evidence gathered and the packet ready to submit within minutes of the order. A procedure order requiring prior authorisation is placed in the EHR → Complete packet submitted to the payer, auth number written to the EHR, team informed Healthcareapiemailweb # Prior authorization Payer rules searched, clinical evidence gathered and the packet ready to submit within minutes of the order. In short Prior authorization is an agent scenario in the Healthcare sector running on the api, email, web channels. Trigger: A procedure order requiring prior authorisation is placed in the EHR. The Prior Auth Supervisor coordinates 3 specialist agents (Order Agent, Payer Rules Agent, Submission Agent) across 4 systems (EHR (HL7/FHIR), Payer Policy Vector Store, Payer Portal, Exchange / M365 Mail); of 10 steps, 2 write to systems and 2 are checks or approvals. Outcome: Complete packet submitted to the payer, auth number written to the EHR, team informed. Approval rule: Surgical procedure or cost > 25,000 TL → utilisation management nurse approval. Healthcare ### Prior authorization [Prior authorization] run #41,286,623elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Surgical procedure or cost > 25,000 TL → utilisation management nurse approval Agents in this run Prior Auth Supervisorsupervisorplans, routes, approves Order Agentaginies-konrul-1 · on-premExtracts fields from the order and clinical documents Payer Rules Agentaginies-tulpar-4 · on-premSearches payer policy and assesses medical necessity Submission Agentaginies-talay-2 · on-premAssembles the packet, submits it and updates the record Audit trail00:00.00 - ◦Order OR-58213 for lumbar fusion received from the EHR, patient identity matched - ∿CPT 22612 · ICD-10 M43.16 · Physician: Dr Yılmaz · Payer: Plan A · Cost 185,000 TL - ⌘Plan: clinical notes → payer rule → necessity check → submit - ⌕Last 6 months of notes, MRI report and 8 weeks of physiotherapy records pulled - ⌕Payer policy SP-L-2026 §4.2 found: ≥ 6 weeks of conservative treatment required - ✓Necessity: 8 weeks PT documented, neurological deficit present → criteria met (0.93) - ⏸Surgical procedure → utilisation management nurse reviewed and approved the packet - ⇢Packet (12 pages) submitted to the payer portal, reference PA-2026-118402 received - ⇢Authorisation number and 90-day validity written to the encounter - ➤Physician office and surgical scheduling informed, decision trail stored - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger A procedure order requiring prior authorisation is placed in the EHR Outcome Complete packet submitted to the payer, auth number written to the EHR, team informed ## Agents SUPERVISOR ### Prior Auth Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Order Agent Extracts fields from the order and clinical documents extract_order_fieldsextract_clinical_notes aginies-tulpar-4 · on-prem ### Payer Rules Agent Searches payer policy and assesses medical necessity search_payer_policycheck_medical_necessity aginies-talay-2 · on-prem ### Submission Agent Assembles the packet, submits it and updates the record build_auth_packetsubmit_prior_authupdate_encountersend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Prior Auth Supervisor | Order OR-58213 for lumbar fusion received from the EHR, patient identity matched | — | 200 | — | | 02 | reason | Order Agent | CPT 22612 · ICD-10 M43.16 · Physician: Dr Yılmaz · Payer: Plan A · Cost 185,000 TL | — | 720 | 640 | | 03 | plan | Prior Auth Supervisor | Plan: clinical notes → payer rule → necessity check → submit | — | 300 | 210 | | 04 | retrieve | Order Agent | Last 6 months of notes, MRI report and 8 weeks of physiotherapy records pulled | EHR (HL7/FHIR) | 460 | — | | 05 | retrieve | Payer Rules Agent | Payer policy SP-L-2026 §4.2 found: ≥ 6 weeks of conservative treatment required | Payer Policy Vector Store | 520 | 880 | | 06 | verify | Payer Rules Agent | Necessity: 8 weeks PT documented, neurological deficit present → criteria met (0.93) | — | 840 | 1,100 | | 07 | approval | Prior Auth Supervisor | Surgical procedure → utilisation management nurse reviewed and approved the packet | — | 2,600 | — | | 08 | write | Submission Agent | Packet (12 pages) submitted to the payer portal, reference PA-2026-118402 received | Payer Portal | 900 | — | | 09 | write | Submission Agent | Authorisation number and 90-day validity written to the encounter | EHR (HL7/FHIR) | 380 | — | | 10 | notify | Submission Agent | Physician office and surgical scheduling informed, decision trail stored | Exchange / M365 Mail | 240 | — | Simulation metrics (estimated) 6,200runs / month 94.1%success 7.2 sp50 $0.0111est. cost / run Approval rule Surgical procedure or cost > 25,000 TL → utilisation management nurse approval Step profile ingest · 1plan · 1retrieve · 2reason · 1verify · 1approval · 1write · 2notify · 1 SYSTEMS - EHR (HL7/FHIR)core - Payer Policy Vector Storedb - Payer Portalapi - Exchange / M365 Mailemail What you need to run it - Read/write API or MCP server and a service account for the core system - Read access to the database or vector store; a data owner - Credentials and a test environment for the enterprise APIs involved - Mailbox access (Microsoft 365 or Google Workspace) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the prior-auth automation programmes of US health systems and payers Packages that include this scenario Information management & decision support →Healthcare operations & clinical guidance → Sector: Healthcare → Related agents HealthcareHITL ### Patient journey coordinator: appointment, preparation, approval and results The patient writes or calls; appointment, preparation instructions, insurer pre-approval and result notice come together in one coordinator. 58,000 runs / month 95.1% success 5 agents HealthcareHITL ### Patient intake and triage The patient types or talks; intake and urgency scoring take seconds, and a nurse always makes the final call. 24,500 runs / month 95.8% success 5 agents Healthcare ### Appointment scheduling and no-show reduction High no-show-risk patients are reached on WhatsApp in advance; freed slots are refilled from the waiting list. 62,000 runs / month 98.2% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Procure-to-pay desk: request, quotes, contract and invoice in one place URL: https://www.aginies.com/en/agents/procure-to-pay-desk Language: en Summary: The request arrives by email; quotes compared, contract clauses checked, PO opened, invoice three-way matched and payment scheduled. Purchase request email or a request form opened in the ERP → PO opened, contract deviations flagged, invoice matched, payment scheduled at due date Enterprise Functionsemailapi # Procure-to-pay desk: request, quotes, contract and invoice in one place The request arrives by email; quotes compared, contract clauses checked, PO opened, invoice three-way matched and payment scheduled. In short Procure-to-pay desk: request, quotes, contract and invoice in one place is an agent scenario in the Enterprise Functions sector running on the email, api channels. Trigger: Purchase request email or a request form opened in the ERP. The Procurement Supervisor coordinates 4 specialist agents (Request Agent, Sourcing Agent, Contract Agent, Invoice Agent) across 5 systems (Procurement Inbox, ERP, Supplier Portal, Contract Repository, Payment Hub); of 10 steps, 3 write to systems and 4 are checks or approvals. Outcome: PO opened, contract deviations flagged, invoice matched, payment scheduled at due date. Approval rule: Amount > 250,000, single quote or non-framework supplier → procurement committee. Enterprise Functions ### Procure-to-pay desk: request, quotes, contract and invoice in one place [Procure-to-pay desk: request, quotes, contract and invoice in one place] run #41,290,058elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Amount > 250,000, single quote or non-framework supplier → procurement committee Agents in this run Procurement Supervisorsupervisorplans, routes, approves Request Agentaginies-sunkar-1 · on-premClassifies the request, finds the budget line and approval chain Sourcing Agentaginies-tulpar-4 · on-premCollects quotes, normalises them, compares on total cost of ownership Contract Agentaginies-talay-2 · on-premCompares quote terms with the company template and the existing framework contract Invoice Agentaginies-konrul-1 · on-premReads the invoice, matches it with PO and receipt, schedules payment Audit trail00:00.00 - ◦Email: "40 laptops, from the Q4 budget, within 3 weeks" + specification attached - ✓Category IT hardware, budget line IT-CAPEX-Q4 available, approval chain: IT manager + finance - ⌘Plan: quotes from 3 suppliers → normalise → contract check → approval → PO → invoice → payment - ⇢Quote request sent through the portal to 3 framework suppliers, deadline 3 business days - ∿Quotes: 1.84M / 1.91M / 1.79M; TCO ranking with warranty and delivery: B, A, C - ✓Supplier B: payment term 30 days (framework 45), penalty clause missing → 2 deviations flagged - ⏸Comparison and deviations presented to IT manager and finance; B approved at 45-day terms - ⇢PO #PO-88213 opened and sent to the supplier, delivery in 18 days - ✓Invoice read: 40 units, 1.79M; three-way match with PO and goods receipt complete - ⇢Payment scheduled for day 45, no early-payment discount; closure email to the requester - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Purchase request email or a request form opened in the ERP Outcome PO opened, contract deviations flagged, invoice matched, payment scheduled at due date ## Agents SUPERVISOR ### Procurement Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Request Agent Classifies the request, finds the budget line and approval chain classify_requestcheck_budgetroute_approval aginies-tulpar-4 · on-prem ### Sourcing Agent Collects quotes, normalises them, compares on total cost of ownership request_quotesnormalise_quotescompare_tco aginies-talay-2 · on-prem ### Contract Agent Compares quote terms with the company template and the existing framework contract get_frameworkcompare_clausesflag_deviation aginies-konrul-1 · on-prem ### Invoice Agent Reads the invoice, matches it with PO and receipt, schedules payment extract_invoicethree_way_matchschedule_payment ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Procurement Supervisor | Email: "40 laptops, from the Q4 budget, within 3 weeks" + specification attached | Procurement Inbox | 180 | — | | 02 | verify | Request Agent | Category IT hardware, budget line IT-CAPEX-Q4 available, approval chain: IT manager + finance | ERP | 600 | 260 | | 03 | plan | Procurement Supervisor | Plan: quotes from 3 suppliers → normalise → contract check → approval → PO → invoice → payment | — | 240 | 220 | | 04 | write | Sourcing Agent | Quote request sent through the portal to 3 framework suppliers, deadline 3 business days | Supplier Portal | 700 | — | | 05 | reason | Sourcing Agent | Quotes: 1.84M / 1.91M / 1.79M; TCO ranking with warranty and delivery: B, A, C | Supplier Portal | 2,400 | 1,400 | | 06 | verify | Contract Agent | Supplier B: payment term 30 days (framework 45), penalty clause missing → 2 deviations flagged | Contract Repository | 1,500 | 820 | | 07 | approval | Request Agent | Comparison and deviations presented to IT manager and finance; B approved at 45-day terms | ERP | 1,200 | — | | 08 | write | Sourcing Agent | PO #PO-88213 opened and sent to the supplier, delivery in 18 days | ERP | 560 | — | | 09 | verify | Invoice Agent | Invoice read: 40 units, 1.79M; three-way match with PO and goods receipt complete | ERP | 1,800 | 900 | | 10 | write | Invoice Agent | Payment scheduled for day 45, no early-payment discount; closure email to the requester | Payment Hub | 480 | — | Simulation metrics (estimated) 14,000runs / month 92.7%success 9.1 sp50 $0.0055est. cost / run Approval rule Amount > 250,000, single quote or non-framework supplier → procurement committee Step profile ingest · 1plan · 1reason · 1verify · 3approval · 1write · 3 SYSTEMS - Procurement Inboxemail - ERPerp - Supplier Portalapi - Contract Repositorystorage - Payment Hubpayments What you need to run it - Mailbox access (Microsoft 365 or Google Workspace) - Posting and query API access to ERP/MES - Credentials and a test environment for the enterprise APIs involved - Read access to the document store and the document types - Authorised API access and limits for payment services - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Large enterprises' source-to-pay automation pattern, with contract checks and three-way matching Packages that include this scenario Invoice & document extraction → Related agents Enterprise Functions ### Email classification and attachment check Every e-mail in the shared mailbox is classified, attachments checked against a list, gaps asked for at once. 86,000 runs / month 97.6% success 4 agents Enterprise FunctionsHITL ### Call centre quality monitoring Every call scored against the rubric, not a sample; compliance breaches and coaching flags ready by morning. 90,000 runs / month 98.8% success 4 agents Enterprise FunctionsHITL ### Contract clause extraction Clauses are extracted and compared with the legal playbook; deviations reach counsel with redline suggestions. 1,400 runs / month 95.1% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Product Q&A assistant URL: https://www.aginies.com/en/agents/product-qa-assistant Language: en Summary: Size, compatibility and spec questions answered from catalogue and return data; stock and price verified. Question typed into the chat widget on a product page → Sourced answer and the right size/product card in chat, one tap to add to basket Retail & E-commercewebappwhatsapp # Product Q&A assistant Size, compatibility and spec questions answered from catalogue and return data; stock and price verified. In short Product Q&A assistant is an agent scenario in the Retail & E-commerce sector running on the web, app, whatsapp channels. Trigger: Question typed into the chat widget on a product page. The Shopping Supervisor coordinates 3 specialist agents (Intent Agent, Catalogue Agent, Reply Agent) across 4 systems (Vector Store, PIM / Catalogue, Shopify / OMS, Web Chat Widget); of 9 steps, 0 write to systems and 1 are checks or approvals. Outcome: Sourced answer and the right size/product card in chat, one tap to add to basket. Retail & E-commerce ### Product Q&A assistant [Product Q&A assistant] run #41,287,428elapsed 0 mstokens 0est. cost $0.000status — Agents in this run Shopping Supervisorsupervisorplans, routes, approves Intent Agentaginies-sunkar-1 · on-premExtracts the question and constraints (height, weight, device model) Catalogue Agentaginies-tulpar-4 · on-premBuilds the recommendation from catalogue, size chart and return data Reply Agentaginies-sunkar-1 · on-premGrounds the answer in sources and sends it with a product card Audit trail00:00.00 - ◦Web chat: "Would size 42 of this coat fit me? 1.78 m, 80 kg" - ∿Intent: size fit · product SKU 88-4410 · height 178 · weight 80 - ⌘Plan: size chart → return data → stock → grounding check → answer - ⌕Size chart and fit note retrieved ("slim fit") - ⚙PIM: 80% wool, lined; return data: size 42 "too small" 31% - ∿Recommendation: size 44 (slim fit + return data), confidence 86% - ⚙Stock: size 44 available (7), Istanbul warehouse, next-day delivery - ✓Answer grounded in catalogue sources only; price and delivery claims verified - ➤Answer and size-44 product card sent to chat with an "add to basket" button - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Question typed into the chat widget on a product page Outcome Sourced answer and the right size/product card in chat, one tap to add to basket ## Agents SUPERVISOR ### Shopping Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Intent Agent Extracts the question and constraints (height, weight, device model) detect_intentextract_product_refextract_constraints aginies-tulpar-4 · on-prem ### Catalogue Agent Builds the recommendation from catalogue, size chart and return data search_catalogueget_product_specsget_size_chartcheck_compatibilityget_stock aginies-sunkar-1 · on-prem ### Reply Agent Grounds the answer in sources and sends it with a product card check_groundingcompose_answeradd_product_cardssend_reply ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Shopping Supervisor | Web chat: "Would size 42 of this coat fit me? 1.78 m, 80 kg" | — | 160 | — | | 02 | reason | Intent Agent | Intent: size fit · product SKU 88-4410 · height 178 · weight 80 | — | 380 | 240 | | 03 | plan | Shopping Supervisor | Plan: size chart → return data → stock → grounding check → answer | — | 220 | 160 | | 04 | retrieve | Catalogue Agent | Size chart and fit note retrieved ("slim fit") | Vector Store | 340 | — | | 05 | tool | Catalogue Agent | PIM: 80% wool, lined; return data: size 42 "too small" 31% | PIM / Catalogue | 420 | — | | 06 | reason | Catalogue Agent | Recommendation: size 44 (slim fit + return data), confidence 86% | — | 720 | 810 | | 07 | tool | Catalogue Agent | Stock: size 44 available (7), Istanbul warehouse, next-day delivery | Shopify / OMS | 300 | — | | 08 | verify | Reply Agent | Answer grounded in catalogue sources only; price and delivery claims verified | — | 360 | 280 | | 09 | notify | Reply Agent | Answer and size-44 product card sent to chat with an "add to basket" button | Web Chat Widget | 240 | 210 | Simulation metrics (estimated) 120,000runs / month 98.7%success 3.2 sp50 $0.0004est. cost / run Step profile ingest · 1plan · 1retrieve · 1reason · 2tool · 2verify · 1notify · 1 SYSTEMS - Vector Storedb - PIM / Cataloguedb - Shopify / OMSerp - Web Chat Widgetchannel What you need to run it - Read access to the database or vector store; a data owner - Posting and query API access to ERP/MES - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the size-recommendation pattern at large fashion retailers, grounded in return data Packages that include this scenario E-commerce support & action →Sales & lead management → Sector: Retail & E-commerce → Related agents Retail & E-commerceHITL ### Outfit stylist: a complete look by size, stock, budget and style The customer sends a photo and a sentence; the agent builds three looks that are in stock, fit and within budget, and adds the chosen one to the cart. 142,000 runs / month 96.1% success 5 agents Retail & E-commerce ### Where is my order? (WISMO) Customer asks, order and carrier events join instantly; on delay a new ETA is given and a trace opened. 110,000 runs / month 98.1% success 4 agents Retail & E-commerceHITL ### Returns and refunds Policy and history checked in seconds, label issued; the refund executes when the carrier scans the parcel. 48,000 runs / month 96.7% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Public procurement bid file pre-check URL: https://www.aginies.com/en/agents/public-procurement-bid-check Language: en Summary: Bid files are compared with the specification, missing and non-conforming documents listed; the evaluation decision belongs to the commission. Electronic bid files opened at the tender deadline → Document checklist and conformity summary ready for every bid; commission minutes drafted Government & Public Sectorbatchweb # Public procurement bid file pre-check Bid files are compared with the specification, missing and non-conforming documents listed; the evaluation decision belongs to the commission. In short Public procurement bid file pre-check is an agent scenario in the Government & Public Sector sector running on the batch, web channels. Trigger: Electronic bid files opened at the tender deadline. The Tender Supervisor coordinates 3 specialist agents (Document Agent, Specification Agent, Minutes Agent) across 3 systems (e-Procurement Platform, Specification Vector Store, Document Store); of 9 steps, 1 write to systems and 2 are checks or approvals. Outcome: Document checklist and conformity summary ready for every bid; commission minutes drafted. Approval rule: Conformity and elimination decisions always with the tender commission; the agent only prepares. Government & Public Sector ### Public procurement bid file pre-check [Public procurement bid file pre-check] run #41,289,431elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Conformity and elimination decisions always with the tender commission; the agent only prepares Agents in this run Tender Supervisorsupervisorplans, routes, approves Document Agentaginies-konrul-1 · on-premClassifies and reads the documents in the bid file Specification Agentaginies-tulpar-4 · on-premCompares with administrative and technical specification clauses Minutes Agentaginies-talay-2 · on-premWrites the checklist and the minutes draft Audit trail00:00.00 - ◦Tender 2026/118203: 7 bids, 214 documents, 96 clauses across specifications - ⚙Documents downloaded and classified: guarantee, qualification, experience, technical annexes - ∿Bid 3: provisional guarantee 2.9% (min 3%); Bid 5: e-signature could not be verified - ⌕Specification clauses retrieved: 14 qualification criteria, 31 technical mandatories - ∿7 bids matched: 4 fully conforming, 2 with missing documents, 1 technical non-conformity candidate - ✓Clause number and source page linked for every finding; 3 ambiguous clauses flagged for the commission - ⇢Checklists and minutes draft created; attached to the tender record - ⏸Conformity decision → tender commission reviewed the findings, corrected 2 items, signed the minutes - ➤Signed minutes and findings summary sent to commission members; process record closed - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Electronic bid files opened at the tender deadline Outcome Document checklist and conformity summary ready for every bid; commission minutes drafted ## Agents SUPERVISOR ### Tender Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Document Agent Classifies and reads the documents in the bid file extract_documentclassify_documentcheck_signature aginies-tulpar-4 · on-prem ### Specification Agent Compares with administrative and technical specification clauses search_specificationmatch_requirementlist_gaps aginies-talay-2 · on-prem ### Minutes Agent Writes the checklist and the minutes draft create_checklistdraft_minutesupdate_tender ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Tender Supervisor | Tender 2026/118203: 7 bids, 214 documents, 96 clauses across specifications | e-Procurement Platform | 200 | — | | 02 | tool | Document Agent | Documents downloaded and classified: guarantee, qualification, experience, technical annexes | Document Store | 620 | — | | 03 | reason | Document Agent | Bid 3: provisional guarantee 2.9% (min 3%); Bid 5: e-signature could not be verified | — | 1,100 | 1,600 | | 04 | retrieve | Specification Agent | Specification clauses retrieved: 14 qualification criteria, 31 technical mandatories | Specification Vector Store | 520 | 700 | | 05 | reason | Specification Agent | 7 bids matched: 4 fully conforming, 2 with missing documents, 1 technical non-conformity candidate | — | 1,400 | 2,100 | | 06 | verify | Specification Agent | Clause number and source page linked for every finding; 3 ambiguous clauses flagged for the commission | — | 480 | 380 | | 07 | write | Minutes Agent | Checklists and minutes draft created; attached to the tender record | e-Procurement Platform | 520 | — | | 08 | approval | Tender Supervisor | Conformity decision → tender commission reviewed the findings, corrected 2 items, signed the minutes | — | 3,200 | — | | 09 | notify | Minutes Agent | Signed minutes and findings summary sent to commission members; process record closed | e-Procurement Platform | 240 | — | Simulation metrics (estimated) 380runs / month 94.8%success 8.4 sp50 $0.0201est. cost / run Approval rule Conformity and elimination decisions always with the tender commission; the agent only prepares Step profile ingest · 1retrieve · 1reason · 2tool · 1verify · 1approval · 1write · 1notify · 1 SYSTEMS - e-Procurement Platformapi - Specification Vector Storedb - Document Storestorage What you need to run it - Credentials and a test environment for the enterprise APIs involved - Read access to the database or vector store; a data owner - Read access to the document store and the document types - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors document-check automation with human commission decisions in European public procurement bodies Packages that include this scenario Public procurement & tender compliance → Sector: Government & Public Sector → Related agents Government & Public SectorHITL ### Citizen services assistant: application, documents, appointment and status The citizen writes or calls; eligibility checked, documents read, application opened, appointment given, every step traceable. 230,000 runs / month 93.8% success 5 agents Government & Public SectorHITL ### Citizen request routing e-Government requests classified in seconds, given a legal deadline and routed to the right unit with an SLA. 54,000 runs / month 98.1% success 4 agents Government & Public SectorHITL ### Permit application check Missing documents and regulatory fit reported in minutes; the licensing officer always makes the decision. 3,100 runs / month 95.8% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Regression test cycle (hired QA agent) URL: https://www.aginies.com/en/agents/regression-test-cycle Language: en Summary: Regression runs on every release candidate, new bugs are filed with evidence, a coverage report is written; the “go” decision stays with the release manager. Release-candidate tag from CI → Regression result, filed bugs and coverage report on the release record; work log updated Software Deliveryapibatch # Regression test cycle (hired QA agent) Regression runs on every release candidate, new bugs are filed with evidence, a coverage report is written; the “go” decision stays with the release manager. In short Regression test cycle (hired QA agent) is an agent scenario in the Software Delivery sector running on the api, batch channels. Trigger: Release-candidate tag from CI. The QA Supervisor coordinates 3 specialist agents (Test Plan Agent, Execution Agent, Defect Agent) across 4 systems (CI/CD, Device / Browser Grid, Jira / Xray, Workforce Work Log); of 9 steps, 2 write to systems and 2 are checks or approvals. Outcome: Regression result, filed bugs and coverage report on the release record; work log updated. Approval rule: The release “go / no-go” decision stays with the release manager. Software Delivery ### Regression test cycle (hired QA agent) [Regression test cycle (hired QA agent)] run #41,289,907elapsed 0 mstokens 0est. cost $0.000status —Approval rule: The release “go / no-go” decision stays with the release manager Agents in this run QA Supervisorsupervisorplans, routes, approves Test Plan Agentaginies-tulpar-4 · on-premSelects and extends the test set by changed areas Execution Agentaginies-sunkar-1 · on-premRuns the tests, isolates flaky ones Defect Agentaginies-talay-2 · on-premFiles bugs with evidence and steps, writes the report Audit trail00:00.00 - ◦RC 2.14.0-rc3: 23 commits, changed areas payments, profile, notifications - ∿412 tests selected (full regression), 9 new scenarios derived from acceptance criteria - ⚙Run on 3 browsers + 4 devices: 409 passed, 3 failed; 2 passed on retry (flaky) - ✓Remaining failure consistent: payment summary rounding difference (0.01) - ⇢BUG-2291 filed: steps, screen recording, logs; priority high, assigned to the payments team - ∿Report: coverage 91%, 2 flaky tests quarantined, escaped defects 0 (previous release 1) - ⏸Release manager: “no-go” until BUG-2291 is fixed; re-run scheduled for rc4 - ⇢Work log: 412 tests · 1 bug · 9 new scenarios · hour-equivalent 11 · quality 4.8 - ➤Report link and next run time sent to the team - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Release-candidate tag from CI Outcome Regression result, filed bugs and coverage report on the release record; work log updated ## Agents SUPERVISOR ### QA Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-tulpar-4 · on-prem ### Test Plan Agent Selects and extends the test set by changed areas get_release_diffselect_testsgenerate_cases aginies-sunkar-1 · on-prem ### Execution Agent Runs the tests, isolates flaky ones run_playwrightrun_xcuitestretry_flakycollect_artifacts aginies-talay-2 · on-prem ### Defect Agent Files bugs with evidence and steps, writes the report file_bugwrite_reportlog_work ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | QA Supervisor | RC 2.14.0-rc3: 23 commits, changed areas payments, profile, notifications | CI/CD | 200 | — | | 02 | reason | Test Plan Agent | 412 tests selected (full regression), 9 new scenarios derived from acceptance criteria | — | 900 | 1,300 | | 03 | tool | Execution Agent | Run on 3 browsers + 4 devices: 409 passed, 3 failed; 2 passed on retry (flaky) | Device / Browser Grid | 1,800 | — | | 04 | verify | Execution Agent | Remaining failure consistent: payment summary rounding difference (0.01) | — | 520 | 380 | | 05 | write | Defect Agent | BUG-2291 filed: steps, screen recording, logs; priority high, assigned to the payments team | Jira / Xray | 520 | — | | 06 | reason | Defect Agent | Report: coverage 91%, 2 flaky tests quarantined, escaped defects 0 (previous release 1) | — | 700 | 640 | | 07 | approval | QA Supervisor | Release manager: “no-go” until BUG-2291 is fixed; re-run scheduled for rc4 | — | 2,600 | — | | 08 | write | Defect Agent | Work log: 412 tests · 1 bug · 9 new scenarios · hour-equivalent 11 · quality 4.8 | Workforce Work Log | 300 | — | | 09 | notify | Defect Agent | Report link and next run time sent to the team | Jira / Xray | 240 | — | Simulation metrics (estimated) 260runs / month 95.8%success 8.6 sp50 $0.0165est. cost / run Approval rule The release “go / no-go” decision stays with the release manager Step profile ingest · 1reason · 2tool · 1verify · 1approval · 1write · 2notify · 1 SYSTEMS - CI/CDapi - Device / Browser Gridapi - Jira / Xrayticketing - Workforce Work Logdb What you need to run it - Credentials and a test environment for the enterprise APIs involved - API access and assignment rules for the ticketing or work-order system - Read access to the database or vector store; a data owner - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern The hired QA agent model: an agent that runs every release and files evidenced bugs instead of an outsourced tester Packages that include this scenario Agent outsourcing (agent workforce) → Sector: Software Delivery → Related agents Software DeliveryHITL ### Legacy module modernisation: analysis, tests, rewrite, PR Starts from a Jira epic; the old module is read, its behaviour pinned with tests, rewritten and opened as a PR with evidence. 1,900 runs / month 90.4% success 5 agents Software DeliveryHITL ### Feature delivery pipeline A Jira epic turns into criteria, design, code and tests via agents; humans only review and approve the PR. 1,200 runs / month 91.2% success 5 agents Software DeliveryHITL ### CI/CD release pipeline Build, tests, scans and staging run unattended; production ships only on release-manager approval, via ArgoCD. 950 runs / month 94.7% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Returns and refunds URL: https://www.aginies.com/en/agents/returns-and-refunds Language: en Summary: Policy and history checked in seconds, label issued; the refund executes when the carrier scans the parcel. Return request from the app, web or email → Return record and label ready, refund armed on carrier pick-up, customer informed Retail & E-commerceappwebwhatsappemail # Returns and refunds Policy and history checked in seconds, label issued; the refund executes when the carrier scans the parcel. In short Returns and refunds is an agent scenario in the Retail & E-commerce sector running on the app, web, whatsapp, email channels. Trigger: Return request from the app, web or email. The Returns Supervisor coordinates 3 specialist agents (Intake Agent, Policy Agent, Fulfilment Agent) across 4 systems (Shopify / OMS, Carrier API (Yurtiçi, DHL), Payment Gateway (iyzico), Email (Exchange)); of 10 steps, 3 write to systems and 2 are checks or approvals. Outcome: Return record and label ready, refund armed on carrier pick-up, customer informed. Approval rule: Refund > 2,500 or > 5 returns in 12 months → team lead approval. Retail & E-commerce ### Returns and refunds [Returns and refunds] run #41,287,918elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Refund > 2,500 or > 5 returns in 12 months → team lead approval Agents in this run Returns Supervisorsupervisorplans, routes, approves Intake Agentaginies-sunkar-1 · on-premFinds the order and classifies the return reason Policy Agentaginies-tulpar-4 · on-premAssesses eligibility and abuse signals Fulfilment Agentaginies-talay-2 · on-premCreates the label, records the return, issues the refund Audit trail00:00.00 - ◦In-app return request: order #100377, 2 items, reason "wrong size" - ⚙Order delivered 26 Aug, 2 items totalling 3,180, paid by card - ✓Policy: within 14 days, tags attached, returnable category → eligible - ∿History: 3 returns in 11 orders over 12 months (27%), no abuse signal - ⌘Plan: return label → approval threshold → return record → refund → notify - ⇢Free return label created: YK-8812 0034, valid for 3 days - ⏸Refund > 2,500 → team lead approved in 2 minutes - ⇢Return record created, stock flagged as "expected return" - ⇢3,180 refunded to the card on the carrier pick-up event - ➤Email: return confirmation, label and 3–5 working day refund timeline - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Return request from the app, web or email Outcome Return record and label ready, refund armed on carrier pick-up, customer informed ## Agents SUPERVISOR ### Returns Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Intake Agent Finds the order and classifies the return reason get_orderextract_return_reasonclassify_reason aginies-tulpar-4 · on-prem ### Policy Agent Assesses eligibility and abuse signals get_return_policycheck_eligibilityget_customer_return_history aginies-talay-2 · on-prem ### Fulfilment Agent Creates the label, records the return, issues the refund create_return_labelcreate_returnissue_refundsend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Returns Supervisor | In-app return request: order #100377, 2 items, reason "wrong size" | — | 200 | — | | 02 | tool | Intake Agent | Order delivered 26 Aug, 2 items totalling 3,180, paid by card | Shopify / OMS | 360 | — | | 03 | verify | Policy Agent | Policy: within 14 days, tags attached, returnable category → eligible | — | 620 | 540 | | 04 | reason | Policy Agent | History: 3 returns in 11 orders over 12 months (27%), no abuse signal | — | 480 | 390 | | 05 | plan | Returns Supervisor | Plan: return label → approval threshold → return record → refund → notify | — | 240 | 180 | | 06 | write | Fulfilment Agent | Free return label created: YK-8812 0034, valid for 3 days | Carrier API (Yurtiçi, DHL) | 420 | — | | 07 | approval | Returns Supervisor | Refund > 2,500 → team lead approved in 2 minutes | — | 2,400 | — | | 08 | write | Fulfilment Agent | Return record created, stock flagged as "expected return" | Shopify / OMS | 380 | — | | 09 | write | Fulfilment Agent | 3,180 refunded to the card on the carrier pick-up event | Payment Gateway (iyzico) | 540 | — | | 10 | notify | Fulfilment Agent | Email: return confirmation, label and 3–5 working day refund timeline | Email (Exchange) | 240 | — | Simulation metrics (estimated) 48,000runs / month 96.7%success 5.9 sp50 $0.0012est. cost / run Approval rule Refund > 2,500 or > 5 returns in 12 months → team lead approval Step profile ingest · 1plan · 1reason · 1tool · 1verify · 1approval · 1write · 3notify · 1 SYSTEMS - Shopify / OMSerp - Carrier API (Yurtiçi, DHL)api - Payment Gateway (iyzico)payments - Email (Exchange)email What you need to run it - Posting and query API access to ERP/MES - Credentials and a test environment for the enterprise APIs involved - Authorised API access and limits for payment services - Mailbox access (Microsoft 365 or Google Workspace) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the self-service returns pattern of fashion e-commerce, with refunds triggered on the carrier scan event Packages that include this scenario E-commerce support & action → Sector: Retail & E-commerce → Related agents Retail & E-commerceHITL ### Outfit stylist: a complete look by size, stock, budget and style The customer sends a photo and a sentence; the agent builds three looks that are in stock, fit and within budget, and adds the chosen one to the cart. 142,000 runs / month 96.1% success 5 agents Retail & E-commerce ### Where is my order? (WISMO) Customer asks, order and carrier events join instantly; on delay a new ETA is given and a trace opened. 110,000 runs / month 98.1% success 4 agents Retail & E-commerce ### Product Q&A assistant Size, compatibility and spec questions answered from catalogue and return data; stock and price verified. 120,000 runs / month 98.7% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Review response and quality escalation URL: https://www.aginies.com/en/agents/review-response Language: en Summary: Reviews classified and answered in brand tone; recurring defects joined with return data and escalated to QA. Hourly review pull (site, marketplace, app store) → Reply published, exchange link with the customer, defect cluster open with QA and evidence Retail & E-commercewebapi # Review response and quality escalation Reviews classified and answered in brand tone; recurring defects joined with return data and escalated to QA. In short Review response and quality escalation is an agent scenario in the Retail & E-commerce sector running on the web, api channels. Trigger: Hourly review pull (site, marketplace, app store). The Voice-of-Customer Supervisor coordinates 3 specialist agents (Classification Agent, Response Agent, Quality Agent) across 3 systems (Review Platform API, Shopify / OMS, Jira (QA)); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: Reply published, exchange link with the customer, defect cluster open with QA and evidence. Approval rule: 1-star review replies and defect escalations → community manager approval. Retail & E-commerce ### Review response and quality escalation [Review response and quality escalation] run #41,287,248elapsed 0 mstokens 0est. cost $0.000status —Approval rule: 1-star review replies and defect escalations → community manager approval Agents in this run Voice-of-Customer Supervisorsupervisorplans, routes, approves Classification Agentaginies-sunkar-1 · on-premDetects sentiment, topic and defect signal Response Agentaginies-tulpar-4 · on-premDrafts an on-brand reply using order context Quality Agentaginies-talay-2 · on-premClusters defects, compares with return rate, opens the QA ticket Audit trail00:00.00 - ◦Hourly pull: 184 new reviews, 23 rated ≤ 2 stars - ∿Review #R-77120: 1 star, "zip broke in 2 weeks" → defect signal, SKU 55-2210 - ⌘Plan: order context → response draft → defect cluster → approval → QA escalation - ⚙Order #99841 delivered 18 Aug, under warranty, exchange available - ∿Response drafted: apology + free exchange offer, brand tone, 68 words - ⚙SKU 55-2210: 17 "zip" reviews in 30 days, return rate 9.4% (average 3.1%) - ✓Rule: ≥ 10 similar defects or return rate > 2× average → QA escalation - ⏸1-star reply → community manager approved from the queue - ⇢Jira QA-1184 opened with supplier batch, 17 reviews and return data attached - ➤Reply published on the platform with the exchange link included - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Hourly review pull (site, marketplace, app store) Outcome Reply published, exchange link with the customer, defect cluster open with QA and evidence ## Agents SUPERVISOR ### Voice-of-Customer Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Classification Agent Detects sentiment, topic and defect signal classify_sentimentdetect_topicsdetect_defect_signal aginies-tulpar-4 · on-prem ### Response Agent Drafts an on-brand reply using order context get_order_contextdraft_responsecheck_tonepublish_reply aginies-talay-2 · on-prem ### Quality Agent Clusters defects, compares with return rate, opens the QA ticket cluster_defectsget_sku_return_ratecreate_qa_ticket ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Voice-of-Customer Supervisor | Hourly pull: 184 new reviews, 23 rated ≤ 2 stars | — | 260 | — | | 02 | reason | Classification Agent | Review #R-77120: 1 star, "zip broke in 2 weeks" → defect signal, SKU 55-2210 | — | 520 | 380 | | 03 | plan | Voice-of-Customer Supervisor | Plan: order context → response draft → defect cluster → approval → QA escalation | — | 240 | 170 | | 04 | tool | Response Agent | Order #99841 delivered 18 Aug, under warranty, exchange available | Shopify / OMS | 340 | — | | 05 | reason | Response Agent | Response drafted: apology + free exchange offer, brand tone, 68 words | — | 680 | 620 | | 06 | tool | Quality Agent | SKU 55-2210: 17 "zip" reviews in 30 days, return rate 9.4% (average 3.1%) | Shopify / OMS | 380 | — | | 07 | verify | Quality Agent | Rule: ≥ 10 similar defects or return rate > 2× average → QA escalation | — | 420 | 310 | | 08 | approval | Voice-of-Customer Supervisor | 1-star reply → community manager approved from the queue | — | 2,200 | — | | 09 | write | Quality Agent | Jira QA-1184 opened with supplier batch, 17 reviews and return data attached | Jira (QA) | 460 | — | | 10 | notify | Response Agent | Reply published on the platform with the exchange link included | Review Platform API | 320 | — | Simulation metrics (estimated) 26,000runs / month 97.2%success 5.8 sp50 $0.0009est. cost / run Approval rule 1-star review replies and defect escalations → community manager approval Step profile ingest · 1plan · 1reason · 2tool · 2verify · 1approval · 1write · 1notify · 1 SYSTEMS - Review Platform APIapi - Shopify / OMSerp - Jira (QA)ticketing What you need to run it - Credentials and a test environment for the enterprise APIs involved - Posting and query API access to ERP/MES - API access and assignment rules for the ticketing or work-order system - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the review-to-product-quality feedback loop at global marketplaces Packages that include this scenario Call-centre quality monitoring →Sales & lead management → Sector: Retail & E-commerce → Related agents Retail & E-commerceHITL ### Outfit stylist: a complete look by size, stock, budget and style The customer sends a photo and a sentence; the agent builds three looks that are in stock, fit and within budget, and adds the chosen one to the cart. 142,000 runs / month 96.1% success 5 agents Retail & E-commerce ### Where is my order? (WISMO) Customer asks, order and carrier events join instantly; on delay a new ETA is given and a trace opened. 110,000 runs / month 98.1% success 4 agents Retail & E-commerceHITL ### Returns and refunds Policy and history checked in seconds, label issued; the refund executes when the carrier scans the parcel. 48,000 runs / month 96.7% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Rooftop solar connection application URL: https://www.aginies.com/en/agents/rooftop-solar-connection-application Language: en Summary: Application documents are read, transformer capacity and regulation checked, gaps reported the same day; the connection opinion belongs to the engineer. Rooftop solar connection application from the DSO portal → Technical pre-assessment and gap list ready; connection opinion published with engineer approval Energy & Utilitieswebemailapi # Rooftop solar connection application Application documents are read, transformer capacity and regulation checked, gaps reported the same day; the connection opinion belongs to the engineer. In short Rooftop solar connection application is an agent scenario in the Energy & Utilities sector running on the web, email, api channels. Trigger: Rooftop solar connection application from the DSO portal. The Connection Supervisor coordinates 4 specialist agents (Document Agent, Grid Agent, Regulation Agent, Case Agent) across 4 systems (Connection Portal, GIS / Network Model, Regulation Vector Store (EPDK), E-mail); of 9 steps, 1 write to systems and 2 are checks or approvals. Outcome: Technical pre-assessment and gap list ready; connection opinion published with engineer approval. Approval rule: The connection opinion is always signed by a network planning engineer. Energy & Utilities ### Rooftop solar connection application [Rooftop solar connection application] run #41,289,256elapsed 0 mstokens 0est. cost $0.000status —Approval rule: The connection opinion is always signed by a network planning engineer Agents in this run Connection Supervisorsupervisorplans, routes, approves Document Agentaginies-konrul-1 · on-premReads the single-line diagram, deed and permit Grid Agentaginies-tulpar-4 · on-premEvaluates transformer and feeder capacity Regulation Agentaginies-talay-2 · on-premChecks against unlicensed-generation regulation Case Agentaginies-sunkar-1 · on-premUpdates the application and informs the applicant Audit trail00:00.00 - ◦Application B-2026-31877: 48 kWp rooftop solar, commercial customer, 6 documents uploaded - ∿Documents read: single-line diagram valid, deed matches, building permit missing a page - ⚙Transformer TR-2291: 400 kVA, peak load 71%, connected generation 96 kWp - ∿Hosting capacity sufficient: with 48 kWp total generation is 36% of transformer rating - ⌕Regulation: same metering point as consumption, within contracted power limit → eligible - ✓Opinion draft: positive, conditional on protection settings and meter change; 1 missing document - ⏸Connection opinion → network planning engineer reviewed and signed - ⇢Application set to “conditionally positive”, opinion document attached - ➤Opinion and missing-document list e-mailed to the applicant; 15-day completion window - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Rooftop solar connection application from the DSO portal Outcome Technical pre-assessment and gap list ready; connection opinion published with engineer approval ## Agents SUPERVISOR ### Connection Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Document Agent Reads the single-line diagram, deed and permit extract_documentclassify_documentcheck_completeness aginies-tulpar-4 · on-prem ### Grid Agent Evaluates transformer and feeder capacity get_transformer_loadget_connected_generationassess_hosting_capacity aginies-talay-2 · on-prem ### Regulation Agent Checks against unlicensed-generation regulation search_regulationcheck_eligibilitydraft_opinion aginies-sunkar-1 · on-prem ### Case Agent Updates the application and informs the applicant update_applicationsend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Connection Supervisor | Application B-2026-31877: 48 kWp rooftop solar, commercial customer, 6 documents uploaded | Connection Portal | 200 | — | | 02 | reason | Document Agent | Documents read: single-line diagram valid, deed matches, building permit missing a page | — | 980 | 1,300 | | 03 | tool | Grid Agent | Transformer TR-2291: 400 kVA, peak load 71%, connected generation 96 kWp | GIS / Network Model | 460 | — | | 04 | reason | Grid Agent | Hosting capacity sufficient: with 48 kWp total generation is 36% of transformer rating | — | 700 | 760 | | 05 | retrieve | Regulation Agent | Regulation: same metering point as consumption, within contracted power limit → eligible | Regulation Vector Store (EPDK) | 480 | 520 | | 06 | verify | Regulation Agent | Opinion draft: positive, conditional on protection settings and meter change; 1 missing document | — | 620 | 640 | | 07 | approval | Connection Supervisor | Connection opinion → network planning engineer reviewed and signed | — | 3,000 | — | | 08 | write | Case Agent | Application set to “conditionally positive”, opinion document attached | Connection Portal | 440 | — | | 09 | notify | Case Agent | Opinion and missing-document list e-mailed to the applicant; 15-day completion window | E-mail | 240 | — | Simulation metrics (estimated) 5,400runs / month 95.3%success 7.1 sp50 $0.0064est. cost / run Approval rule The connection opinion is always signed by a network planning engineer Step profile ingest · 1retrieve · 1reason · 2tool · 1verify · 1approval · 1write · 1notify · 1 SYSTEMS - Connection Portalapi - GIS / Network Modeldb - Regulation Vector Store (EPDK)db - E-mailemail What you need to run it - Credentials and a test environment for the enterprise APIs involved - Read access to the database or vector store; a data owner - Mailbox access (Microsoft 365 or Google Workspace) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors distributed-generation connection application automation at European DSOs Packages that include this scenario Utility customer operations → Sector: Energy & Utilities → Related agents Energy & UtilitiesHITL ### Utility customer assistant: bill, outage, consumption and instalments in one dialogue The customer calls or writes; the high bill is explained, outage ETA given, meter data shown, an instalment plan set up. 520,000 runs / month 94.6% success 5 agents Energy & UtilitiesHITL ### Outage report handling Reports are clustered by feeder, one OSS ticket is raised and everyone affected hears before the crew leaves. 38,000 runs / month 97.7% success 5 agents Energy & UtilitiesHITL ### Meter reading dispute Customer sends a meter photo; the index is read by vision, the estimated bill fixed, big credits approved. 9,200 runs / month 94.9% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Sales lead qualification URL: https://www.aginies.com/en/agents/sales-lead-qualification Language: en Summary: Inbound leads enriched, scored and in the CRM within seconds; the SDR starts the call with a context brief. "Request a demo" form submitted on the website → Lead at SQL stage, SDR assigned, call task within 4 h and context brief ready Enterprise Functionswebapiemail # Sales lead qualification Inbound leads enriched, scored and in the CRM within seconds; the SDR starts the call with a context brief. In short Sales lead qualification is an agent scenario in the Enterprise Functions sector running on the web, api, email channels. Trigger: "Request a demo" form submitted on the website. The Sales Supervisor coordinates 3 specialist agents (Enrichment Agent, Scoring Agent, CRM Agent) across 3 systems (HubSpot, Enrichment API, Microsoft Teams); of 9 steps, 2 write to systems and 2 are checks or approvals. Outcome: Lead at SQL stage, SDR assigned, call task within 4 h and context brief ready. Approval rule: Enterprise (> 1,000 employees) or suspected competitor → SDR lead acceptance. Enterprise Functions ### Sales lead qualification [Sales lead qualification] run #41,289,930elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Enterprise (> 1,000 employees) or suspected competitor → SDR lead acceptance Agents in this run Sales Supervisorsupervisorplans, routes, approves Enrichment Agentaginies-tulpar-4 · on-premEnriches company and contact data Scoring Agentaginies-tulpar-4 · on-premScores on ICP fit and intent signals CRM Agentaginies-talay-2 · on-premUpdates the CRM and hands off to the SDR Audit trail00:00.00 - ◦Web form: "request a demo" from ayse.demir@…, company field "Norm Holding" - ⚙Company: 2,400 employees, manufacturing, İzmir; title: Digital Transformation Lead - ⚙CRM history: 1 closed-lost deal in 2024 (budget), 3 webinar attendances - ∿ICP fit 88/100 · intent high (pricing page ×3) · SQL threshold 75 → qualified - ✓Guardrail: corporate domain, not a competitor, KVKK/GDPR consent present - ⇢Lead upserted, stage "SQL", owner: SDR Team Marmara (round-robin) - ⏸Enterprise (> 1,000 employees) → SDR lead accepted before outreach - ⇢Task: call within 4 h; 3-line context brief attached to the record - ➤SDR pinged in Teams with the brief; personalised follow-up e-mail drafted, not sent - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger "Request a demo" form submitted on the website Outcome Lead at SQL stage, SDR assigned, call task within 4 h and context brief ready ## Agents SUPERVISOR ### Sales Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-tulpar-4 · on-prem ### Enrichment Agent Enriches company and contact data enrich_companyfind_contact_profilecheck_domain aginies-tulpar-4 · on-prem ### Scoring Agent Scores on ICP fit and intent signals search_crm_historymatch_icpscore_lead aginies-talay-2 · on-prem ### CRM Agent Updates the CRM and hands off to the SDR upsert_leadassign_ownercreate_tasksend_teams_message ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Sales Supervisor | Web form: "request a demo" from ayse.demir@…, company field "Norm Holding" | — | 180 | — | | 02 | tool | Enrichment Agent | Company: 2,400 employees, manufacturing, İzmir; title: Digital Transformation Lead | Enrichment API | 520 | — | | 03 | tool | Scoring Agent | CRM history: 1 closed-lost deal in 2024 (budget), 3 webinar attendances | HubSpot | 380 | — | | 04 | reason | Scoring Agent | ICP fit 88/100 · intent high (pricing page ×3) · SQL threshold 75 → qualified | — | 760 | 820 | | 05 | verify | Scoring Agent | Guardrail: corporate domain, not a competitor, KVKK/GDPR consent present | — | 300 | 180 | | 06 | write | CRM Agent | Lead upserted, stage "SQL", owner: SDR Team Marmara (round-robin) | HubSpot | 480 | — | | 07 | approval | Sales Supervisor | Enterprise (> 1,000 employees) → SDR lead accepted before outreach | — | 2,200 | — | | 08 | write | CRM Agent | Task: call within 4 h; 3-line context brief attached to the record | HubSpot | 360 | — | | 09 | notify | CRM Agent | SDR pinged in Teams with the brief; personalised follow-up e-mail drafted, not sent | Microsoft Teams | 240 | — | Simulation metrics (estimated) 15,600runs / month 97.9%success 5.4 sp50 $0.0021est. cost / run Approval rule Enterprise (> 1,000 employees) or suspected competitor → SDR lead acceptance Step profile ingest · 1reason · 1tool · 2verify · 1approval · 1write · 2notify · 1 SYSTEMS - HubSpotcrm - Enrichment APIapi - Microsoft Teamschannel What you need to run it - CRM API access and a customer matching key - Credentials and a test environment for the enterprise APIs involved - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors inbound speed-to-lead automation at SaaS revenue teams Packages that include this scenario Sales & lead management → Related agents Enterprise FunctionsHITL ### Procure-to-pay desk: request, quotes, contract and invoice in one place The request arrives by email; quotes compared, contract clauses checked, PO opened, invoice three-way matched and payment scheduled. 14,000 runs / month 92.7% success 5 agents Enterprise Functions ### Email classification and attachment check Every e-mail in the shared mailbox is classified, attachments checked against a list, gaps asked for at once. 86,000 runs / month 97.6% success 4 agents Enterprise FunctionsHITL ### Call centre quality monitoring Every call scored against the rubric, not a sample; compliance breaches and coaching flags ready by morning. 90,000 runs / month 98.8% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Shipment exception handling URL: https://www.aginies.com/en/agents/shipment-exception-handling Language: en Summary: When a delay or damage lands, impact is costed, the carrier rebooked and the customer hears before they ask. Delay or damage exception from the carrier webhook → Shipment re-planned, exception closed with a cause code, customer told the new ETA Logisticsapiemailwhatsapp # Shipment exception handling When a delay or damage lands, impact is costed, the carrier rebooked and the customer hears before they ask. In short Shipment exception handling is an agent scenario in the Logistics sector running on the api, email, whatsapp channels. Trigger: Delay or damage exception from the carrier webhook. The Shipment Supervisor coordinates 3 specialist agents (Exception Agent, Impact Agent, Action Agent) across 4 systems (TMS (Transport Management), Carrier API, Salesforce CRM, WhatsApp Business API); of 10 steps, 2 write to systems and 2 are checks or approvals. Outcome: Shipment re-planned, exception closed with a cause code, customer told the new ETA. Approval rule: Extra cost > 1,000 € → dispatch lead approval. Logistics ### Shipment exception handling [Shipment exception handling] run #41,288,821elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Extra cost > 1,000 € → dispatch lead approval Agents in this run Shipment Supervisorsupervisorplans, routes, approves Exception Agentaginies-sunkar-1 · on-premClassifies the exception and its severity Impact Agentaginies-tulpar-4 · on-premCosts SLA and penalty exposure Action Agentaginies-talay-2 · on-premRebooks the carrier and communicates with the customer Audit trail00:00.00 - ◦Carrier webhook: shipment SH-4471820 "delayed – hub congestion", ETA +36 h - ⚙Shipment: 14 pallets, cold chain, customer C-2081, due Wed 08:00 - ∿Exception class: delay · severity high (cold chain, contractual SLA) - ⚙Customer SLA: 98% on-time, penalty 250 € per late pallet-day - ∿Estimated impact: 3,500 € penalty + reefer risk; alternative carrier +1,200 € - ✓Policy: rebooking allowed if extra cost < 40% of penalty exposure → 34% → allowed - ⏸Extra cost > 1,000 € → dispatch lead approval (approved in 6 min) - ⇢Pickup rebooked with carrier B, new ETA Wed 06:30, reference RB-88231 - ⇢Shipment re-planned, exception closed with cause code "hub congestion" - ➤Customer logistics contact told the new ETA, account manager copied by e-mail - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Delay or damage exception from the carrier webhook Outcome Shipment re-planned, exception closed with a cause code, customer told the new ETA ## Agents SUPERVISOR ### Shipment Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Exception Agent Classifies the exception and its severity classify_exceptionget_shipment_status aginies-tulpar-4 · on-prem ### Impact Agent Costs SLA and penalty exposure get_customer_slaestimate_delay_costsearch_contract_terms aginies-talay-2 · on-prem ### Action Agent Rebooks the carrier and communicates with the customer rebook_carrierupdate_shipmentcreate_claimsend_whatsappsend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Shipment Supervisor | Carrier webhook: shipment SH-4471820 "delayed – hub congestion", ETA +36 h | — | 180 | — | | 02 | tool | Exception Agent | Shipment: 14 pallets, cold chain, customer C-2081, due Wed 08:00 | TMS (Transport Management) | 400 | — | | 03 | reason | Exception Agent | Exception class: delay · severity high (cold chain, contractual SLA) | — | 540 | 460 | | 04 | tool | Impact Agent | Customer SLA: 98% on-time, penalty 250 € per late pallet-day | Salesforce CRM | 380 | — | | 05 | reason | Impact Agent | Estimated impact: 3,500 € penalty + reefer risk; alternative carrier +1,200 € | — | 720 | 880 | | 06 | verify | Impact Agent | Policy: rebooking allowed if extra cost 1,000 € → dispatch lead approval (approved in 6 min) | — | 2,600 | — | | 08 | write | Action Agent | Pickup rebooked with carrier B, new ETA Wed 06:30, reference RB-88231 | Carrier API | 620 | — | | 09 | write | Action Agent | Shipment re-planned, exception closed with cause code "hub congestion" | TMS (Transport Management) | 420 | — | | 10 | notify | Action Agent | Customer logistics contact told the new ETA, account manager copied by e-mail | WhatsApp Business API | 260 | — | Simulation metrics (estimated) 12,800runs / month 95.5%success 6.5 sp50 $0.0037est. cost / run Approval rule Extra cost > 1,000 € → dispatch lead approval Step profile ingest · 1reason · 2tool · 2verify · 1approval · 1write · 2notify · 1 SYSTEMS - TMS (Transport Management)core - Carrier APIapi - Salesforce CRMcrm - WhatsApp Business APIchannel What you need to run it - Read/write API or MCP server and a service account for the core system - Credentials and a test environment for the enterprise APIs involved - CRM API access and a customer matching key - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors control-tower exception management at global 3PLs and parcel networks Packages that include this scenario Logistics control tower → Sector: Logistics → Related agents LogisticsHITL ### Control tower: delay prediction, rerouting, customer notice, SLA The telematics signal catches the delay early; alternative route and carrier chosen, customer informed, SLA compensation computed. 96,000 runs / month 93.4% success 5 agents LogisticsHITL ### Customs document preparation HS codes, invoice and packing list prepared in minutes; only low-confidence lines reach a customs specialist. 4,300 runs / month 94.7% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # SIM swap fraud check URL: https://www.aginies.com/en/agents/sim-swap-fraud-check Language: en Summary: Every SIM swap is risk-scored before provisioning; suspicious requests are held and banks are signalled. SIM swap request from a dealer, the app or the API → Request provisioned or held, banks signalled, customer alerted Telecombranchappapi # SIM swap fraud check Every SIM swap is risk-scored before provisioning; suspicious requests are held and banks are signalled. In short SIM swap fraud check is an agent scenario in the Telecom sector running on the branch, app, api channels. Trigger: SIM swap request from a dealer, the app or the API. The SIM Security Supervisor coordinates 3 specialist agents (Risk Agent, Verification Agent, Provisioning Agent) across 4 systems (Salesforce CRM, HLR / Provisioning, Bank SIM-Swap Signal API, SMS Gateway); of 10 steps, 2 write to systems and 2 are checks or approvals. Outcome: Request provisioned or held, banks signalled, customer alerted. Approval rule: Risk score ≥ 70 or credential change within 72 hours → fraud desk approval. Telecom ### SIM swap fraud check [SIM swap fraud check] run #41,290,457elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Risk score ≥ 70 or credential change within 72 hours → fraud desk approval Agents in this run SIM Security Supervisorsupervisorplans, routes, approves Risk Agentaginies-talay-2 · on-premScores risk from account events and device history Verification Agentaginies-sunkar-1 · on-premSends OTP to the old SIM and verifies the ID document Provisioning Agentaginies-sunkar-1 · on-premActivates or holds the SIM, informs banks and the customer Audit trail00:00.00 - ◦Dealer request: SIM swap for 0532 *** 8810, reason "lost" - ⚙Account events: password reset 2 days ago, email changed 6 hours ago - ⚙Device history: same IMEI active for 14 months, 3 login attempts in 24 hours - ∿Risk score 78/100: recent credential changes + dealer channel → high - ⌘Plan: OTP to old SIM → ID document → approval → provision or hold - ✓OTP to old SIM unanswered; in-store ID OCR match 71% → insufficient - ⏸Risk ≥ 70 → fraud desk: 24-hour hold, video verification requested - ⇢SIM swap recorded as "on hold", activation blocked - ⇢Per-line SIM-swap attempt signal shared with banks via the API - ➤Alert to the old SIM and registered email: "if this wasn't you, call 444" - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger SIM swap request from a dealer, the app or the API Outcome Request provisioned or held, banks signalled, customer alerted ## Agents SUPERVISOR ### SIM Security Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Risk Agent Scores risk from account events and device history get_account_eventsget_device_historyscore_sim_swap_risk aginies-sunkar-1 · on-prem ### Verification Agent Sends OTP to the old SIM and verifies the ID document send_otp_old_simverify_id_documentcheck_recent_password_reset aginies-sunkar-1 · on-prem ### Provisioning Agent Activates or holds the SIM, informs banks and the customer activate_simhold_requestnotify_banks_apisend_smssend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | SIM Security Supervisor | Dealer request: SIM swap for 0532 *** 8810, reason "lost" | — | 200 | — | | 02 | tool | Risk Agent | Account events: password reset 2 days ago, email changed 6 hours ago | Salesforce CRM | 420 | — | | 03 | tool | Risk Agent | Device history: same IMEI active for 14 months, 3 login attempts in 24 hours | HLR / Provisioning | 380 | — | | 04 | reason | Risk Agent | Risk score 78/100: recent credential changes + dealer channel → high | — | 680 | 720 | | 05 | plan | SIM Security Supervisor | Plan: OTP to old SIM → ID document → approval → provision or hold | — | 240 | 180 | | 06 | verify | Verification Agent | OTP to old SIM unanswered; in-store ID OCR match 71% → insufficient | — | 1,600 | 420 | | 07 | approval | SIM Security Supervisor | Risk ≥ 70 → fraud desk: 24-hour hold, video verification requested | — | 3,000 | — | | 08 | write | Provisioning Agent | SIM swap recorded as "on hold", activation blocked | HLR / Provisioning | 400 | — | | 09 | write | Provisioning Agent | Per-line SIM-swap attempt signal shared with banks via the API | Bank SIM-Swap Signal API | 360 | — | | 10 | notify | Provisioning Agent | Alert to the old SIM and registered email: "if this wasn't you, call 444" | SMS Gateway | 240 | — | Simulation metrics (estimated) 41,000runs / month 97.7%success 7.5 sp50 $0.001est. cost / run Approval rule Risk score ≥ 70 or credential change within 72 hours → fraud desk approval Step profile ingest · 1plan · 1reason · 1tool · 2verify · 1approval · 1write · 2notify · 1 SYSTEMS - Salesforce CRMcrm - HLR / Provisioningcore - Bank SIM-Swap Signal APIapi - SMS Gatewaychannel What you need to run it - CRM API access and a customer matching key - Read/write API or MCP server and a service account for the core system - Credentials and a test environment for the enterprise APIs involved - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the SIM-swap risk-signal pattern UK/EU operators share with banks Packages that include this scenario Subscriber care and network support → Sector: Telecom → Related agents TelecomHITL ### Subscriber care assistant: bill, plan, fault and contract The subscriber calls or writes; bill dispute, plan change and line fault are resolved in BSS and network systems in one session. 410,000 runs / month 94.3% success 5 agents TelecomHITL ### Bill dispute resolution Invoice lines and usage explained in seconds; goodwill credit applied by policy, larger amounts need a lead. 88,000 runs / month 96.3% success 4 agents TelecomHITL ### Network outage communications When an outage ticket opens, affected subscribers are found and the approved message sent before they call. 1,400 runs / month 98.5% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Social assistance eligibility pre-assessment URL: https://www.aginies.com/en/agents/social-assistance-eligibility Language: en Summary: The application is verified against agency records, eligibility scored, gaps reported to the citizen; the decision belongs to the case officer. Assistance application via e-Government or a social service centre → Verified file and eligibility summary reached the officer; citizen received gap and timeline information Government & Public Sectorwebappvoicebatch # Social assistance eligibility pre-assessment The application is verified against agency records, eligibility scored, gaps reported to the citizen; the decision belongs to the case officer. In short Social assistance eligibility pre-assessment is an agent scenario in the Government & Public Sector sector running on the web, app, voice, batch channels. Trigger: Assistance application via e-Government or a social service centre. The Social Assistance Supervisor coordinates 3 specialist agents (Verification Agent, Eligibility Agent, Citizen Comms Agent) across 4 systems (Social Assistance Information System, Civil Registry / Income Services, Eligibility Criteria (vector), SMS Gateway); of 9 steps, 1 write to systems and 2 are checks or approvals. Outcome: Verified file and eligibility summary reached the officer; citizen received gap and timeline information. Approval rule: The assistance decision is always with the case officer; the agent pre-assesses. Government & Public Sector ### Social assistance eligibility pre-assessment [Social assistance eligibility pre-assessment] run #41,289,986elapsed 0 mstokens 0est. cost $0.000status —Approval rule: The assistance decision is always with the case officer; the agent pre-assesses Agents in this run Social Assistance Supervisorsupervisorplans, routes, approves Verification Agentaginies-sunkar-1 · on-premQueries civil registry, income and social security records Eligibility Agentaginies-tulpar-4 · on-premScores the criteria and writes the rationale Citizen Comms Agentaginies-talay-2 · on-premInforms the applicant in plain language Audit trail00:00.00 - ◦Application SY-2026-77410: household of 4, regular assistance request, identity verified - ⚙Household composition and address verified; 2 adults, 2 children (7 and 11) - ⚙Income records: 1 registered employee, monthly gross 21,400; social security active - ⌕Criteria: per-capita income threshold, property, household status, prior assistance - ∿Per-capita income 8% below threshold; no property record → pre-eligibility positive, home visit required - ✓Rationale written in plain language; sensitive fields masked on the officer screen - ⇢File set to “pre-eligible”, home visit scheduled within 3 working days - ⏸Assistance decision → case officer approved the file after the visit - ➤SMS to the citizen: decision, first payment date and right of appeal; record closed with an audit trail - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Assistance application via e-Government or a social service centre Outcome Verified file and eligibility summary reached the officer; citizen received gap and timeline information ## Agents SUPERVISOR ### Social Assistance Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Verification Agent Queries civil registry, income and social security records get_householdget_income_recordsget_social_security_status aginies-tulpar-4 · on-prem ### Eligibility Agent Scores the criteria and writes the rationale search_criteriascore_eligibilityexplain_decision aginies-talay-2 · on-prem ### Citizen Comms Agent Informs the applicant in plain language send_smsupdate_caseschedule_home_visit ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Social Assistance Supervisor | Application SY-2026-77410: household of 4, regular assistance request, identity verified | Social Assistance Information System | 200 | — | | 02 | tool | Verification Agent | Household composition and address verified; 2 adults, 2 children (7 and 11) | Civil Registry / Income Services | 480 | — | | 03 | tool | Verification Agent | Income records: 1 registered employee, monthly gross 21,400; social security active | Civil Registry / Income Services | 520 | — | | 04 | retrieve | Eligibility Agent | Criteria: per-capita income threshold, property, household status, prior assistance | Eligibility Criteria (vector) | 460 | 520 | | 05 | reason | Eligibility Agent | Per-capita income 8% below threshold; no property record → pre-eligibility positive, home visit required | — | 900 | 1,200 | | 06 | verify | Eligibility Agent | Rationale written in plain language; sensitive fields masked on the officer screen | — | 420 | 340 | | 07 | write | Citizen Comms Agent | File set to “pre-eligible”, home visit scheduled within 3 working days | Social Assistance Information System | 480 | — | | 08 | approval | Social Assistance Supervisor | Assistance decision → case officer approved the file after the visit | — | 3,200 | — | | 09 | notify | Citizen Comms Agent | SMS to the citizen: decision, first payment date and right of appeal; record closed with an audit trail | SMS Gateway | 260 | — | Simulation metrics (estimated) 21,000runs / month 96.6%success 6.9 sp50 $0.0034est. cost / run Approval rule The assistance decision is always with the case officer; the agent pre-assesses Step profile ingest · 1retrieve · 1reason · 1tool · 2verify · 1approval · 1write · 1notify · 1 SYSTEMS - Social Assistance Information Systemcore - Civil Registry / Income Servicesapi - Eligibility Criteria (vector)db - SMS Gatewaychannel What you need to run it - Read/write API or MCP server and a service account for the core system - Credentials and a test environment for the enterprise APIs involved - Read access to the database or vector store; a data owner - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors record verification and explainable pre-assessment models of European social service agencies Packages that include this scenario Citizen services & file pre-review → Sector: Government & Public Sector → Related agents Government & Public SectorHITL ### Citizen services assistant: application, documents, appointment and status The citizen writes or calls; eligibility checked, documents read, application opened, appointment given, every step traceable. 230,000 runs / month 93.8% success 5 agents Government & Public SectorHITL ### Citizen request routing e-Government requests classified in seconds, given a legal deadline and routed to the right unit with an SLA. 54,000 runs / month 98.1% success 4 agents Government & Public SectorHITL ### Permit application check Missing documents and regulatory fit reported in minutes; the licensing officer always makes the decision. 3,100 runs / month 95.8% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Subrogation recovery URL: https://www.aginies.com/en/agents/subrogation-recovery Language: en Summary: Closed claims scanned, the at-fault party's insurer identified, and the demand sent after counsel approval. Weekly batch: motor claims closed in the last 90 days → Recovery case open, demand letter with the counterparty insurer, 30-day follow-up scheduled Insurancebatchemailapi # Subrogation recovery Closed claims scanned, the at-fault party's insurer identified, and the demand sent after counsel approval. In short Subrogation recovery is an agent scenario in the Insurance sector running on the batch, email, api channels. Trigger: Weekly batch: motor claims closed in the last 90 days. The Subrogation Supervisor coordinates 3 specialist agents (Liability Scan Agent, Evidence Agent, Recovery Agent) across 4 systems (Guidewire ClaimCenter, Document Store (S3), Insurance Registry (SBM), Email (Exchange)); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: Recovery case open, demand letter with the counterparty insurer, 30-day follow-up scheduled. Approval rule: Recovery demand > 50,000 → subrogation counsel approval. Insurance ### Subrogation recovery [Subrogation recovery] run #41,290,016elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Recovery demand > 50,000 → subrogation counsel approval Agents in this run Subrogation Supervisorsupervisorplans, routes, approves Liability Scan Agentaginies-talay-2 · on-premScores third-party liability across closed claims Evidence Agentaginies-tulpar-4 · on-premSummarises evidence and drafts the demand letter with legal basis Recovery Agentaginies-sunkar-1 · on-premFinds the counterparty insurer, opens the recovery case, sends the letter Audit trail00:00.00 - ◦Weekly batch: 1,240 closed motor claims scanned for subrogation - ∿Claim #CL-2026-104877: 86,000 paid, third party 100% at fault (accident report) - ⌘Plan: evidence pack → counterparty insurer → demand letter → approval → send - ⌕Accident report, photos and repair invoices retrieved (11 documents) - ⚙Third-party vehicle 06 XYZ 456 → liability insurer found in the registry, policy active - ∿Demand letter drafted: 86,000, interest from 14 Apr, legal basis cited - ✓Limitation: within 2 years · amount reconciles with the file · legal basis correct - ⏸Demand > 50,000 → subrogation counsel approved the letter - ⇢Recovery case #SR-8817 opened, expected recovery 86,000 - ➤Demand letter emailed to the counterparty insurer, 30-day follow-up scheduled - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Weekly batch: motor claims closed in the last 90 days Outcome Recovery case open, demand letter with the counterparty insurer, 30-day follow-up scheduled ## Agents SUPERVISOR ### Subrogation Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Liability Scan Agent Scores third-party liability across closed claims get_closed_claimsscore_third_party_liabilityget_accident_report aginies-tulpar-4 · on-prem ### Evidence Agent Summarises evidence and drafts the demand letter with legal basis fetch_claim_documentssummarise_evidencedraft_demand_lettercheck_limitation_period aginies-sunkar-1 · on-prem ### Recovery Agent Finds the counterparty insurer, opens the recovery case, sends the letter lookup_insurercreate_recovery_casesend_emailschedule_followup ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Subrogation Supervisor | Weekly batch: 1,240 closed motor claims scanned for subrogation | — | 340 | — | | 02 | reason | Liability Scan Agent | Claim #CL-2026-104877: 86,000 paid, third party 100% at fault (accident report) | — | 820 | 760 | | 03 | plan | Subrogation Supervisor | Plan: evidence pack → counterparty insurer → demand letter → approval → send | — | 280 | 200 | | 04 | retrieve | Evidence Agent | Accident report, photos and repair invoices retrieved (11 documents) | Document Store (S3) | 420 | — | | 05 | tool | Recovery Agent | Third-party vehicle 06 XYZ 456 → liability insurer found in the registry, policy active | Insurance Registry (SBM) | 460 | — | | 06 | reason | Evidence Agent | Demand letter drafted: 86,000, interest from 14 Apr, legal basis cited | — | 1,300 | 1,580 | | 07 | verify | Evidence Agent | Limitation: within 2 years · amount reconciles with the file · legal basis correct | — | 520 | 440 | | 08 | approval | Subrogation Supervisor | Demand > 50,000 → subrogation counsel approved the letter | — | 3,200 | — | | 09 | write | Recovery Agent | Recovery case #SR-8817 opened, expected recovery 86,000 | Guidewire ClaimCenter | 480 | — | | 10 | notify | Recovery Agent | Demand letter emailed to the counterparty insurer, 30-day follow-up scheduled | Email (Exchange) | 260 | — | Simulation metrics (estimated) 3,400runs / month 95.5%success 8.1 sp50 $0.0087est. cost / run Approval rule Recovery demand > 50,000 → subrogation counsel approval Step profile ingest · 1plan · 1retrieve · 1reason · 2tool · 1verify · 1approval · 1write · 1notify · 1 SYSTEMS - Guidewire ClaimCentercore - Document Store (S3)storage - Insurance Registry (SBM)api - Email (Exchange)email What you need to run it - Read/write API or MCP server and a service account for the core system - Read access to the document store and the document types - Credentials and a test environment for the enterprise APIs involved - Mailbox access (Microsoft 365 or Google Workspace) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the missed-subrogation detection pattern US carriers use to lift recovery rates Packages that include this scenario Claims operations end to end → Sector: Insurance → Related agents InsuranceHITL ### End-to-end motor claim: from notice to payment Voice notice, photo assessment, repair-shop assignment and payment order in one flow; the adjuster steps in only above the threshold. 21,000 runs / month 91.2% success 5 agents InsuranceHITL ### Voice claims intake (FNOL) The customer calls and describes it; policy found by plate, cover checked, claim opened and adjuster assigned. 34,000 runs / month 95.8% success 5 agents InsuranceHITL ### Policy renewal offer Policies re-priced, sharp increases go to approval; customer accepts with one tap and the policy is issued. 22,000 runs / month 94.4% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Subscriber care assistant: bill, plan, fault and contract URL: https://www.aginies.com/en/agents/subscriber-care-assistant Language: en Summary: The subscriber calls or writes; bill dispute, plan change and line fault are resolved in BSS and network systems in one session. Multi-topic request over IVR, app chat or WhatsApp → Bill corrected, plan changed, modem repaired remotely, contract terms explained Telecomvoiceappwhatsapp # Subscriber care assistant: bill, plan, fault and contract The subscriber calls or writes; bill dispute, plan change and line fault are resolved in BSS and network systems in one session. In short Subscriber care assistant: bill, plan, fault and contract is an agent scenario in the Telecom sector running on the voice, app, whatsapp channels. Trigger: Multi-topic request over IVR, app chat or WhatsApp. The Subscriber Care Supervisor coordinates 4 specialist agents (Dialogue Agent, Account Agent, Network Agent, Retention Agent) across 5 systems (IVR / SIP, BSS / Billing, OSS Network, CRM, SMS Gateway); of 10 steps, 3 write to systems and 1 are checks or approvals. Outcome: Bill corrected, plan changed, modem repaired remotely, contract terms explained. Approval rule: Contract cancellation, bill correction above 500 or technician priority → representative approval. Telecom ### Subscriber care assistant: bill, plan, fault and contract [Subscriber care assistant: bill, plan, fault and contract] run #41,288,639elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Contract cancellation, bill correction above 500 or technician priority → representative approval Agents in this run Subscriber Care Supervisorsupervisorplans, routes, approves Dialogue Agentaginies-talay-2 · on-premRuns the conversation, separates topics, replies in voice and text Account Agentaginies-sunkar-1 · on-premExplains bill lines, applies corrections and plan changes Network Agentaginies-tulpar-4 · on-premDiagnoses line and modem, decides remote repair or technician dispatch Retention Agentaginies-sunkar-1 · on-premComputes contract and exit terms, selects the eligible offer Audit trail00:00.00 - ◦Call: "my bill doubled, the internet keeps dropping for two days, I want to change my plan" - ∿Three topics: bill dispute · fault · plan change; identity verified by IVR PIN - ⌘Plan: bill lines → correction → line test → repair → contract → plan → summary - ⚙Bill 486: roaming 212 (3 days), no out-of-plan item - ⇢Roaming pack applied retroactively, 212 → 89; difference on the next bill - ⚙Line test: SNR marginal, modem not restarted for 41 days, firmware outdated - ⇢Modem updated and restarted remotely; test: 94 Mbps stable, no technician needed - ✓7 months left on contract; upgrade carries no exit fee, +60 monthly, 3 months discounted - ⇢Plan change confirmed, effective next billing period, ref #PL-99102 - ➤Spoken summary read out, SMS with the three references sent, satisfaction survey triggered - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Multi-topic request over IVR, app chat or WhatsApp Outcome Bill corrected, plan changed, modem repaired remotely, contract terms explained ## Agents SUPERVISOR ### Subscriber Care Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Dialogue Agent Runs the conversation, separates topics, replies in voice and text detect_intentscompose_replytts_speak aginies-sunkar-1 · on-prem ### Account Agent Explains bill lines, applies corrections and plan changes get_billadjust_billchange_planget_commitment aginies-tulpar-4 · on-prem ### Network Agent Diagnoses line and modem, decides remote repair or technician dispatch line_testmodem_resetschedule_technician aginies-sunkar-1 · on-prem ### Retention Agent Computes contract and exit terms, selects the eligible offer calc_exit_feeget_offerspolicy_check ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Subscriber Care Supervisor | Call: "my bill doubled, the internet keeps dropping for two days, I want to change my plan" | IVR / SIP | 200 | — | | 02 | reason | Dialogue Agent | Three topics: bill dispute · fault · plan change; identity verified by IVR PIN | — | 640 | 380 | | 03 | plan | Subscriber Care Supervisor | Plan: bill lines → correction → line test → repair → contract → plan → summary | — | 240 | 220 | | 04 | tool | Account Agent | Bill 486: roaming 212 (3 days), no out-of-plan item | BSS / Billing | 440 | — | | 05 | write | Account Agent | Roaming pack applied retroactively, 212 → 89; difference on the next bill | BSS / Billing | 560 | — | | 06 | tool | Network Agent | Line test: SNR marginal, modem not restarted for 41 days, firmware outdated | OSS Network | 1,600 | — | | 07 | write | Network Agent | Modem updated and restarted remotely; test: 94 Mbps stable, no technician needed | OSS Network | 2,400 | — | | 08 | verify | Retention Agent | 7 months left on contract; upgrade carries no exit fee, +60 monthly, 3 months discounted | CRM | 480 | 260 | | 09 | write | Account Agent | Plan change confirmed, effective next billing period, ref #PL-99102 | BSS / Billing | 420 | — | | 10 | notify | Dialogue Agent | Spoken summary read out, SMS with the three references sent, satisfaction survey triggered | SMS Gateway | 400 | 200 | Simulation metrics (estimated) 410,000runs / month 94.3%success 7.4 sp50 $0.0019est. cost / run Approval rule Contract cancellation, bill correction above 500 or technician priority → representative approval Step profile ingest · 1plan · 1reason · 1tool · 2verify · 1write · 3notify · 1 SYSTEMS - IVR / SIPvoice - BSS / Billingcore - OSS Networkapi - CRMcrm - SMS Gatewaychannel What you need to run it - SIP integration with the switch and recording access - Read/write API or MCP server and a service account for the core system - Credentials and a test environment for the enterprise APIs involved - CRM API access and a customer matching key - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern The transacting version of operators' digital assistants, combined with network diagnostics Packages that include this scenario Voice analysis & quality →Subscriber care and network support → Sector: Telecom → Related agents TelecomHITL ### Bill dispute resolution Invoice lines and usage explained in seconds; goodwill credit applied by policy, larger amounts need a lead. 88,000 runs / month 96.3% success 4 agents TelecomHITL ### SIM swap fraud check Every SIM swap is risk-scored before provisioning; suspicious requests are held and banks are signalled. 41,000 runs / month 97.7% success 4 agents TelecomHITL ### Network outage communications When an outage ticket opens, affected subscribers are found and the approved message sent before they call. 1,400 runs / month 98.5% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Supplier invoice three-way match URL: https://www.aginies.com/en/agents/supplier-invoice-three-way-match Language: en Summary: Invoice, PO and goods receipt matched in SAP automatically; only out-of-tolerance lines reach a buyer. A supplier invoice PDF arrives in the ap@ mailbox → Invoice posted in SAP, payment block lifted, deviation reason in the audit trail Manufacturing & Automotiveemailapibatch # Supplier invoice three-way match Invoice, PO and goods receipt matched in SAP automatically; only out-of-tolerance lines reach a buyer. In short Supplier invoice three-way match is an agent scenario in the Manufacturing & Automotive sector running on the email, api, batch channels. Trigger: A supplier invoice PDF arrives in the ap@ mailbox. The Invoice Supervisor coordinates 3 specialist agents (Capture Agent, Matching Agent, Posting Agent) across 3 systems (SAP S/4HANA, Exchange / M365 Mail, Tolerance Policy Vector Store); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: Invoice posted in SAP, payment block lifted, deviation reason in the audit trail. Approval rule: Price deviation > 2% or > 1,000 TL → buyer approval. Manufacturing & Automotive ### Supplier invoice three-way match [Supplier invoice three-way match] run #41,290,364elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Price deviation > 2% or > 1,000 TL → buyer approval Agents in this run Invoice Supervisorsupervisorplans, routes, approves Capture Agentaginies-konrul-1 · on-premExtracts invoice fields with OCR Matching Agentaginies-tulpar-4 · on-premCompares against PO and goods receipt, applies tolerances Posting Agentaginies-talay-2 · on-premPosts or parks the invoice in SAP Audit trail00:00.00 - ◦Invoice PDF from Demir Metal A.Ş. received in the ap@ mailbox - ∿OCR: invoice DM-2026-4471 · 184,320 TL · 3 lines · PO 4500123877 referenced - ⌘Plan: PO → goods receipt → tolerance check → post or park - ⚙PO 4500123877 pulled: 3 lines, 180,000 TL, vendor 100428 - ⚙Goods receipt 5000871122: 3 lines, 500/500/250 units, delivered 2 Sep - ⌕Tolerance policy: price ±2%, quantity ±0%, freight up to 3,000 TL - ✓Line 2 price +2.4% (> 2% tolerance) → 4,320 TL deviation flagged - ⏸Deviation > 2% and > 1,000 TL → buyer approval requested, approved with a note - ⇢Invoice posted (doc 5105634201), payment block lifted, due 2 Oct - ➤Supplier acknowledgement sent, OCR pages and buyer note added to the audit trail - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger A supplier invoice PDF arrives in the ap@ mailbox Outcome Invoice posted in SAP, payment block lifted, deviation reason in the audit trail ## Agents SUPERVISOR ### Invoice Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Capture Agent Extracts invoice fields with OCR ocr_documentextract_invoice_fields aginies-tulpar-4 · on-prem ### Matching Agent Compares against PO and goods receipt, applies tolerances get_purchase_orderget_goods_receiptsearch_tolerance_policyapply_tolerance_rules aginies-talay-2 · on-prem ### Posting Agent Posts or parks the invoice in SAP post_invoicepark_invoicesend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Invoice Supervisor | Invoice PDF from Demir Metal A.Ş. received in the ap@ mailbox | — | 200 | — | | 02 | reason | Capture Agent | OCR: invoice DM-2026-4471 · 184,320 TL · 3 lines · PO 4500123877 referenced | — | 880 | 720 | | 03 | plan | Invoice Supervisor | Plan: PO → goods receipt → tolerance check → post or park | — | 280 | 190 | | 04 | tool | Matching Agent | PO 4500123877 pulled: 3 lines, 180,000 TL, vendor 100428 | SAP S/4HANA | 430 | — | | 05 | tool | Matching Agent | Goods receipt 5000871122: 3 lines, 500/500/250 units, delivered 2 Sep | SAP S/4HANA | 410 | — | | 06 | retrieve | Matching Agent | Tolerance policy: price ±2%, quantity ±0%, freight up to 3,000 TL | Tolerance Policy Vector Store | 360 | 300 | | 07 | verify | Matching Agent | Line 2 price +2.4% (> 2% tolerance) → 4,320 TL deviation flagged | — | 760 | 890 | | 08 | approval | Invoice Supervisor | Deviation > 2% and > 1,000 TL → buyer approval requested, approved with a note | — | 2,600 | — | | 09 | write | Posting Agent | Invoice posted (doc 5105634201), payment block lifted, due 2 Oct | SAP S/4HANA | 560 | — | | 10 | notify | Posting Agent | Supplier acknowledgement sent, OCR pages and buyer note added to the audit trail | Exchange / M365 Mail | 240 | — | Simulation metrics (estimated) 21,000runs / month 97.1%success 6.7 sp50 $0.0042est. cost / run Approval rule Price deviation > 2% or > 1,000 TL → buyer approval Step profile ingest · 1plan · 1retrieve · 1reason · 1tool · 2verify · 1approval · 1write · 1notify · 1 SYSTEMS - SAP S/4HANAerp - Exchange / M365 Mailemail - Tolerance Policy Vector Storedb What you need to run it - Posting and query API access to ERP/MES - Mailbox access (Microsoft 365 or Google Workspace) - Read access to the database or vector store; a data owner - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the AP automation pattern of large shared-service centres Packages that include this scenario Invoice & document extraction →Financial analysis & operational intelligence →Production and maintenance operations → Sector: Manufacturing & Automotive → Related agents Manufacturing & AutomotiveHITL ### Line stoppage response: from alarm to work order and shift report An MES alarm lands; root cause found from history, spare part checked, work order and plan update ready within minutes. 3,100 runs / month 92.2% success 5 agents Manufacturing & AutomotiveHITL ### Maintenance ticket triage Sensor alarm and technician voice note become one work order, spare parts and a safe stop window included. 3,400 runs / month 95.2% success 5 agents Manufacturing & AutomotiveHITL ### Warranty claim processing Dealer claims validated for coverage, parts and labour in minutes; big amounts go to the regional manager. 8,600 runs / month 96.3% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Supplier security and compliance assessment URL: https://www.aginies.com/en/agents/supplier-security-assessment Language: en Summary: Supplier questionnaires and documents are read, compared against facility security requirements, gaps listed; the qualification decision belongs to the security manager. New supplier qualification request from procurement → Assessment report, gap list and risk class ready; decision recorded Defence & Aerospaceemailwebapi # Supplier security and compliance assessment Supplier questionnaires and documents are read, compared against facility security requirements, gaps listed; the qualification decision belongs to the security manager. In short Supplier security and compliance assessment is an agent scenario in the Defence & Aerospace sector running on the email, web, api channels. Trigger: New supplier qualification request from procurement. The Supplier Security Supervisor coordinates 3 specialist agents (Document Agent, Assessment Agent, Case Agent) across 4 systems (Supplier Management (SRM), Security Requirements (vector), Document Store, E-mail); of 9 steps, 1 write to systems and 2 are checks or approvals. Outcome: Assessment report, gap list and risk class ready; decision recorded. Approval rule: The qualification decision always sits with the facility security manager; the agent only prepares. Defence & Aerospace ### Supplier security and compliance assessment [Supplier security and compliance assessment] run #41,290,328elapsed 0 mstokens 0est. cost $0.000status —Approval rule: The qualification decision always sits with the facility security manager; the agent only prepares Agents in this run Supplier Security Supervisorsupervisorplans, routes, approves Document Agentaginies-konrul-1 · on-premReads questionnaires, certificates and policy documents Assessment Agentaginies-tulpar-4 · on-premCompares answers with the requirement set Case Agentaginies-talay-2 · on-premWrites the report and sends gaps to the supplier Audit trail00:00.00 - ◦Qualification request: composite parts supplier, 3 documents + 84-question questionnaire - ⚙Documents received: ISO 27001 certificate, information security policy, site plan - ∿Certificate valid (2027-03); policy version 2024, access-control section missing - ⌕Requirement set: facility security, personnel security, information security, subcontractor control - ∿84 answers scored: 71 compliant, 9 partial, 4 missing; risk class: medium - ✓Critical control: no subcontractor disclosure → mandatory before qualification - ⏸Medium risk + critical gap → facility security manager approved conditional qualification - ⇢Assessment SA-2026-117 recorded; supplier status “conditional”, 30-day window - ➤Gap list and evidence templates sent to the supplier; procurement informed - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger New supplier qualification request from procurement Outcome Assessment report, gap list and risk class ready; decision recorded ## Agents SUPERVISOR ### Supplier Security Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Document Agent Reads questionnaires, certificates and policy documents extract_documentclassify_documentcheck_certificate_validity aginies-tulpar-4 · on-prem ### Assessment Agent Compares answers with the requirement set search_requirementsscore_controllist_gaps aginies-talay-2 · on-prem ### Case Agent Writes the report and sends gaps to the supplier create_assessmentupdate_supplier_statussend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Supplier Security Supervisor | Qualification request: composite parts supplier, 3 documents + 84-question questionnaire | — | 200 | — | | 02 | tool | Document Agent | Documents received: ISO 27001 certificate, information security policy, site plan | Document Store | 460 | — | | 03 | reason | Document Agent | Certificate valid (2027-03); policy version 2024, access-control section missing | — | 900 | 1,200 | | 04 | retrieve | Assessment Agent | Requirement set: facility security, personnel security, information security, subcontractor control | Security Requirements (vector) | 520 | 640 | | 05 | reason | Assessment Agent | 84 answers scored: 71 compliant, 9 partial, 4 missing; risk class: medium | — | 1,100 | 1,800 | | 06 | verify | Assessment Agent | Critical control: no subcontractor disclosure → mandatory before qualification | — | 380 | 260 | | 07 | approval | Supplier Security Supervisor | Medium risk + critical gap → facility security manager approved conditional qualification | — | 3,000 | — | | 08 | write | Case Agent | Assessment SA-2026-117 recorded; supplier status “conditional”, 30-day window | Supplier Management (SRM) | 480 | — | | 09 | notify | Case Agent | Gap list and evidence templates sent to the supplier; procurement informed | E-mail | 260 | — | Simulation metrics (estimated) 900runs / month 95.2%success 7.8 sp50 $0.0147est. cost / run Approval rule The qualification decision always sits with the facility security manager; the agent only prepares Step profile ingest · 1retrieve · 1reason · 2tool · 1verify · 1approval · 1write · 1notify · 1 SYSTEMS - Supplier Management (SRM)erp - Security Requirements (vector)db - Document Storestorage - E-mailemail What you need to run it - Posting and query API access to ERP/MES - Read access to the database or vector store; a data owner - Read access to the document store and the document types - Mailbox access (Microsoft 365 or Google Workspace) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors supply-chain security programmes of defence primes: questionnaire, documents, human decision Packages that include this scenario Public procurement & tender compliance →Defence export control & supply-chain security → Sector: Defence & Aerospace → Related agents Defence & AerospaceHITL ### Technical publication assistant: maintenance manual, parts, procedure and revision control The technician asks in the hangar; the right-revision procedure, part number and safety note arrive with clearance checks, every access traced. 9,800 runs / month 95.7% success 5 agents Defence & AerospaceHITL ### Export control and end-user screening Part, buyer and end use are classified in one flow; licence need is detected, the decision always stays with the compliance officer. 2,600 runs / month 96.1% success 4 agents Defence & AerospaceHITL ### Platform maintenance (MRO) work order preparation The fault report is matched to technical publications, task card and parts list are produced, the work order opens with planner approval. 4,100 runs / month 95.7% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Technical publication assistant: maintenance manual, parts, procedure and revision control URL: https://www.aginies.com/en/agents/technical-publication-assistant Language: en Summary: The technician asks in the hangar; the right-revision procedure, part number and safety note arrive with clearance checks, every access traced. Maintenance question from the tablet app on the closed network → Procedure and part information given, revision verified, linked to the work package, access logged Defence & Aerospaceapp # Technical publication assistant: maintenance manual, parts, procedure and revision control The technician asks in the hangar; the right-revision procedure, part number and safety note arrive with clearance checks, every access traced. In short Technical publication assistant: maintenance manual, parts, procedure and revision control is an agent scenario in the Defence & Aerospace sector running on the app channel. Trigger: Maintenance question from the tablet app on the closed network. The Technical Publication Supervisor coordinates 4 specialist agents (Clearance Agent, Publication Agent, Procedure Agent, Parts Agent) across 5 systems (Hangar Tablet App, Access Control, Tech Pub Repository (S1000D), MRO System, Audit Log); of 10 steps, 2 write to systems and 2 are checks or approvals. Outcome: Procedure and part information given, revision verified, linked to the work package, access logged. Approval rule: Proposed deviation from the procedure → authorised engineer approval; export-controlled content → clearance required. Defence & Aerospace ### Technical publication assistant: maintenance manual, parts, procedure and revision control [Technical publication assistant: maintenance manual, parts, procedure and revision control] run #41,286,632elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Proposed deviation from the procedure → authorised engineer approval; export-controlled content → clearance required Agents in this run Technical Publication Supervisorsupervisorplans, routes, approves Clearance Agentaginies-sunkar-1 · on-premCompares the user's clearance with the content's export-control marking Publication Agentaginies-sunkar-1 · on-premFinds the right revision across S1000D data modules Procedure Agentaginies-tulpar-4 · on-premInterprets the procedure for the platform configuration, surfaces safety notes Parts Agentaginies-sunkar-1 · on-premFetches the part number, alternates and stock from the MRO system Audit trail00:00.00 - ◦Tablet: "tail no 07-14, hydraulic pump replacement, torque values and safety steps?" - ✓User clearance: level 2; module marking level 2, not export-controlled → access permitted - ⌘Plan: applicability → current revision → configuration → procedure → parts → work package → trace - ⌕Data module DM-29-10-05 found; revision 14 (3 months ago), applicable to tail 07-14 - ✓Difference from revision 13: torque value updated, one new warning added → change summary prepared - ∿Procedure adapted to the 07-14 configuration: 11 steps, 3 WARNINGS, 2 CAUTIONS; torque 48 ± 2 Nm - ⚙Pump P/N 7742-A2: 1 in stock, alternate 7742-A3 (bulletin required); seals in stock - ⇢Part and procedure linked to work package #WP-3391; QC sign-off step added - ➤To the tablet: step list with warnings highlighted, revision difference and part card - ⇢Access record: user, module, revision, time and work package written to the audit log - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Maintenance question from the tablet app on the closed network Outcome Procedure and part information given, revision verified, linked to the work package, access logged ## Agents SUPERVISOR ### Technical Publication Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-sunkar-1 · on-prem ### Clearance Agent Compares the user's clearance with the content's export-control marking get_clearancecheck_markinglog_access aginies-sunkar-1 · on-prem ### Publication Agent Finds the right revision across S1000D data modules search_modulesget_revisionget_applicability aginies-tulpar-4 · on-prem ### Procedure Agent Interprets the procedure for the platform configuration, surfaces safety notes apply_configurationextract_warningscompose_answer aginies-sunkar-1 · on-prem ### Parts Agent Fetches the part number, alternates and stock from the MRO system lookup_partget_alternatescheck_stock ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Technical Publication Supervisor | Tablet: "tail no 07-14, hydraulic pump replacement, torque values and safety steps?" | Hangar Tablet App | 160 | — | | 02 | verify | Clearance Agent | User clearance: level 2; module marking level 2, not export-controlled → access permitted | Access Control | 380 | — | | 03 | plan | Technical Publication Supervisor | Plan: applicability → current revision → configuration → procedure → parts → work package → trace | — | 240 | 220 | | 04 | retrieve | Publication Agent | Data module DM-29-10-05 found; revision 14 (3 months ago), applicable to tail 07-14 | Tech Pub Repository (S1000D) | 900 | 320 | | 05 | verify | Publication Agent | Difference from revision 13: torque value updated, one new warning added → change summary prepared | — | 700 | 380 | | 06 | reason | Procedure Agent | Procedure adapted to the 07-14 configuration: 11 steps, 3 WARNINGS, 2 CAUTIONS; torque 48 ± 2 Nm | — | 2,200 | 1,500 | | 07 | tool | Parts Agent | Pump P/N 7742-A2: 1 in stock, alternate 7742-A3 (bulletin required); seals in stock | MRO System | 520 | — | | 08 | write | Parts Agent | Part and procedure linked to work package #WP-3391; QC sign-off step added | MRO System | 480 | — | | 09 | notify | Procedure Agent | To the tablet: step list with warnings highlighted, revision difference and part card | Hangar Tablet App | 420 | 260 | | 10 | write | Clearance Agent | Access record: user, module, revision, time and work package written to the audit log | Audit Log | 200 | — | Simulation metrics (estimated) 9,800runs / month 95.7%success 6.2 sp50 $0.0024est. cost / run Approval rule Proposed deviation from the procedure → authorised engineer approval; export-controlled content → clearance required Step profile ingest · 1plan · 1retrieve · 1reason · 1tool · 1verify · 2write · 2notify · 1 SYSTEMS - Hangar Tablet Appchannel - Access Controlapi - Tech Pub Repository (S1000D)storage - MRO Systemerp - Audit Logdb What you need to run it - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - Credentials and a test environment for the enterprise APIs involved - Read access to the document store and the document types - Posting and query API access to ERP/MES - Read access to the database or vector store; a data owner - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Aerospace and defence MROs' S1000D-based intelligent technical publication pattern, on a closed network Packages that include this scenario Defence export control & supply-chain security → Sector: Defence & Aerospace → Related agents Defence & AerospaceHITL ### Export control and end-user screening Part, buyer and end use are classified in one flow; licence need is detected, the decision always stays with the compliance officer. 2,600 runs / month 96.1% success 4 agents Defence & AerospaceHITL ### Platform maintenance (MRO) work order preparation The fault report is matched to technical publications, task card and parts list are produced, the work order opens with planner approval. 4,100 runs / month 95.7% success 5 agents Defence & AerospaceHITL ### Supplier security and compliance assessment Supplier questionnaires and documents are read, compared against facility security requirements, gaps listed; the qualification decision belongs to the security manager. 900 runs / month 95.2% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Underwriting document extraction URL: https://www.aginies.com/en/agents/underwriting-document-extraction Language: en Summary: Medical reports and labs read and assessed against the manual; underwriters only see files above threshold. Life insurance application with attachments emailed by an agent → Decision and loading draft on the workbench, rationale and extracted fields on file Insuranceemailwebbatch # Underwriting document extraction Medical reports and labs read and assessed against the manual; underwriters only see files above threshold. In short Underwriting document extraction is an agent scenario in the Insurance sector running on the email, web, batch channels. Trigger: Life insurance application with attachments emailed by an agent. The Underwriting Supervisor coordinates 3 specialist agents (Document Agent, Medical Assessment Agent, Underwriting Agent) across 4 systems (Underwriting Workbench, Document Store (S3), Vector Store, Email (Exchange)); of 9 steps, 1 write to systems and 2 are checks or approvals. Outcome: Decision and loading draft on the workbench, rationale and extracted fields on file. Approval rule: Loading ≥ 50% or sum assured > 5,000,000 → senior underwriter approval. Insurance ### Underwriting document extraction [Underwriting document extraction] run #41,287,441elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Loading ≥ 50% or sum assured > 5,000,000 → senior underwriter approval Agents in this run Underwriting Supervisorsupervisorplans, routes, approves Document Agentaginies-konrul-1 · on-premClassifies pages, extracts fields and lab tables Medical Assessment Agentaginies-tulpar-4 · on-premMaps findings to ICD codes and summarises medical history Underwriting Agentaginies-talay-2 · on-premRetrieves the manual, applies rules, updates the application Audit trail00:00.00 - ◦Email: life application #UW-55120 with 6 attachments (medical report, labs, declaration) - ⌕Attachments stored in S3; 14 pages classified, 32-row lab table extracted (98.4%) - ⌘Plan: medical summary → underwriting manual → rule set → approval → decision - ∿Findings: HbA1c 6.9 · BMI 31 · on blood-pressure medication → ICD E11, I10 - ⌕Underwriting manual v7.2: type 2 diabetes and hypertension sections retrieved - ✓Rule: HbA1c 6.5–7.5 + controlled BP → +75% loading, no decline - ⏸Loading ≥ 50% → senior underwriter approved the decision - ⇢Application set to "accepted with loading", extracted fields and rationale attached - ➤Decision summary and loaded-premium offer emailed to the agent - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Life insurance application with attachments emailed by an agent Outcome Decision and loading draft on the workbench, rationale and extracted fields on file ## Agents SUPERVISOR ### Underwriting Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Document Agent Classifies pages, extracts fields and lab tables classify_documentextract_fieldsextract_tables aginies-tulpar-4 · on-prem ### Medical Assessment Agent Maps findings to ICD codes and summarises medical history map_icd_codessummarise_historyscore_medical_risk aginies-talay-2 · on-prem ### Underwriting Agent Retrieves the manual, applies rules, updates the application get_uw_manualapply_uw_rulesupdate_applicationsend_email ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Underwriting Supervisor | Email: life application #UW-55120 with 6 attachments (medical report, labs, declaration) | — | 220 | — | | 02 | retrieve | Document Agent | Attachments stored in S3; 14 pages classified, 32-row lab table extracted (98.4%) | Document Store (S3) | 1,600 | 3,100 | | 03 | plan | Underwriting Supervisor | Plan: medical summary → underwriting manual → rule set → approval → decision | — | 280 | 200 | | 04 | reason | Medical Assessment Agent | Findings: HbA1c 6.9 · BMI 31 · on blood-pressure medication → ICD E11, I10 | — | 940 | 1,120 | | 05 | retrieve | Underwriting Agent | Underwriting manual v7.2: type 2 diabetes and hypertension sections retrieved | Vector Store | 360 | — | | 06 | verify | Underwriting Agent | Rule: HbA1c 6.5–7.5 + controlled BP → +75% loading, no decline | — | 720 | 860 | | 07 | approval | Underwriting Supervisor | Loading ≥ 50% → senior underwriter approved the decision | — | 3,000 | — | | 08 | write | Underwriting Agent | Application set to "accepted with loading", extracted fields and rationale attached | Underwriting Workbench | 480 | — | | 09 | notify | Underwriting Agent | Decision summary and loaded-premium offer emailed to the agent | Email (Exchange) | 240 | — | Simulation metrics (estimated) 4,800runs / month 94.9%success 7.8 sp50 $0.0128est. cost / run Approval rule Loading ≥ 50% or sum assured > 5,000,000 → senior underwriter approval Step profile ingest · 1plan · 1retrieve · 2reason · 1verify · 1approval · 1write · 1notify · 1 SYSTEMS - Underwriting Workbenchcore - Document Store (S3)storage - Vector Storedb - Email (Exchange)email What you need to run it - Read/write API or MCP server and a service account for the core system - Read access to the document store and the document types - Read access to the database or vector store; a data owner - Mailbox access (Microsoft 365 or Google Workspace) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the document-extraction pattern global life insurers use to raise straight-through underwriting rates Packages that include this scenario Invoice & document extraction →KVKK / GDPR personal data protection →Underwriting and renewal operations → Sector: Insurance → Related agents InsuranceHITL ### End-to-end motor claim: from notice to payment Voice notice, photo assessment, repair-shop assignment and payment order in one flow; the adjuster steps in only above the threshold. 21,000 runs / month 91.2% success 5 agents InsuranceHITL ### Voice claims intake (FNOL) The customer calls and describes it; policy found by plate, cover checked, claim opened and adjuster assigned. 34,000 runs / month 95.8% success 5 agents InsuranceHITL ### Policy renewal offer Policies re-priced, sharp increases go to approval; customer accepts with one tap and the policy is issued. 22,000 runs / month 94.4% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Utility customer assistant: bill, outage, consumption and instalments in one dialogue URL: https://www.aginies.com/en/agents/utility-customer-assistant Language: en Summary: The customer calls or writes; the high bill is explained, outage ETA given, meter data shown, an instalment plan set up. Multi-topic request over IVR, WhatsApp or the customer app → Bill explained, outage information given, instalment plan active, reminder set Energy & Utilitiesvoicewhatsappapp # Utility customer assistant: bill, outage, consumption and instalments in one dialogue The customer calls or writes; the high bill is explained, outage ETA given, meter data shown, an instalment plan set up. In short Utility customer assistant: bill, outage, consumption and instalments in one dialogue is an agent scenario in the Energy & Utilities sector running on the voice, whatsapp, app channels. Trigger: Multi-topic request over IVR, WhatsApp or the customer app. The Customer Service Supervisor coordinates 4 specialist agents (Dialogue Agent, Billing Agent, Grid Agent, Collections Agent) across 5 systems (IVR / SIP, Billing (CIS), Outage Management, Smart Meter Data, SMS Gateway); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: Bill explained, outage information given, instalment plan active, reminder set. Approval rule: Instalments > 6 months, bill cancellation or meter suspicion → customer service approval. Energy & Utilities ### Utility customer assistant: bill, outage, consumption and instalments in one dialogue [Utility customer assistant: bill, outage, consumption and instalments in one dialogue] run #41,289,293elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Instalments > 6 months, bill cancellation or meter suspicion → customer service approval Agents in this run Customer Service Supervisorsupervisorplans, routes, approves Dialogue Agentaginies-talay-2 · on-premVerifies the customer, separates topics, writes plain replies Billing Agentaginies-sunkar-1 · on-premExplains bill lines and consumption profile, applies corrections Grid Agentaginies-tulpar-4 · on-premGets outage record and ETA from the OMS, reports planned works Collections Agentaginies-sunkar-1 · on-premApplies the instalment policy, sets up the plan, schedules reminders Audit trail00:00.00 - ◦Call: "why is my bill triple, the neighbourhood has no power, can I split the payment?" - ✓Verified by customer number and registered phone; three topics separated - ⌘Plan: outage status → bill analysis → meter data → instalments → spoken summary - ⚙OMS: transformer fault 11:20, 1,260 customers affected, crew on site, ETA 13:45 - ⚙Bill 2,140: 1,980 kWh, same period last year 690 kWh; actual reading, not estimated - ⌕Meter series: night consumption 4× since 14 August; steady load profile (water heater / AC) - ∿Conclusion: bill correct, consumption real; dated hourly consumption summary prepared for the customer - ✓Instalment policy: 2,140 → 4 interest-free instalments eligible; no prior arrears - ⇢Instalment plan created, first instalment in 15 days; reminder SMS scheduled - ➤Spoken summary: outage 13:45, consumption explanation, instalment plan; SMS with consumption chart link - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger Multi-topic request over IVR, WhatsApp or the customer app Outcome Bill explained, outage information given, instalment plan active, reminder set ## Agents SUPERVISOR ### Customer Service Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-talay-2 · on-prem ### Dialogue Agent Verifies the customer, separates topics, writes plain replies verify_customerdetect_intentscompose_replytts_speak aginies-sunkar-1 · on-prem ### Billing Agent Explains bill lines and consumption profile, applies corrections get_billget_meter_seriescompare_periodsadjust_bill aginies-tulpar-4 · on-prem ### Grid Agent Gets outage record and ETA from the OMS, reports planned works outage_lookupget_etaplanned_works aginies-sunkar-1 · on-prem ### Collections Agent Applies the instalment policy, sets up the plan, schedules reminders policy_checkcreate_instalment_planschedule_reminder ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Customer Service Supervisor | Call: "why is my bill triple, the neighbourhood has no power, can I split the payment?" | IVR / SIP | 200 | — | | 02 | verify | Dialogue Agent | Verified by customer number and registered phone; three topics separated | — | 600 | 320 | | 03 | plan | Customer Service Supervisor | Plan: outage status → bill analysis → meter data → instalments → spoken summary | — | 240 | 220 | | 04 | tool | Grid Agent | OMS: transformer fault 11:20, 1,260 customers affected, crew on site, ETA 13:45 | Outage Management | 520 | — | | 05 | tool | Billing Agent | Bill 2,140: 1,980 kWh, same period last year 690 kWh; actual reading, not estimated | Billing (CIS) | 480 | — | | 06 | retrieve | Billing Agent | Meter series: night consumption 4× since 14 August; steady load profile (water heater / AC) | Smart Meter Data | 1,100 | 540 | | 07 | reason | Billing Agent | Conclusion: bill correct, consumption real; dated hourly consumption summary prepared for the customer | — | 900 | 480 | | 08 | verify | Collections Agent | Instalment policy: 2,140 → 4 interest-free instalments eligible; no prior arrears | — | 420 | 180 | | 09 | write | Collections Agent | Instalment plan created, first instalment in 15 days; reminder SMS scheduled | Billing (CIS) | 560 | — | | 10 | notify | Dialogue Agent | Spoken summary: outage 13:45, consumption explanation, instalment plan; SMS with consumption chart link | SMS Gateway | 440 | 240 | Simulation metrics (estimated) 520,000runs / month 94.6%success 6.4 sp50 $0.0015est. cost / run Approval rule Instalments > 6 months, bill cancellation or meter suspicion → customer service approval Step profile ingest · 1plan · 1retrieve · 1reason · 1tool · 2verify · 2write · 1notify · 1 SYSTEMS - IVR / SIPvoice - Billing (CIS)core - Outage Managementapi - Smart Meter Datadb - SMS Gatewaychannel What you need to run it - SIP integration with the switch and recording access - Read/write API or MCP server and a service account for the core system - Credentials and a test environment for the enterprise APIs involved - Read access to the database or vector store; a data owner - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Distribution companies' digital customer centre pattern, with smart-meter-based bill explanation Packages that include this scenario Voice AI for claims intake →Utility customer operations → Sector: Energy & Utilities → Related agents Energy & UtilitiesHITL ### Outage report handling Reports are clustered by feeder, one OSS ticket is raised and everyone affected hears before the crew leaves. 38,000 runs / month 97.7% success 5 agents Energy & UtilitiesHITL ### Meter reading dispute Customer sends a meter photo; the index is read by vision, the estimated bill fixed, big credits approved. 9,200 runs / month 94.9% success 4 agents Energy & UtilitiesHITL ### Rooftop solar connection application Application documents are read, transformer capacity and regulation checked, gaps reported the same day; the connection opinion belongs to the engineer. 5,400 runs / month 95.3% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Voice banking money transfer URL: https://www.aginies.com/en/agents/voice-banking-transfer Language: en Summary: Customer calls and talks; voiceprint, security question and OTP verify them; transfer completes agent-free. "I want to send money" call arriving at the call centre over SIP → Transfer executed in core banking, spoken and SMS confirmation given, recording archived Bankingvoice # Voice banking money transfer Customer calls and talks; voiceprint, security question and OTP verify them; transfer completes agent-free. In short Voice banking money transfer is an agent scenario in the Banking sector running on the voice channel. Trigger: "I want to send money" call arriving at the call centre over SIP. The Voice Transaction Supervisor coordinates 4 specialist agents (Speech Agent, Identity Agent, Transfer Agent, Dialogue Agent) across 4 systems (Genesys SIP, Core Banking, Fraud Engine, SMS Gateway); of 10 steps, 1 write to systems and 2 are checks or approvals. Outcome: Transfer executed in core banking, spoken and SMS confirmation given, recording archived. Approval rule: Amount > 50,000 or new payee → warm transfer to a call-centre agent. Banking ### Voice banking money transfer [Voice banking money transfer] run #41,289,138elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Amount > 50,000 or new payee → warm transfer to a call-centre agent Agents in this run Voice Transaction Supervisorsupervisorplans, routes, approves Speech Agentaginies-voice-tr-v3 · on-premTranscribes live audio, detects language and confidence Identity Agentaginies-sunkar-1 · on-premVerifies identity by voiceprint, security question and OTP Transfer Agentaginies-talay-2 · on-premResolves the payee, checks limits and fraud, executes the transfer Dialogue Agentaginies-tulpar-4 · on-premExtracts intent, confirms details, produces spoken replies Audit trail00:00.00 - ◦Call connected over SIP, caller ID matched to a customer record - ∿Transcript: "send 12,500 lira to Ayşe Demir" · language: TR · confidence 0.96 - ✓Voiceprint match 94% + security question correct → identity verified - ⌘Plan: resolve payee → limit/fraud → OTP → transfer → spoken confirmation - ⚙Saved payee "Ayşe Demir" TR12 0001 …8834 found, daily limit 50,000 - ⚙Fraud engine: score 12/100, payee registered for 8 months - ✓Policy: amount ≤ 50,000 and payee saved → eligible, OTP mandatory - ➤OTP sent by SMS, customer read the 6-digit code aloud → verified - ⇢12,500 transfer executed in core banking, ref #TR-5583021 - ➤Spoken confirmation read out, summary SMS sent, call recording archived - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger "I want to send money" call arriving at the call centre over SIP Outcome Transfer executed in core banking, spoken and SMS confirmation given, recording archived ## Agents SUPERVISOR ### Voice Transaction Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-voice-tr-v3 · on-prem ### Speech Agent Transcribes live audio, detects language and confidence transcribe_streamdetect_language aginies-sunkar-1 · on-prem ### Identity Agent Verifies identity by voiceprint, security question and OTP verify_voiceprintask_security_questionsend_otpverify_otp aginies-talay-2 · on-prem ### Transfer Agent Resolves the payee, checks limits and fraud, executes the transfer get_accountsresolve_beneficiarycheck_limitsget_fraud_scoreexecute_transfer aginies-tulpar-4 · on-prem ### Dialogue Agent Extracts intent, confirms details, produces spoken replies extract_intentconfirm_detailssynthesize_speech ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Voice Transaction Supervisor | Call connected over SIP, caller ID matched to a customer record | — | 200 | — | | 02 | reason | Speech Agent | Transcript: "send 12,500 lira to Ayşe Demir" · language: TR · confidence 0.96 | — | 540 | 180 | | 03 | verify | Identity Agent | Voiceprint match 94% + security question correct → identity verified | — | 900 | 260 | | 04 | plan | Voice Transaction Supervisor | Plan: resolve payee → limit/fraud → OTP → transfer → spoken confirmation | — | 280 | 200 | | 05 | tool | Transfer Agent | Saved payee "Ayşe Demir" TR12 0001 …8834 found, daily limit 50,000 | Core Banking | 380 | — | | 06 | tool | Transfer Agent | Fraud engine: score 12/100, payee registered for 8 months | Fraud Engine | 360 | — | | 07 | verify | Transfer Agent | Policy: amount ≤ 50,000 and payee saved → eligible, OTP mandatory | — | 420 | 300 | | 08 | notify | Identity Agent | OTP sent by SMS, customer read the 6-digit code aloud → verified | SMS Gateway | 1,800 | — | | 09 | write | Transfer Agent | 12,500 transfer executed in core banking, ref #TR-5583021 | Core Banking | 560 | — | | 10 | notify | Dialogue Agent | Spoken confirmation read out, summary SMS sent, call recording archived | Genesys SIP | 420 | 160 | Simulation metrics (estimated) 27,000runs / month 93.1%success 5.9 sp50 $0.0033est. cost / run Approval rule Amount > 50,000 or new payee → warm transfer to a call-centre agent Step profile ingest · 1plan · 1reason · 1tool · 2verify · 2write · 1notify · 2 SYSTEMS - Genesys SIPvoice - Core Bankingcore - Fraud Engineapi - SMS Gatewaychannel What you need to run it - SIP integration with the switch and recording access - Read/write API or MCP server and a service account for the core system - Credentials and a test environment for the enterprise APIs involved - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the conversational IVR-replacement pattern at large banks, extended to execute transactions Packages that include this scenario Conversational banking → Sector: Banking → Related agents BankingHITL ### Dialogue banking: voice and text transaction assistant The customer talks or types; bill, limit, dispute and transfer are completed in core banking in one session. 186,000 runs / month 94.7% success 5 agents BankingHITL ### Credit card limit increase The customer types in the app or talks to the call centre; the decision takes seconds, and anything above threshold goes to a manager. 18,400 runs / month 96.4% success 4 agents BankingHITL ### Digital account opening (KYC) ID, liveness and sanctions checks in one flow; clean applications open an account in minutes, no branch visit. 42,000 runs / month 93.8% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Voice claims intake (FNOL) URL: https://www.aginies.com/en/agents/voice-claims-intake Language: en Summary: The customer calls and describes it; policy found by plate, cover checked, claim opened and adjuster assigned. "Someone hit my car" call on the claims line → Claim opened, reserve set, adjuster appointment and photo link with the customer Insurancevoicewhatsapp # Voice claims intake (FNOL) The customer calls and describes it; policy found by plate, cover checked, claim opened and adjuster assigned. In short Voice claims intake (FNOL) is an agent scenario in the Insurance sector running on the voice, whatsapp channels. Trigger: "Someone hit my car" call on the claims line. The Claims Supervisor coordinates 4 specialist agents (Speech Agent, Intake Agent, Policy Agent, Claims Registration Agent) across 4 systems (Genesys SIP, Policy Admin System, Guidewire ClaimCenter, WhatsApp Business API); of 10 steps, 2 write to systems and 1 are checks or approvals. Outcome: Claim opened, reserve set, adjuster appointment and photo link with the customer. Approval rule: Injury or reserve > 250,000 → warm transfer to a claims specialist. Insurance ### Voice claims intake (FNOL) [Voice claims intake (FNOL)] run #41,288,372elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Injury or reserve > 250,000 → warm transfer to a claims specialist Agents in this run Claims Supervisorsupervisorplans, routes, approves Speech Agentaginies-voice-tr-v3 · on-premTranscribes live audio and filters noise Intake Agentaginies-tulpar-4 · on-premGathers incident facts, asks follow-ups, gives a spoken summary Policy Agentaginies-talay-2 · on-premFinds the policy, checks cover and claim history Claims Registration Agentaginies-sunkar-1 · on-premOpens the claim, sets the reserve, assigns the adjuster Audit trail00:00.00 - ◦Call connected: "someone hit my car last night" · plate 34 ABC 123 - ∿Live transcript with noise filtering; tone: calm, confidence 0.95 - ⚙Policy found by plate: comprehensive #KS-2291045, active, excess 2,000 - ⌘Plan: gather incident facts → coverage → open claim → assign adjuster - ∿Incident: hit while parked, 7 Sep 23:30, Kadıköy · no injuries · third party unknown - ✓Cover: collision within comprehensive · 0 claims in 12 months · no waiting period - ⇢Claim #CL-2026-118342 opened, reserve set at 45,000 - ⇢Nearest partner adjuster (Kadıköy) assigned, appointment tomorrow 10:00 - ➤Spoken summary read out, customer confirmed the claim number - ➤Claim number and photo upload link sent on WhatsApp - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger "Someone hit my car" call on the claims line Outcome Claim opened, reserve set, adjuster appointment and photo link with the customer ## Agents SUPERVISOR ### Claims Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-voice-tr-v3 · on-prem ### Speech Agent Transcribes live audio and filters noise transcribe_streamdetect_emotion aginies-tulpar-4 · on-prem ### Intake Agent Gathers incident facts, asks follow-ups, gives a spoken summary extract_claim_factsask_followupsynthesize_speech aginies-talay-2 · on-prem ### Policy Agent Finds the policy, checks cover and claim history get_policy_by_platecheck_coverageget_claim_history aginies-sunkar-1 · on-prem ### Claims Registration Agent Opens the claim, sets the reserve, assigns the adjuster create_claimset_reserveassign_adjustersend_whatsapp ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Claims Supervisor | Call connected: "someone hit my car last night" · plate 34 ABC 123 | — | 200 | — | | 02 | reason | Speech Agent | Live transcript with noise filtering; tone: calm, confidence 0.95 | — | 480 | 170 | | 03 | tool | Policy Agent | Policy found by plate: comprehensive #KS-2291045, active, excess 2,000 | Policy Admin System | 420 | — | | 04 | plan | Claims Supervisor | Plan: gather incident facts → coverage → open claim → assign adjuster | — | 280 | 210 | | 05 | reason | Intake Agent | Incident: hit while parked, 7 Sep 23:30, Kadıköy · no injuries · third party unknown | — | 1,100 | 1,240 | | 06 | verify | Policy Agent | Cover: collision within comprehensive · 0 claims in 12 months · no waiting period | — | 560 | 480 | | 07 | write | Claims Registration Agent | Claim #CL-2026-118342 opened, reserve set at 45,000 | Guidewire ClaimCenter | 620 | — | | 08 | write | Claims Registration Agent | Nearest partner adjuster (Kadıköy) assigned, appointment tomorrow 10:00 | Guidewire ClaimCenter | 380 | — | | 09 | notify | Intake Agent | Spoken summary read out, customer confirmed the claim number | Genesys SIP | 380 | 160 | | 10 | notify | Claims Registration Agent | Claim number and photo upload link sent on WhatsApp | WhatsApp Business API | 240 | — | Simulation metrics (estimated) 34,000runs / month 95.8%success 4.7 sp50 $0.0026est. cost / run Approval rule Injury or reserve > 250,000 → warm transfer to a claims specialist Step profile ingest · 1plan · 1reason · 2tool · 1verify · 1write · 2notify · 2 SYSTEMS - Genesys SIPvoice - Policy Admin Systemcore - Guidewire ClaimCentercore - WhatsApp Business APIchannel What you need to run it - SIP integration with the switch and recording access - Read/write API or MCP server and a service account for the core system - Channel integration (WhatsApp Business API, SMS gateway or in-app widget) - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors the FNOL automation pattern used by large US/EU carriers to cut call-centre handling load Packages that include this scenario Voice AI for claims intake →Voice analysis & quality →Claims operations end to end →Underwriting and renewal operations → Sector: Insurance → Related agents InsuranceHITL ### End-to-end motor claim: from notice to payment Voice notice, photo assessment, repair-shop assignment and payment order in one flow; the adjuster steps in only above the threshold. 21,000 runs / month 91.2% success 5 agents InsuranceHITL ### Policy renewal offer Policies re-priced, sharp increases go to approval; customer accepts with one tap and the policy is issued. 22,000 runs / month 94.4% success 4 agents InsuranceHITL ### Underwriting document extraction Medical reports and labs read and assessed against the manual; underwriters only see files above threshold. 4,800 runs / month 94.9% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Warranty claim processing URL: https://www.aginies.com/en/agents/warranty-claim-processing Language: en Summary: Dealer claims validated for coverage, parts and labour in minutes; big amounts go to the regional manager. A warranty claim with photos is submitted on the dealer portal → Claim approved with adjustment, credit memo posted in SAP, dealer informed with reasons Manufacturing & Automotivewebapiemail # Warranty claim processing Dealer claims validated for coverage, parts and labour in minutes; big amounts go to the regional manager. In short Warranty claim processing is an agent scenario in the Manufacturing & Automotive sector running on the web, api, email channels. Trigger: A warranty claim with photos is submitted on the dealer portal. The Warranty Supervisor coordinates 3 specialist agents (Claim Agent, Validation Agent, Settlement Agent) across 4 systems (Dealer Portal, SAP S/4HANA, Warranty Policy Vector Store, Document Store (S3)); of 10 steps, 3 write to systems and 2 are checks or approvals. Outcome: Claim approved with adjustment, credit memo posted in SAP, dealer informed with reasons. Approval rule: Warranty claim > 5,000 € → regional warranty manager approval. Manufacturing & Automotive ### Warranty claim processing [Warranty claim processing] run #41,288,224elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Warranty claim > 5,000 € → regional warranty manager approval Agents in this run Warranty Supervisorsupervisorplans, routes, approves Claim Agentaginies-konrul-1 · on-premReads claim fields and damage photos Validation Agentaginies-tulpar-4 · on-premValidates coverage, part price and labour standard Settlement Agentaginies-talay-2 · on-premPosts the credit memo and updates the dealer record Audit trail00:00.00 - ◦Claim WC-2026-30917 submitted by dealer D-114 with 4 photos - ∿VIN WBA…7712 · 31,400 km · part: turbo actuator · labour 2.4 h · 6,850 € - ⚙Coverage: 3 years / 100,000 km, valid until 03/2027 → in scope - ⌕Policy WP-14: actuator labour standard 1.8 h, part price list 2026-Q3 - ✓Labour 2.4 h vs standard 1.8 h (+33%) → reduced; part price and photos consistent - ⇢Photos and diagnostic log archived under the claim folder - ⏸Adjusted claim 6,100 € > 5,000 € → regional warranty manager approved - ⇢Credit memo 1700442310 posted for 6,100 € to the dealer account - ⇢Claim status set to "approved – adjusted", reason code L02 (labour) - ➤Dealer notified in the portal with the adjustment breakdown, audit trail stored - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus Trigger A warranty claim with photos is submitted on the dealer portal Outcome Claim approved with adjustment, credit memo posted in SAP, dealer informed with reasons ## Agents SUPERVISOR ### Warranty Supervisor Decomposes the objective, delegates to agents, manages approval points, merges the result. aginies-konrul-1 · on-prem ### Claim Agent Reads claim fields and damage photos extract_claim_fieldsanalyse_damage_photosarchive_documents aginies-tulpar-4 · on-prem ### Validation Agent Validates coverage, part price and labour standard get_vehicle_coverageget_labour_standardcheck_parts_cataloguesearch_warranty_policy aginies-talay-2 · on-prem ### Settlement Agent Posts the credit memo and updates the dealer record post_credit_memoupdate_claim_statussend_dealer_message ## Steps | # | Kind | Agent | What happens | System | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | ingest | Warranty Supervisor | Claim WC-2026-30917 submitted by dealer D-114 with 4 photos | — | 200 | — | | 02 | reason | Claim Agent | VIN WBA…7712 · 31,400 km · part: turbo actuator · labour 2.4 h · 6,850 € | — | 860 | 690 | | 03 | tool | Validation Agent | Coverage: 3 years / 100,000 km, valid until 03/2027 → in scope | SAP S/4HANA | 420 | — | | 04 | retrieve | Validation Agent | Policy WP-14: actuator labour standard 1.8 h, part price list 2026-Q3 | Warranty Policy Vector Store | 400 | 520 | | 05 | verify | Validation Agent | Labour 2.4 h vs standard 1.8 h (+33%) → reduced; part price and photos consistent | — | 820 | 1,040 | | 06 | write | Claim Agent | Photos and diagnostic log archived under the claim folder | Document Store (S3) | 300 | — | | 07 | approval | Warranty Supervisor | Adjusted claim 6,100 € > 5,000 € → regional warranty manager approved | — | 2,800 | — | | 08 | write | Settlement Agent | Credit memo 1700442310 posted for 6,100 € to the dealer account | SAP S/4HANA | 540 | — | | 09 | write | Settlement Agent | Claim status set to "approved – adjusted", reason code L02 (labour) | Dealer Portal | 380 | — | | 10 | notify | Settlement Agent | Dealer notified in the portal with the adjustment breakdown, audit trail stored | Dealer Portal | 220 | — | Simulation metrics (estimated) 8,600runs / month 96.3%success 7 sp50 $0.0062est. cost / run Approval rule Warranty claim > 5,000 € → regional warranty manager approval Step profile ingest · 1retrieve · 1reason · 1tool · 1verify · 1approval · 1write · 3notify · 1 SYSTEMS - Dealer Portalapi - SAP S/4HANAerp - Warranty Policy Vector Storedb - Document Store (S3)storage What you need to run it - Credentials and a test environment for the enterprise APIs involved - Posting and query API access to ERP/MES - Read access to the database or vector store; a data owner - Read access to the document store and the document types - An authorised role and approval channel for the approval rule (mobile approval card, e-mail or ticket) - SSO with the organisation’s identity provider and a scenario-specific authorisation group Global pattern Mirrors dealer warranty claim automation at automotive OEMs and tier-1 suppliers Packages that include this scenario Production and maintenance operations → Sector: Manufacturing & Automotive → Related agents Manufacturing & AutomotiveHITL ### Line stoppage response: from alarm to work order and shift report An MES alarm lands; root cause found from history, spare part checked, work order and plan update ready within minutes. 3,100 runs / month 92.2% success 5 agents Manufacturing & AutomotiveHITL ### Supplier invoice three-way match Invoice, PO and goods receipt matched in SAP automatically; only out-of-tolerance lines reach a buyer. 21,000 runs / month 97.1% success 4 agents Manufacturing & AutomotiveHITL ### Maintenance ticket triage Sensor alarm and technician voice note become one work order, spare parts and a safe stop window included. 3,400 runs / month 95.2% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Compare URL: https://www.aginies.com/en/compare Language: en Summary: Compare Aginies with workflow automation tools, agent frameworks, chatbot platforms, hyperscaler agent studios, AI coding agents and infrastructure / model services at category level: orchestration, models, voice, knowledge, governance, deployment and operating model. Compare # Which category does Aginies belong to? Aginies is the execution layer where all of the enterprise’s AI scenarios run, deployable under full sovereignty; it does not map one-to-one onto any of the six tool categories below. The comparison is at category level, feature by feature. ## Workflow automation tools Trigger-action tools that connect applications, with large integration catalogues. What we mean Low-code automation tools and integration platforms (iPaaS), mostly SaaS, that wire applications together with trigger-action logic; AI is added to them as one step. When it fits Well-defined, rule-based automations; small teams; quick prototypes. The gap in production AI is bolted on as a step; supervisor architecture, human approval, audit trail and on-prem sovereignty stay at the product’s edges. 4 yes · 12 partial · 29 no ## Agent development frameworks Open-source libraries where developers build agent graphs in code. What we mean Open-source Python or TypeScript libraries in which developers build the agent graph in code: multi-agent orchestration, tool-calling and state-management frameworks. When it fits Organisations with a strong engineering team that want to build their own platform. The gap in production Observability, cost attribution, RBAC, voice and the operating model are written from scratch; accountability stays entirely with the enterprise. 7 yes · 12 partial · 26 no ## Chatbot and virtual assistant platforms Visual conversation designers for customer-service dialogues. What we mean Chatbot and virtual-assistant platforms built for customer-service dialogues, with a visual conversation designer, intent recognition and channel integrations. When it fits Customer assistants that inform, route and perform limited transactions. The gap in production Non-dialogue back-office processes (documents, reconciliation, SDLC) and write operations in core systems are out of scope. 5 yes · 20 partial · 20 no ## Hyperscaler agent studios Agent tooling from large cloud and CRM vendors, tied to their own ecosystems. What we mean Agent studios embedded in the products of large cloud, office-software and CRM vendors; they run on that vendor’s models and cloud. When it fits Enterprises standardised on one ecosystem that can keep their data in that cloud. The gap in production Model and cloud lock-in; on-prem / air-gap and in-country GPU capacity are mostly unavailable; local-language voice and local regulation need extra projects. 8 yes · 27 partial · 10 no ## AI coding agents Developer agents in the terminal and IDE that read the codebase, edit files and run commands; from single-vendor cloud tools to model-agnostic open-source tools. What we mean Developer agents that run in the terminal or IDE: commercial coding assistants tied to one vendor’s cloud model, and open-source tools that connect to any model endpoint. When it fits Individual developer productivity; teams whose code may go to a cloud model, or engineering teams willing to operate their own model endpoint. The gap in production A developer tool: no supervisor architecture for business processes, no voice or dialogue channels, no cost attribution per unit; governance is limited to a per-developer permission prompt, with no org-level audit trail or approval flow; for local models the enterprise builds the GPU infrastructure itself. Code Aginies offers the same IDE surface inside the sovereign platform, with in-country GPUs, an organisation licence and masking. 2 yes · 17 partial · 26 no ## Infrastructure and model services Infrastructure layers that serve models on any hardware, manage GPU and Kubernetes capacity and give token-based model access; delivered as an in-country cloud service or as software running in the enterprise’s own data centre. What we mean The AI platforms of enterprise Linux and Kubernetes vendors, and the GPU and model-as-a-service offerings of in-country cloud providers; they serve the model, the enterprise writes the agent. When it fits Organisations that want only model access and GPU capacity and will write agents with their own engineers; regulated sectors that prefer shared infrastructure. The gap in production Runs the model but does not bring the agent that does the work: the business-process library, local-language voice and dialogue channels, the code assistant, cost attribution per unit and accountability for outcomes stay with the enterprise. Aginies runs on top of this layer; the infrastructure choice does not change the Aginies choice. 11 yes · 19 partial · 15 no Positioning ## Where we sit on build model and governance The horizontal axis is how the agent is built (code → visual builder → build by dialogue); the vertical axis is governance and sovereignty (audit trail, human approval, on-prem). Aginies sits at the far end of both: agents are built by dialogue and run in the enterprise’s own data centre. [Positioning chart] ## Feature comparison Category level, as of October 2026. Products change constantly; for a one-to-one comparison with a specific product, let us look together in a discovery session. | Feature | Aginies | Workflow automation tools | Agent development frameworks | Chatbot and virtual assistant platforms | Hyperscaler agent studios | AI coding agents | Infrastructure and model services | | --- | --- | --- | --- | --- | --- | --- | --- | | Agent workforce | | Hired agents per role (developer, analyst, accountant, tester)Outsourced agents instead of outsourced people; the agent works in your tools | Yes | No | No | No | Partial | No | No | | Work log: task, output, hour-equivalent, quality, approval, cost | Yes | No | No | No | No | No | No | | Shadow period → sampled review; a named team lead | Yes | No | No | No | No | No | No | | Orchestration | | Supervisor → sub-agent architecture | Yes | No | Yes | Partial | Partial | Partial | Partial | | Visual agent builder (no glue code) | Yes | Yes | No | Yes | Yes | No | Partial | | Stateful multi-step execution with parallel branches | Yes | Partial | Yes | Partial | Partial | Partial | Partial | | Agent-to-agent communication (A2A) | Yes | No | Partial | No | Partial | No | Partial | | Autonomous Agent: build, test and deploy agents by dialogue | Yes | No | No | No | Partial | No | No | | Models | | Model-agnostic gateway: proprietary and open-sourceModelHub | Yes | Partial | Yes | Partial | No | Partial | Yes | | On-prem serving of open-source models | Yes | No | Partial | No | No | Partial | Yes | | Fallback model strategy and prompt versioning | Yes | No | Partial | Partial | Partial | No | Partial | | Model proxy: cloud models with your own keys, observed and masked | Yes | Partial | Partial | No | No | Partial | Partial | | Vision and document models | Yes | No | Partial | No | Partial | No | Partial | | OpenAI-compatible model API served to enterprise applications: REST, SSE and WebSocketClient code survives a provider change | Yes | No | No | No | Partial | No | Yes | | Model-as-a-Service: token-based usage with group and user quotas | Yes | No | No | No | Partial | No | Yes | | Channels | | Voice agent: streaming speech and call-centre SIP | Yes | No | No | Partial | Partial | No | No | | Dialogue: session memory, intent and entity blocks | Yes | No | Partial | Yes | Yes | No | Partial | | WhatsApp, SMS, email, web widget | Yes | Yes | No | Yes | Partial | No | No | | Turkish speech and local-language quality | Yes | No | No | Partial | Partial | No | Partial | | Code: IDE extensions, code completion, IDE registry | Yes | No | No | No | Partial | Yes | No | | Closed-circuit code assistant: VPN, in-country model, code context stays inside the perimeterScenario on the Security page | Yes | No | No | No | No | Partial | No | | Knowledge | | Multi-source RAG with hybrid search | Yes | Partial | Yes | Yes | Yes | Partial | Partial | | Chunk-level inspection and editing | Yes | No | No | Partial | Partial | No | No | | Integration | | MCP servers and custom tools with per-tool policies | Yes | Partial | Yes | Partial | Partial | Yes | Partial | | Write operations in core systems (core, ERP, CRM) | Yes | Yes | Yes | Partial | Partial | No | No | | Prebuilt connector catalogue (thousands of apps)Aginies connects to enterprise systems through REST, databases and MCP; every connection is defined as a tool with its own policy. Native connectors cover email, calendar, files and collaboration tools; a marketplace of prebuilt connectors to thousands of SaaS apps is outside the scope | Partial | Yes | No | Partial | Yes | No | No | | Governance | | Human-in-the-loop checkpoints (threshold and approver based) | Yes | Partial | Partial | Partial | Partial | Partial | No | | Full audit trail for every agent action and model call | Yes | Partial | No | Partial | Partial | Partial | Partial | | PII masking, guardrails, kill-switch | Yes | No | Partial | Partial | Yes | Partial | Partial | | SSO / SAML, fine-grained RBAC, workspace isolation | Yes | Partial | No | Yes | Yes | Partial | Yes | | Workspace / tenant separation; agent sharing and co-development inside the organisation | Yes | Partial | No | Partial | Yes | Partial | Partial | | Masking and anonymisation of all requests | Yes | No | No | Partial | Partial | No | Partial | | Model and agent evaluation: test corpus, safety screening, regulatory compliance testing | Yes | No | Partial | Partial | Partial | No | Partial | | Observability | | Per-run traces, latency and token tracking | Yes | Partial | Partial | Partial | Partial | Partial | Yes | | Cost attribution by scenario, unit and channel | Yes | No | No | No | Partial | No | Partial | | Per-model usage and cost, platform health dashboard | Yes | No | No | Partial | Partial | Partial | Yes | | Deployment | | On-prem / air-gapped deployment | Yes | Partial | Yes | No | No | Partial | Partial | | Hybrid and dedicated-region options | Yes | Partial | Partial | Partial | Partial | Partial | Yes | | In-country GPU-as-a-Service | Yes | No | No | No | No | No | Yes | | Deployment aligned with KVKK, banking IT regulation and GDPR | Yes | No | No | Partial | Partial | No | Partial | | Self-service GPU and Kubernetes capacity, pay as you goAginies reserves capacity by contract | Partial | No | No | No | Yes | No | Yes | | Inference engine and distributed GPU scheduling as a product (vLLM-based serving, multi-cluster)Aginies operates open engines, does not productise them | Partial | No | No | No | Partial | No | Yes | | Operating model | | One accountable partner: team + platform + infrastructure | Yes | No | No | No | No | No | No | | Five-layer assessment methodology | Yes | No | No | No | No | No | No | | SDLC agent team: PM, analyst, architect, developer, tester | Yes | No | Partial | No | Partial | Partial | No | Evidence from the library ## The claims in the matrix are backed by scenario definitions. Every figure is derived at build time from the agent library definitions and can be read as evidence for the matching row. 73scenarios with a human-approval rule 127write steps in enterprise systems 23scenarios on the voice channel 200enterprise systems connected 14sectors, one execution layer 5model families, in-country GPUs Full version with example products This page compares categories and names no products. The full comparison with each category’s example products is in the documentation area, behind an access code. Docs › Compare ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Contact URL: https://www.aginies.com/en/contact Language: en Summary: A 30-minute discovery session: your priority business problems, a live demo of a similar scenario, and a roadmap draft that starts with the Value layer. Next step # See Aginies on your own workflows. Book a 30-minute session with our team. We walk through your use case, show the platform and outline a path to production. What happens in the session - 1We take your 2–3 priority business problems together. - 2You see a live demo of a similar scenario on Aginies. - 3We draft a roadmap for you that starts with the Value layer analysis. All you need to bring - One business problem and its current volume (monthly transactions, calls, documents) - The names of the systems to act in; you do not need to know whether they have APIs - Your first view on which decisions must stay with people What you get after the session - The closest library scenario and solution package to your case - A draft scorecard for the value and readiness layers - A pilot scope and deployment-model recommendation for the first 90 days Email dahi@aginies.com Location Istanbul, Türkiye · Serving clients globally Starting points for your sector Before the discovery session you can look at your sector’s first-90-days plan. - Bankingfirst scenario: Dialogue banking: voice and text transaction assistant - Fintech & Paymentsfirst scenario: Merchant operations assistant: POS, settlement, refund and limit - Insurancefirst scenario: End-to-end motor claim: from notice to payment - Retail & E-commercefirst scenario: Outfit stylist: a complete look by size, stock, budget and style - Telecomfirst scenario: Subscriber care assistant: bill, plan, fault and contract - Healthcarefirst scenario: Patient journey coordinator: appointment, preparation, approval and results - Manufacturing & Automotivefirst scenario: Line stoppage response: from alarm to work order and shift report - Logisticsfirst scenario: Control tower: delay prediction, rerouting, customer notice, SLA - Energy & Utilitiesfirst scenario: Utility customer assistant: bill, outage, consumption and instalments in one dialogue - Government & Public Sectorfirst scenario: Citizen services assistant: application, documents, appointment and status - Defence & Aerospacefirst scenario: Technical publication assistant: maintenance manual, parts, procedure and revision control - IoT & Connected Devicesfirst scenario: Field device support assistant: telemetry, remote diagnosis, OTA and technician - Software Deliveryfirst scenario: Feature delivery pipeline --- # Aginies overview · Documentation URL: https://www.aginies.com/en/docs Language: en Summary: Enterprise AI transformation: from pilot to production, end-to-end. What Aginies is, the problem it solves, how it works and who it is for. Documentation On this page What is Aginies?The problem: reaching productionThe offer: one accountable partnerHow it works: four stepsProduct familyProduction proofWho it is forSolution packages Overview # Aginies overview Enterprise AI transformation: from pilot to production, end-to-end. What Aginies is, the problem it solves, how it works and who it is for. Updated: September 2026 ## What is Aginies? Aginies is an enterprise agentic AI orchestration platform. Supervisor agents plan the objective, specialised agents act inside core systems; every step is traced, passes human approval where needed, and is costed. It is model-agnostic and runs on-premise, hybrid or in the cloud. It is the single execution layer for all of the enterprise’s AI scenarios. Aginies is taken in two forms: as a platform (the execution layer on which the enterprise builds and runs its own agents) and as an agent workforce (Workforce Aginies: roles such as iOS developer, business analyst, accountant and tester hired as agents). Both run on the same platform under the same governance; the second offers outsourced agents instead of outsourced people. The thesis is simple: enterprise AI transformation must be handled from pilot to production, end-to-end. Most enterprises experiment with AI; very few run it in production, at scale, under control. Aginies was built to close that gap. ## The problem: reaching production The real problem with AI is reaching production. Pilots impress; the path to production is usually undefined. Four failure modes repeat across enterprises: - The PoC gap: the pilot works, the path to production stays undefined. - Scattered initiatives: every unit trials its own tool; no enterprise coherence. - Unmeasured value: the gap between “it works” and “it pays off” goes unreported. - Governance gap: security, compliance and auditability are afterthoughts. Result: budget spent, transformation missing. ## The offer: one accountable partner A partner that owns enterprise AI transformation and runs it end-to-end. Three components under one roof; accountability in one place. Consultants do not come and go; the product is operated together. | Component | What it provides | Why it matters | | --- | --- | --- | | Expert team | Strategy, data, engineering and change management in one team | Scenario selection, build and operation stay with one team | | The Aginies platform | Enterprise-grade agentic orchestration layer | Every AI workload on one execution layer; observed and governed | | Sovereign infrastructure | GPU-as-a-Service inside national borders | Data and model control stay with the customer; scales without capital outlay | ## How it works: four steps Agentic AI runs real business processes in four steps. Every step happens on the same orchestration layer and is recorded in the same audit trail. - Understand signals: inputs from applications, APIs, documents, voice and events become execution-ready tasks. - Supervisor plans: the complex objective is decomposed into structured steps; responsibilities go to specialised agents. - Agents act: knowledge is retrieved, documents processed, tools invoked, real operations triggered in enterprise systems. - Everything is observed: agent decisions, latency, cost and tool usage are monitored and optimised with full visibility. ## Product family Platform Aginies is the umbrella product; the others are its modules. Every agent is built, observed and governed on one control plane. | Product | Scope | Highlights | | --- | --- | --- | | Platform Aginies | Orchestration core and control plane | Visual agent builder, Autonomous Agent, knowledge base, logs, dashboards, workspace separation | | Voice Aginies | STT · TTS · ASR | Speech models trained for Turkish; call-centre SIP integration; on-prem serving | | Code Aginies | Engineering teams | IDE registry, IDE extensions, in-house code completion, SDLC agent team | | Hub Aginies | Model Hub · MCP Hub · proxy | Enterprise fine-tuned models, on-prem oss models, proxy with your own keys, approved MCP servers | | Vision Aginies | Documents · images · masking | Invoice, contract and form extraction; damage photos; personal-data detection and masking | | Workforce Aginies | Agent workforce, on demand | Roles such as iOS developer, business analyst, accountant and tester hired as agents; work log, shadow period, agent-month billing | | Dialog Aginies | Dialogue UI | Web widget, mobile, WhatsApp; session memory; human takeover; per-message audit trail | ## Production proof Production figures for the workloads running on Aginies today, last 30 days: - 13.1M+ executions in the last 30 days - 97.4% success rate - 1,920+ agents active in production - 13,100+ agents across all deployments These workloads run at retail, insurance, automotive and banking enterprises in Türkiye. Alongside the production figures, facts derived from the agent library on this site: | Indicator | Value | Source | | --- | --- | --- | | Live scenarios | 77 | Agent library | | Sectors | 14 | Scenario definitions | | Specialist agents | 261 | Scenario definitions | | Distinct tool definitions | 657 | Agent tool lists | | Systems connected | 200 | Scenario system lists | | Scenarios with a human-approval rule | 73 | Approval rules | | Write steps in enterprise systems | 127 | Step kinds | | Simulated runs / month (total) | 3,320,970 | Scenario metrics (estimated) | This table is derived at build time from the agent library’s scenario definitions and changes as scenarios are added. Volume and cost figures are simulation definitions and estimates. ## Who it is for Aginies is designed for large and mid-sized enterprises that want to move AI out of pilots and into production under control. The requirements of regulated sectors are part of the architecture. - CIOs, CDOs and transformation leads: consolidating scattered initiatives on one platform - Business unit leaders: measurable outcomes in customer, operations and finance processes - Risk, compliance and internal audit: traceability and human approval for every decision - Engineering teams: the agent library, the SDLC agent team and in-house code completion The agent library covers fourteen sectors: banking, fintech and payments, insurance, retail and e-commerce, telecom, healthcare, manufacturing and automotive, logistics, energy and utilities, government and public sector, defence and aerospace, IoT and connected devices, enterprise functions and software delivery. ## Solution packages 30 solution packages are grouped in seven areas. Each is defined with scope, prerequisites, measured targets, governance rules and a rollout plan; its architecture and volumes are derived from the linked scenarios. | Package | Sector | Area | Linked scenarios | First measured target | | --- | --- | --- | --- | --- | | E-commerce support & action | Retail & E-commerce | Customer interaction & experience | 4 | First-contact resolution: +15–25 points | | Invoice & document extraction | Enterprise Functions | Financial analysis & operational intelligence | 4 | Touchless document share: 70–85% | | Voice AI for claims intake | Insurance | Customer interaction & experience | 4 | Notices completed without a representative: 60–75% | | Email classification & attachment verification | Enterprise Functions | Operational automation | 4 | Correct first routing: 95% and above | | Call-centre quality monitoring | Enterprise Functions | Customer interaction & experience | 3 | Calls evaluated: From a 3–5% sample to 100% coverage | | Conversational banking | Banking | Customer interaction & experience | 4 | Transactions completed by conversation: 55–70% | | SDLC agent team | Software Delivery | Software development & technology productivity | 5 | PRs accepted on first review: 60–75% | | Sales & lead management | Retail & E-commerce | Customer interaction & experience | 4 | Time to first contact: From hours to minutes | | Financial analysis & operational intelligence | Fintech & Payments | Financial analysis & operational intelligence | 4 | Monthly report preparation time: 60–70% shorter | | Operational automation & demand management | Enterprise Functions | Operational automation | 4 | Requests closed automatically: 45–60% | | Information management & decision support | Government & Public Sector | Information management & decision support | 4 | Answers with a citation: 100% | | Healthcare operations & clinical guidance | Healthcare | Customer interaction & experience | 4 | No-show rate: 30–40% lower | | HR & employee experience | Enterprise Functions | HR & employee experience | 3 | Onboarding completion time: From weeks to days | | KVKK / GDPR personal data protection | Enterprise Functions | Data protection & compliance | 4 | Personal-data detection accuracy: 97% and above | | Voice analysis & quality | Telecom | Customer interaction & experience | 4 | Calls analysed: 100% | | Field service & dispatch automation | IoT & Connected Devices | Operational automation | 4 | Requests resolved remotely: 30–45% | | IoT device fleet operations | IoT & Connected Devices | Software development & technology productivity | 4 | Alarm-to-incident clustering: 80% less noise | | Utility customer operations | Energy & Utilities | Customer interaction & experience | 4 | Calls resolved without a representative: 55–65% | | Citizen services & file pre-review | Government & Public Sector | Customer interaction & experience | 4 | Documents asked of the citizen again: 70% fewer | | Public procurement & tender compliance | Government & Public Sector | Data protection & compliance | 3 | File pre-review time: 70% shorter | | Defence export control & supply-chain security | Defence & Aerospace | Data protection & compliance | 4 | Party screening time: From days to minutes | | Agent outsourcing (agent workforce) | Enterprise Functions | HR & employee experience | 4 | Accepted output share: 85% and above | | Credit and onboarding operations | Banking | Financial analysis & operational intelligence | 4 | Accounts opened without a branch visit: 70–85% | | Fraud and dispute operations | Banking | Operational automation | 4 | Review time per alert: 70% shorter | | Claims operations end to end | Insurance | Operational automation | 4 | File closure time (simple claims): From weeks to days | | Underwriting and renewal operations | Insurance | Financial analysis & operational intelligence | 3 | Policies renewed automatically: 60–75% | | Merchant and payment operations | Fintech & Payments | Operational automation | 4 | Merchant support first-contact resolution: 70% and above | | Subscriber care and network support | Telecom | Customer interaction & experience | 5 | Requests resolved without a representative: 60–70% | | Logistics control tower | Logistics | Operational automation | 4 | Delay detection lead: Hours ahead | | Production and maintenance operations | Manufacturing & Automotive | Operational automation | 4 | Unplanned downtime: 20–30% lower | Targets are ranges from comparable deployments; baseline and target are fixed with the organisation in the first week of the pilot. They are recalculated with the organisation’s own data in the discovery session. --- # Agent outsourcing: outsourced agents instead of outsourced people · Documentation URL: https://www.aginies.com/en/docs/agent-outsourcing Language: en Summary: With Workforce Aginies, roles such as iOS developer, business analyst, accountant and test engineer are hired as agents. This page explains the model, the roles, the work log, the engagement model, pricing and governance. Documentation On this page What is the model?Role catalogueThe work log: what it did and how muchEngagement model: five stepsCommercial modelGovernance and securityMeasured targets and rollout Agent outsourcing # Agent outsourcing: outsourced agents instead of outsourced people With Workforce Aginies, roles such as iOS developer, business analyst, accountant and test engineer are hired as agents. This page explains the model, the roles, the work log, the engagement model, pricing and governance. Updated: September 2026 ## What is the model? The classic outsourcing model sells person-months: a vendor gives you a developer, an analyst or an accountant; you pay by the hour or by the month. In agent outsourcing an agent holds the same role. It gets accounts in your tools (Jira, Git, ERP, CRM), reports to your team and writes its tasks to the work log. Decisions stay with your approver; a named team lead on the Aginies side manages the agent. The positioning line: “Hire the role, let an agent hold it.” An agent is the holder of a role; yet its authority is bounded per tool, every step is recorded and it can always be stopped. - Starting takes days: role spec, access, a two-week shadow period. - Knowledge stays in the workspace: the next agent starts with the same test corpus and connections. - Traceability comes from the work log: task, output, hour-equivalent, quality, approval, cost. - Scaling takes minutes: a second agent of the same role; when demand drops, capacity and invoice drop. ## Role catalogue The catalogue holds twelve roles today; each is an agent spec: what it does, where it works, which outputs it produces, which KPIs track it, where human approval is required. Roles outside the catalogue are defined with the Autonomous Agent within days in a discovery session. | Role | Replaces | Example output | Human control | | --- | --- | --- | --- | | iOS Developer Agent | Outsourced iOS developer | Tested PR, story points closed | Every PR human-reviewed; deployment approved | | Backend Developer Agent | Outsourced backend developer | Endpoints, migrations, integration tests | Schema changes and production deployment approved | | QA Engineer Agent | Outsourced tester | Regression run, evidenced bugs, coverage report | Release go / no-go with the release manager | | DevOps / SRE Agent | Outsourced DevOps engineer | Pipeline changes, resolved incidents, cost reports | Production infrastructure changes approved | | Business Analyst Agent | Outsourced business analyst | Ready stories, acceptance criteria, process maps | The product owner approves the stories | | Data Analyst Agent | Outsourced data analyst | Reports, dashboards, data-quality findings | Externally shared reports approved | | Accountant Agent | Outsourced accountant | Entries, reconciliations, close tasks | Entries with the accountant’s approval; filings with people | | Procurement Specialist Agent | Outsourced procurement specialist | RFQs, matched invoices, savings reports | Order approval and supplier selection with the manager | | Customer Support Rep Agent | Outsourced contact-centre agent | Tickets resolved, first-contact resolutions | Refunds and above-limit actions approved; 10% sampling | | Sales Development Rep Agent | Outsourced SDR | Qualified leads, meetings booked | External outreach always human-approved | | Recruiter Agent | Outsourced recruiter | Candidates screened, interviews scheduled | Screening and offer decisions with HR | | Contract Analyst Agent | Outsourced legal assistant | Contracts reviewed, redlines, summaries | Every redline with counsel approval | ## The work log: what it did and how much Every task is a row in the work log. The row carries: ticket or story id, start and finish, the concrete output (PR, report, entry, ticket), hour-equivalent (estimated human time for the same work), quality (review result, rework, check-family scores), approval (who, when, rationale) and cost (tokens, tool calls, model). - Daily: work log and approval queue; your team lead sees what awaits approval. - Weekly: timesheet, KPIs and quality; in the same shape as a classic outsourcing report, but per task. - Monthly: invoice, review and scale decision; invoice lines are produced from the work log. Hour-equivalent is a comparison measure: it shows how many hours the same work would take a person. Billing is per agent-month or per output. ## Engagement model: five steps - Role scoping: the job description becomes an agent spec (tasks, tools, outputs, KPIs, approval points). - Access and onboarding: the agent gets accounts in your tools; SSO, roles, read / write permissions, data boundaries. - Shadow period (two weeks): every output reviewed by a person; mistakes added to the test corpus. - Sampled review: after the quality threshold, 10% sampling and threshold-based approval; the kill-switch always on. - Monthly review and scale: work log, KPIs and cost read together; the role is widened, added to or ended. Responsibility is clear at every step: role scoping and review are shared between you and the Aginies team lead; access belongs to IT and security; review and approval to your team. ## Commercial model | Model | How it works | Best fit | | --- | --- | --- | | Agent-month | A fixed monthly fee per role; unlimited tasks, defined KPIs and SLA. | Roles with continuous workload | | Output-based | Fee per merged PR, resolved ticket or processed invoice; the work log is the invoice line. | Variable or project-based work | | Agent pod | Several roles + an Aginies team lead; works like a product team. | Teams that want end-to-end delivery | There are no list prices on the site; pricing is set together in discovery by role, volume and approval intensity. The customer owns the outputs and the knowledge produced; contracts can be ended monthly and the test corpus stays with the customer on exit. ## Governance and security - The agent works in your tenant and your tools; code and data never leave. - Read / write permission per tool; every step over threshold goes to approval (approval matrix on the Security page). - Approver and rationale recorded for every output; audit export in CSV / JSON. - Kill-switch: immediate stop per role or task; a rollback plan on every write. - A named Aginies team lead: weekly planning, monthly review, first response. How to tell it: “Managed like an employee, audited like a system.” Show the customer the work log first, then the role catalogue; pricing belongs to the discovery session. ## Measured targets and rollout | Indicator | Target range | | --- | --- | | Accepted output share | 85% and above | | Cost per role | 30–50% of an outsourced person-month | | Knowledge loss at turnover | Zero; knowledge stays in the workspace | What you need to start: - Tool access the role needs (Jira, Git, ERP, ticketing) - Task definition and acceptance criteria - A review owner on the organisation side Rollout: - 1 · Policy and role catalogue · 2 weeks: HR policies, process owners and system access defined; personal-data boundaries written. - 2 · Pilot · 4 weeks: One department or one role; quality score and user feedback tracked weekly. - 3 · Scale-out: All departments; thresholds, review sampling and kill-switch owned by HR management. Targets are ranges from comparable deployments; baseline and target are fixed together in the first week. --- # How agents are built and run · Documentation URL: https://www.aginies.com/en/docs/agents Language: en Summary: Primitives, the supervisor → sub-agent pattern, tools and MCP policies, approvals, the run lifecycle, the agent library and the SDLC agent team. Documentation On this page PrimitivesThe supervisor → sub-agent → supervisor patternTools and MCP policiesApprovals and thresholdsRun lifecycleThe agent libraryThe SDLC agent team Agents # How agents are built and run Primitives, the supervisor → sub-agent pattern, tools and MCP policies, approvals, the run lifecycle, the agent library and the SDLC agent team. Updated: September 2026 ## Primitives In the visual agent builder every agent is composed from eight primitives. The Autonomous Agent uses the same primitives; an agent produced by dialogue runs and is observed exactly like one designed by hand. | Primitive | Role | Typical use | | --- | --- | --- | | Agent | A specialist with its own model, tools and memory | Credit agent, claims agent, invoice agent | | Supervisor | Breaks the objective into steps, delegates, merges the result | Entry point of every multi-agent scenario | | Router | Branches the process by intent, rule or score | Intent classification, path choice by risk score | | Memory | Shares session and long-term context across agents | Customer context, earlier turns | | Guardrail | Checks input and output against policy; blocks or masks | PII masking, unauthorised requests, prompt injection | | Human approval | Pauses on critical decisions, presents to the approver, records the decision | Above-threshold amounts, policy exceptions | | Tool / MCP | Read or write to an enterprise system; policy per tool | Core, CRM, ERP, ticketing calls | | Parallel & loop | Runs independent steps simultaneously and repeated work until a condition is met | Multi-source lookups, batch document processing | ## The supervisor → sub-agent → supervisor pattern This is the core execution pattern of Aginies. The supervisor plans and merges results; specialist agents take narrow, well-defined tasks. Control returns to the supervisor after every step, so one sub-agent’s failure does not break the whole process. - The supervisor receives the request and decomposes the objective into structured steps. - It delegates each step to the right specialist; independent steps run in parallel. - The sub-agent works with its own model and tools and returns the result with its rationale. - The supervisor verifies, merges and decides the next step. - Where approval is required the process pauses; the decision is recorded, then it resumes. - On completion the outcome goes to the channel; the full trace is written to the audit log. Execution is stateful: steps, intermediate results and tool responses are kept. When a dependency fails the supervisor retries, switches to the fallback model or escalates to a person; the process resumes where it stopped. ## Tools and MCP policies Agents reach enterprise systems through tools: REST APIs, databases and Model Context Protocol (MCP) servers. Every tool is classified as read or write and carries its own invocation policy. The MCP Hub catalogues approved servers only. - Read / write class: write tools are bound to an approval or threshold rule by default - Per-tool invocation policy: allowed parameters, rate limit, which agents may call it - Input and output masking: sensitive data to and from the tool is redacted - Every call appears in the trace: parameters, duration, response code and result - MCP Hub: approved servers, versions and access rights under central management - Agent-to-agent communication (A2A) works under the same policy framework Rights on core systems open in stages: read first, then controlled write, then autonomous write. Each stage is measured and approved. ## Approvals and thresholds Human approval is a primitive of the agent definition. On a critical decision the process pauses, the decision is presented to the approver with context and rationale, and the outcome is written to the audit trail. Thresholds and approvers are set in the agent definition. - Amount threshold: transactions above a value go to the approver - Risk threshold: a score or guardrail signal triggers approval - Policy exception: out-of-policy requests always reach a person - Irreversible actions: mandatory approval for deletion, payment and contract changes - Approver role: defined roles such as branch manager, compliance team, release manager - Kill-switch: immediate stop and automation limit per agent or scenario Example: in the card-debt payment agent, “amount > 100,000 → branch manager approval”. The rule lives in the agent definition; every decision leaves a trace. ## Run lifecycle Every run is made of nine step kinds. Each agent page in the library shows its steps with these kinds; the duration, token usage and touched system of every step are recorded in the trace. - Ingest: the signal arrives; voice, email, API, message or event - Plan: the supervisor decomposes and routes - Retrieve: lookup in the knowledge base or database (RAG) - Reason: classification, extraction, scoring - Tool: a read call to an enterprise system or an MCP tool - Verify: guardrail, policy and consistency checks - Approval: the human-in-the-loop checkpoint - Write: a state-changing action in an enterprise system - Notify: customer or staff communication A run starts with an ingest step and ends with a write or notify step. End-to-end latency is derived from the step durations; cost per run is computed from the token usage of model calls and attributed by scenario, unit and channel. ## The agent library The library holds 77 agents across 14 sectors. Each is a simulation derived from real agent definitions: supervisor, specialist agents, systems, steps and approval rule are all defined. Every one can be produced with the Autonomous Agent and adapted to the enterprise’s systems. How to read an agent page: - Title and tagline: the agent’s purpose and outcome-oriented summary - Trigger and outcome: what starts the run and what is true when it finishes - Channels: voice, email, app, API, WhatsApp, web or batch - Supervisor and agents: each agent’s role, model and tools - Systems: the core, CRM, ERP, ticketing, payments and other systems touched - Steps: in execution order with kind, target system, duration and tokens - Approval rule: the threshold and approver of the human checkpoint, where present - Metrics: monthly runs, success rate, median duration, cost per run - Inspired by: the proven global pattern the agent mirrors Metrics are realistic indicators for an enterprise deployment; they are recalculated in a discovery session for the enterprise’s volume and systems. The full list of the library; every row is a scenario page. | Scenario | Sector | Channels | Agents | Systems | Approval rule | | --- | --- | --- | --- | --- | --- | | Dialogue banking: voice and text transaction assistant | Banking | app, voice, whatsapp | 4 | 5 | Limit increase > 60%, dispute amount > 5,000 or transfer to a new payee → operations approval | | Merchant operations assistant: POS, settlement, refund and limit | Fintech & Payments | whatsapp, web, voice | 4 | 5 | Refund > 25,000 or limit increase > 50% → risk team approval | | End-to-end motor claim: from notice to payment | Insurance | voice, app | 4 | 5 | Estimate > 150,000, unclear fault share or overlap with prior damage → adjuster approval | | Outfit stylist: a complete look by size, stock, budget and style | Retail & E-commerce | app, web, whatsapp | 4 | 5 | Discount outside campaign rules or a bespoke price request → customer service | | Subscriber care assistant: bill, plan, fault and contract | Telecom | voice, app, whatsapp | 4 | 5 | Contract cancellation, bill correction above 500 or technician priority → representative approval | | Legacy module modernisation: analysis, tests, rewrite, PR | Software Delivery | api, batch | 4 | 4 | Production schema change or out-of-scope file → architect approval; PR merge always human | | Patient journey coordinator: appointment, preparation, approval and results | Healthcare | whatsapp, voice, app | 4 | 5 | Drug interaction, suspected contraindication or extra test need → physician approval | | Line stoppage response: from alarm to work order and shift report | Manufacturing & Automotive | api, app | 4 | 5 | Schedule change > 2 hours or urgent purchase above 50,000 → shift lead approval | | Control tower: delay prediction, rerouting, customer notice, SLA | Logistics | api, email, whatsapp | 4 | 5 | Route cost increase > 15% or compensation > 10,000 → operations manager approval | | Utility customer assistant: bill, outage, consumption and instalments in one dialogue | Energy & Utilities | voice, whatsapp, app | 4 | 5 | Instalments > 6 months, bill cancellation or meter suspicion → customer service approval | | Citizen services assistant: application, documents, appointment and status | Government & Public Sector | whatsapp, voice, web | 4 | 5 | Borderline eligibility, inconsistent document or another application in the household → officer review | | Procure-to-pay desk: request, quotes, contract and invoice in one place | Enterprise Functions | email, api | 4 | 5 | Amount > 250,000, single quote or non-framework supplier → procurement committee | | Technical publication assistant: maintenance manual, parts, procedure and revision control | Defence & Aerospace | app | 4 | 5 | Proposed deviation from the procedure → authorised engineer approval; export-controlled content → clearance required | | Field device support assistant: telemetry, remote diagnosis, OTA and technician | IoT & Connected Devices | whatsapp, web, api | 4 | 5 | Bulk OTA (>100 devices) or out-of-warranty device replacement → product-owner approval | | Credit card limit increase | Banking | app, voice | 3 | 3 | Limit > 100,000 → branch manager approval | | Digital account opening (KYC) | Banking | app, web | 4 | 5 | Sanctions/PEP fuzzy match score > 0.65 → compliance analyst approval | | Card fraud alert triage | Banking | whatsapp, sms, voice, api | 3 | 5 | Confirmed fraud loss > 2,500 → analyst approval before permanent cancellation | | Chargeback dispute handling | Banking | app, email, branch | 3 | 4 | Amount > 2,500 → dispute team lead approval before provisional credit | | Early-stage collections outreach | Banking | batch, whatsapp, sms, voice | 3 | 4 | Plan balance > 25,000 or > 3 instalments → collections team lead approval | | Mortgage document review | Banking | web, email, branch | 3 | 4 | Loan > 3,000,000 or DTI > 45% → underwriting officer approval | | Voice banking money transfer | Banking | voice | 4 | 4 | Amount > 50,000 or new payee → warm transfer to a call-centre agent | | Credit card debt payment | Banking | app, voice | 3 | 4 | Amount > 100,000 → branch manager approval | | Merchant KYB onboarding | Fintech & Payments | web, email | 4 | 5 | Risk tier high or UBO fuzzy match > 0.65 → compliance manager approval | | AML transaction monitoring alert triage | Fintech & Payments | api, batch | 4 | 4 | Every alert → compliance analyst decision mandatory; the agent never files an STR or closes a case alone | | BNPL instant instalment decision | Fintech & Payments | api, app, sms | 4 | 5 | Basket > 15,000 or bureau score 10,000 or total > 50,000 → finance manager approval before adjustment | | Voice claims intake (FNOL) | Insurance | voice, whatsapp | 4 | 4 | Injury or reserve > 250,000 → warm transfer to a claims specialist | | Policy renewal offer | Insurance | batch, whatsapp, email, app | 3 | 4 | Premium increase > 20% or discount > 10% → renewal team lead approval | | Underwriting document extraction | Insurance | email, web, batch | 3 | 4 | Loading ≥ 50% or sum assured > 5,000,000 → senior underwriter approval | | Motor claim photo assessment | Insurance | app, whatsapp | 3 | 4 | Estimate > 15,000 or model confidence 50,000 → subrogation counsel approval | | Where is my order? (WISMO) | Retail & E-commerce | whatsapp, web, app, voice | 3 | 4 | — | | Returns and refunds | Retail & E-commerce | app, web, whatsapp, email | 3 | 4 | Refund > 2,500 or > 5 returns in 12 months → team lead approval | | Product Q&A assistant | Retail & E-commerce | web, app, whatsapp | 3 | 4 | — | | Catalogue enrichment (batch) | Retail & E-commerce | batch, email | 3 | 4 | New category or supplier price deviation > 15% → category manager approval | | Review response and quality escalation | Retail & E-commerce | web, api | 3 | 3 | 1-star review replies and defect escalations → community manager approval | | Bill dispute resolution | Telecom | app, voice, whatsapp | 3 | 3 | Goodwill credit > 75 → team lead approval | | SIM swap fraud check | Telecom | branch, app, api | 3 | 4 | Risk score ≥ 70 or credential change within 72 hours → fraud desk approval | | Network outage communications | Telecom | sms, whatsapp, api, batch | 3 | 4 | Affected subscribers > 10,000 or P1 outage → network operations manager approval | | Plan upgrade advisor | Telecom | app, whatsapp, voice | 3 | 5 | Commitment-based or discounted bespoke offer → campaign manager approval | | Feature delivery pipeline | Software Delivery | api, web | 4 | 4 | Every PR → at least 1 human code review approval; agents never push to main directly | | CI/CD release pipeline | Software Delivery | api, batch | 4 | 4 | Every production deploy → release-manager approval mandatory; a critical CVE closes the gate automatically | | Incident response and hotfix | Software Delivery | api, web | 4 | 5 | Every emergency production deploy → on-call engineer approval; the hotfix PR still gets a human review | | Code review and security gate | Software Delivery | api, web | 4 | 4 | Every PR → human CODEOWNERS approval; a high SAST finding or coverage 25,000 TL → utilisation management nurse approval | | Patient intake and triage | Healthcare | app, voice, whatsapp | 4 | 4 | A triage nurse confirms the urgency level on every case; "emergency" cases are directed to 112 immediately | | Appointment scheduling and no-show reduction | Healthcare | whatsapp, sms, batch | 3 | 3 | — | | Supplier invoice three-way match | Manufacturing & Automotive | email, api, batch | 3 | 3 | Price deviation > 2% or > 1,000 TL → buyer approval | | Maintenance ticket triage | Manufacturing & Automotive | api, app, voice | 4 | 4 | Planned stop > 4 hours → shift supervisor approval | | Warranty claim processing | Manufacturing & Automotive | web, api, email | 3 | 4 | Warranty claim > 5,000 € → regional warranty manager approval | | Shipment exception handling | Logistics | api, email, whatsapp | 3 | 4 | Extra cost > 1,000 € → dispatch lead approval | | Customs document preparation | Logistics | api, email, batch | 3 | 5 | HS confidence 2,000 € → customs specialist approval | | Outage report handling | Energy & Utilities | voice, app, whatsapp, web | 4 | 5 | Mass notification > 1,000 customers → control room dispatcher approval | | Meter reading dispute | Energy & Utilities | whatsapp, app, web | 3 | 4 | Credit / correction > 1,000 TL → billing team lead approval | | Rooftop solar connection application | Energy & Utilities | web, email, api | 4 | 4 | The connection opinion is always signed by a network planning engineer | | Demand response event | Energy & Utilities | api, app, whatsapp, batch | 3 | 4 | Event > 10 MW → grid operator approval; settlement with an audit trail | | Citizen request routing | Government & Public Sector | web, app, email, voice | 3 | 4 | Classification confidence 500,000 € → counsel approval | | Employee onboarding | Enterprise Functions | api, email, app | 4 | 5 | Access outside the role template or admin rights → line manager approval | | IT helpdesk L1 | Enterprise Functions | app, voice, web | 3 | 3 | Access to a finance or production group → group owner approval | | Accounts payable inbox | Enterprise Functions | email, batch | 3 | 4 | Non-PO invoice > 25,000 TL → cost centre owner approval | | Sales lead qualification | Enterprise Functions | web, api, email | 3 | 3 | Enterprise (> 1,000 employees) or suspected competitor → SDR lead acceptance | | Month-end close and reconciliation (hired accountant agent) | Enterprise Functions | batch, email, api | 3 | 4 | Every posted entry and every filing with the accountant’s approval | | Export control and end-user screening | Defence & Aerospace | api, email, batch | 3 | 4 | Every controlled item → export compliance officer approval; the agent cannot make licence decisions | | Platform maintenance (MRO) work order preparation | Defence & Aerospace | app, api, batch | 4 | 4 | Every work package affecting airworthiness → maintenance planner approval | | Supplier security and compliance assessment | Defence & Aerospace | email, web, api | 3 | 4 | The qualification decision always sits with the facility security manager; the agent only prepares | | Device fleet anomaly triage | IoT & Connected Devices | api, batch, app | 3 | 5 | Bulk device command > 100 devices → fleet operations lead approval | | Staged OTA update rollout | IoT & Connected Devices | api, batch | 3 | 4 | Rollout to a critical fleet segment → product owner approval; automatic rollback if a health gate fails | | Cold-chain excursion response | IoT & Connected Devices | api, app, whatsapp | 3 | 5 | Product transfer or disposal decision → quality manager approval | ## The SDLC agent team Within Code Aginies five roles work together in the CI/CD pipeline. They own the process from request to production; code review and the production decision always stay with people. Every step is traced in Jira, Git and CI records. - Project Manager Agent: triages requests, clarifies scope and dependencies - Analyst Agent: writes acceptance criteria and test scenarios - Architect Agent: produces the design and impact analysis - Developer Agent: implements on a branch and opens the PR - Tester Agent: generates and runs unit and integration tests, reports coverage In the CI/CD pipeline SAST, dependency scanning and staging smoke tests run automatically. Production deployment goes to release-manager approval; for hotfixes the same gates apply on a shortened timeline. ## Live example: an agent at work banking ### Credit card debt payment [Credit card debt payment] run #41,286,855elapsed 0 mstokens 0est. cost $0.000status —Approval rule: Amount > 100,000 → branch manager approval Agents in this run Card Supervisorsupervisorplans, routes, approves Intent Agentaginies-sunkar-1 · on-premUnderstands the request; extracts card, amount and payment type Payment Agentaginies-tulpar-4 · on-premChecks balances, requests OTP, executes the payment in the core Voice Agentaginies-voice-tr-v3 · on-premTranscribes speech and voices the reply on the voice channel Audit trail00:00.00 - ◦Request received, session identity verified · channel: app - ∿Intent: debt payment · Card: ****2210 · Type: full statement balance - ⚙Card debt fetched: 24,860 · due 14 September - ⌘Plan: account selection → OTP → payment → notify - ⚙Accounts listed: current ****7741 has sufficient balance - ✓Policy: amount ≤ 100,000 → no approval; OTP mandatory - ➤OTP sent and verified (2 attempts left) - ⇢24,860 paid from ****7741 to the card · txn #TX-88120 - ➤Spoken and written confirmation given, decision trail stored - ■completed Simulation · derived from real agent definitions · every agent can be built by dialogue with the Autonomous Agent and validated with a test corpus --- # Platform documentation · Documentation URL: https://www.aginies.com/en/docs/platform Language: en Summary: Eight integrated capability layers, the Autonomous Agent, knowledge base, Hub, voice, vision, code, dialogue, observability and workspace management. Documentation On this page A · Orchestration & Execution IntelligenceB · AI / Model OrchestrationC · Knowledge & RetrievalD · Governance, Risk & ComplianceE · Observability & OptimisationF · Integration & ConnectivityG · Multichannel Experience & VoiceH · Deployment & Enterprise ArchitectureAutonomous AgentKnowledge BaseHub Aginies: Model Hub, MCP Hub and proxyVoice AginiesVision AginiesCode AginiesDialog AginiesObservability and reportingWorkspaces, tenants and collaborationPlatform facts derived from the libraryIntegration catalogueOperating model Platform and products # Platform documentation Eight integrated capability layers, the Autonomous Agent, knowledge base, Hub, voice, vision, code, dialogue, observability and workspace management. Updated: September 2026 ## A · Orchestration & Execution Intelligence The engine layer. Runs complex work end-to-end; routes requests intelligently and keeps multi-step processes stable even when dependencies fail. Every run is stateful: steps, intermediate results and decisions are recorded. - Supervisor → sub-agent → supervisor architecture - Dynamic agent routing - Stateful multi-step execution; parallel branches and loops - Agent-to-agent communication (A2A) - Scheduled and webhook triggers - Agent templates and the visual agent builder - Autonomous Agent: build agents by dialogue with test corpora | Scope | Governance | Evidence | | --- | --- | --- | | Supervisor plan, agent assignment and result merge; retry and fallback path on a failed step · Stateful execution: session and long-term memory, pending approvals, timeout rules · Parallel branches and loops; condition-based termination | Each agent’s authority is defined per tool; the supervisor cannot call out-of-scope tools. | The run plan and every step’s input/output in the trace; replayable. | ## B · AI / Model Orchestration Brain selection and resilience. Uses the right model for each task; reduces vendor lock-in and keeps operations running when a provider is unavailable. Switching models does not require redesigning the agent. - Multi-model gateway: proprietary and open-source models behind one interface - Model proxy: observable, masked access to cloud models with your own keys - Model switching without redesign - Fallback model strategy - On-prem serving for open-source models - Prompt testing environment and versioning - Hub Aginies: fine-tuned model families on Qwen, GLM and Kimi bases, current open releases and speech models | Scope | Governance | Evidence | | --- | --- | --- | | Hub catalogue: fine-tuned families on Qwen, GLM and Kimi bases (tulpar, talay, sunkar, konrul) and current open releases · Model proxy: cloud models with the organisation’s own keys, single-point observation and masking · Per-scenario model choice, fallback model and prompt versioning | Model changes are versioned and never reach production without passing the test corpus. | Model version, tokens and cost recorded per request; catalogue history. | ## C · Knowledge & Retrieval Enterprise memory. Turns fragmented enterprise information into governed, searchable, trustworthy knowledge; answers link back to sources. Chunk-level control makes RAG quality directly manageable. - Document embedding pipeline - Versioned knowledge base management - Tag-based document segmentation - Hybrid search and multi-source RAG - Chunk-level inspection and editing - Hallucination-reduction architecture - Source, tag and access-policy management | Scope | Governance | Evidence | | --- | --- | --- | | Document embedding pipeline, chunk inspection and tagging; versioned knowledge base · Access policy at chunk level: who may see which document through which agent · Cited answers; “I don’t know” when there is no source | Knowledge-base changes are approved by the owner; the previous version can be restored. | Chunks and their versions used in every answer; coverage and quality report. | ## D · Governance, Risk & Compliance The control layer. Makes AI auditable, safe and fit for regulated environments; every decision is traceable. Governance is not bolted on; it is part of the architecture. - Guardrails and policy engine - Human-in-the-loop checkpoints: threshold and approver based - Agent decision traceability - Sensitive data masking and PII redaction - Masking and anonymisation of all requests - Full input/output logging and audit trail - Kill-switch and automation limits | Scope | Governance | Evidence | | --- | --- | --- | | Guardrails and policy engine: input/output checks, masking, forbidden actions · Human-approval checkpoints: threshold, approver, timeout; approval card on mobile and web · Kill-switch per user, team and organisation; role-based authorisation | Policy changes under the four-eyes rule; every rule versioned with a reason. | Approval records (who, when, what), guardrail interventions and kill-switch events. | ## E · Observability & Optimisation Performance management. Provides visibility into accuracy, speed, cost and health; produces the data for continuous improvement. Every run is recorded step by step. - Per-run traces and timelines - Latency and token tracking - Cost attribution by scenario, unit and channel - Usage and cost per model; platform health - Routing accuracy measurement - Dashboard and analytics layer - Live logs, conversation records, per-agent success heatmap | Scope | Governance | Evidence | | --- | --- | --- | | Per-run trace: step, duration, tokens, tool call, error · Dashboards: agent success heatmap, platform health, model usage, cost attribution · Forwarding to SIEM and data warehouse; alert rules | Traces are immutable and retained according to policy. | One-click audit sample; monthly quality report per agent. | ## F · Integration & Connectivity Enterprise reach. Connects to enterprise systems and customer channels without rebuilding the existing technology landscape. Each tool carries its own invocation policy. - REST API and database integrations - Model Context Protocol (MCP) servers - Custom tool definitions and invocation policies - Native connectors: email, calendar, files, collaboration tools - Web widget and messaging integrations - MCP Hub: catalogue of approved tool servers | Scope | Governance | Evidence | | --- | --- | --- | | REST, databases and MCP servers; invocation policy per tool · Ready connectors: core banking, ERP, CRM, ticketing, payments, switch · Read/write separation; write tools require approval by default | Each tool defined with a service account and scope; secrets kept in the organisation’s vault. | Tool-call log: parameters (masked), response code, duration. | ## G · Multichannel Experience & Voice The customer interaction layer. Delivers AI across digital, messaging and voice channels on one consistent orchestration underneath. The same agent runs in text and voice. - Real-time speech-to-text and streaming voice processing - Call-centre SIP integration - WhatsApp, SMS and email services - Intent detection and entity extraction blocks - Session memory - Dialog Aginies: dialogue UI and human takeover - Voice Aginies: Turkish STT, TTS and ASR models | Scope | Governance | Evidence | | --- | --- | --- | | Dialog Aginies on web and mobile; WhatsApp, SMS and e-mail channels · Voice Aginies: streaming ASR, Turkish STT/TTS, call centre over SIP · The same agent on every channel; warm hand-off and conversation summary | Identity level and transaction threshold defined per channel. | Transcripts, recordings and summaries under the retention policy; quality score per channel. | ## H · Deployment & Enterprise Architecture Fit to enterprise constraints. Supports the on-premise, hybrid and sovereignty requirements of large enterprises. Data stays where it must. - On-premise and air-gapped deployment - Hybrid architecture: sensitive workloads inside, elastic capacity outside - Private cloud / dedicated region - SSO / SAML, fine-grained RBAC, workspace isolation - Per-tenant agent sharing and collaborative development - GPU-as-a-Service for in-country elastic capacity | Scope | Governance | Evidence | | --- | --- | --- | | Sovereign shared (our data centres in Türkiye), dedicated tenant/GPU, on-prem or air-gapped in the organisation’s data centre · Access: VPN tunnel or allowlist over the internet · In-country elastic capacity with GPU-as-a-Service | Change management and incident notification in the contract; remote maintenance in windows the organisation opens. | Network diagram, tunnel/allowlist configuration, backup and drill records. | ## Autonomous Agent The Autonomous Agent builds agents by dialogue. You describe what you want; it sets up the supervisor and specialist agents, wires tools and the knowledge base, chooses models, defines approval points, writes and runs a test corpus, and deploys after your approval. Every agent in the library can be produced this way. - Describe: objective, channels and approval rule are given in natural language. - Build: supervisor and specialist agents are created; a version number is assigned. - Wire: tools (REST/MCP) and the knowledge base are attached; chunk counts are reported. - Model selection: separate models for intent, decisions and speech; on-prem preferences respected. - Approval points: threshold and approver defined (e.g. amount > 100,000 → branch manager). - A test corpus is written and run; check families are scored. - Human approval: results are presented; the deployment decision always stays with a person. - Deploy: the new version goes live; logs and dashboard are active immediately. A test corpus has three levels: suites (golden path, branches, robustness injection), scenarios within each suite, and multi-turn conversations within each scenario. Every turn is scored against check families. - customer_resolved: was the customer’s request actually resolved - pii_before_verification: was personal data disclosed before identity verification - State machine: did the conversation stay within defined states and transitions - Memory round-trip: was information from earlier turns recalled correctly - Judge dimensions: qualitative criteria such as accuracy, tone and policy adherence - Tool and approval checks: right tool, right parameters, approval triggered above threshold Coverage shows what share of defined intents, branches and tools the corpus exercises. Low coverage is surfaced as a warning before deployment; results are kept per version and compared across versions. The Autonomous Agent never deploys on its own. Deployment happens after human approval. ## Knowledge Base The knowledge base lets agents answer with sources. Documents pass through the embedding pipeline, are split into chunks, tagged and versioned. Each agent knows which sources it may reach, with which tags and under which access policy. - Sources: files, folders, web, databases and connectors - Chunk editor: inspect, edit or disable any chunk - Versioning: which agent ran with which knowledge base version is recorded - Hybrid search: semantic and keyword retrieval combined - Access policy: visibility by tag and role - Citation: every claim in an answer links to its chunk ## Hub Aginies: Model Hub, MCP Hub and proxy The model catalogue holds three groups in one place: Aginies families fine-tuned on enterprise data from Qwen, GLM and Kimi bases (tulpar, talay, sunkar, konrul); the current open releases of the same bases; Turkish speech and vision models. Family names are stable, version numbers move: when a new base release ships, the fine-tune is repeated and the catalogue version increments; the previous version stays available for the contract term. Each scenario gets the model that balances accuracy, latency and cost. The versions below are the October 2026 catalogue. | Model | Base | Where it runs | Typical use | | --- | --- | --- | --- | | aginies-tulpar-4 | Large tier, enterprise fine-tuned | On-prem · in-country GPU | Complex decisions, multi-step reasoning, supervisor role | | aginies-talay-2 | Medium tier, enterprise fine-tuned | On-prem · in-country GPU | General dialogue, summarisation, extraction, long documents | | aginies-sunkar-1 | Small tier, enterprise fine-tuned | On-prem · in-country GPU | Intent detection, routing, guardrails, code completion | | aginies-konrul-1 | Vision-language, enterprise fine-tuned | On-prem · in-country GPU | Document and image understanding, masking | | Qwen · GLM · Kimi (current open release) | Open-weight base models | On-prem · in-country GPU | Tasks that need no fine-tune, benchmarking and fallback | | Cloud models (proxy) | The organisation’s own key | Cloud, via gateway with masking | Proprietary models the organisation prefers; usage and cost tracked from one point | | aginies-voice-tr-v3 | STT trained for Turkish | On-prem | Speech-to-text, call transcription | | aginies-tts-tr-v2 | TTS trained for Turkish | On-prem | Natural speech synthesis, voice agent replies | | aginies-asr-stream-v1 | Streaming ASR | On-prem | Real-time call-centre recognition | Model proxy: you enter your own cloud keys and the platform reaches cloud models on your behalf as a proxy. Every call passes through one point, is observed and masked. Keys are stored encrypted; usage and cost are reported per agent, unit and channel. - Switch models with one selection; the fallback strategy takes over during provider outages - Prompt testing environment and versioning; comparison across versions - MCP Hub: catalogue of approved tool servers; access and invocation policy per server ## Voice Aginies Speech models trained for Turkish: streaming speech-to-text (STT), natural text-to-speech (TTS) and real-time ASR. Voice agents connect through call-centre SIP integration and run on the same orchestration as text-channel agents. - Models: aginies-voice-tr-v3 (STT), aginies-tts-tr-v2, aginies-asr-stream-v1 - On-prem serving; audio never leaves the enterprise - Voice analysis: quality scoring, sentiment and intent - 100% of calls transcribed and scored against the quality form - Personal data is masked during transcription ## Vision Aginies Image and document understanding: invoice, contract and form extraction, damage-photo assessment, visual quality control. Personal-data detection and document masking run on the same models, inside the enterprise. - aginies-konrul-1 · on-prem - Field extraction with source document and confidence score per field - Document classification, independent of layout - Personal-data detection and document masking (KVKK/GDPR) - Low-confidence fields route to human verification ## Code Aginies For engineering teams: IDE extensions and in-house code completion, an enterprise IDE registry that distributes approved extensions and models, and the SDLC agent team working in the CI/CD pipeline. Code never leaves the enterprise. - In-house code completion on on-prem models - IDE Registry: approved extension and model distribution, version control - SDLC agent team: PM, analyst, architect, developer, tester - Works with Jira, Git and CI; code review and the production decision stay with people ## Dialog Aginies The dialogue surface for customers and employees: web widget, mobile, WhatsApp and messaging. Session memory, human takeover and an audit trail for every message. The same agent runs in text and voice channels. - Human takeover and hand-back; context handed to the representative with a summary - Conversation dashboard and quality scoring - Intent detection and entity extraction blocks - Integration with the identity flow; unauthorised requests stop at the guardrail ## Observability and reporting What happens per agent, how the platform is doing and what each model costs: on one dashboard. Every run is recorded step by step; logs and conversation records are reviewed under role-based access. - Live runs: step, duration, tokens, target system and outcome - Per-agent success heatmap and error timeline - Quality and error timelines; routing accuracy - Cost attribution by scenario, unit, channel and branch - Usage and cost per model; platform health - Capacity projection: cost and resource forecast with usage growth - Audit reporting: input, output, model version, rule set, user Reports can be exported and retained in a form suitable for internal audit and regulatory requests. Conversation records are stored masked; unmasked access requires separate authorisation. ## Workspaces, tenants and collaboration The platform is separated by workspace and tenant. Each team manages its own agents, knowledge bases and keys in isolation; approved agents are shared inside the organisation and developed together. - Identity through SSO / SAML; permissions through fine-grained RBAC - Workspace isolation: data, logs and cost reported separately - Per-tenant agent sharing and collaborative development - Templates, users and licences under central management - Agent versions and release history recorded per workspace ## Platform facts derived from the library The figures and distributions on the platform page are derived from the agent library at build time. The same data is given here as tables. | Indicator | Value | Source | | --- | --- | --- | | Live scenarios | 77 | Agent library | | Sectors | 14 | Scenario definitions | | Specialist agents | 261 | Scenario definitions | | Distinct tool definitions | 657 | Agent tool lists | | Systems connected | 200 | Scenario system lists | | Scenarios with a human-approval rule | 73 | Approval rules | | Write steps in enterprise systems | 127 | Step kinds | | Simulated runs / month (total) | 3,320,970 | Scenario metrics (estimated) | | Channel | Scenarios | | --- | --- | | api | 37 | | app | 34 | | web | 29 | | whatsapp | 27 | | email | 27 | | batch | 25 | | voice | 23 | | sms | 5 | | branch | 3 | | Step kind | Steps | | --- | --- | | write | 127 | | tool call | 110 | | reason | 109 | | verify | 89 | | notify | 80 | | ingest | 77 | | human approval | 61 | | plan | 48 | | retrieve | 38 | | Model family | Agent assignments | | --- | --- | | aginies-tulpar | 83 | | aginies-sunkar | 78 | | aginies-talay | 71 | | aginies-konrul | 21 | | aginies-voice-tr | 8 | This table is derived at build time from the agent library’s scenario definitions and changes as scenarios are added. Volume and cost figures are simulation definitions and estimates. ## Integration catalogue The 200 systems the scenarios touch, in 12 kinds. Connecting a new system means one tool definition and one invocation policy; write tools require approval by default. | System kind | Connections | Examples in the library | | --- | --- | --- | | Enterprise APIs | 76 | Access Control, Appointment System, Bank SIM-Swap Signal API, CI Pipeline, CI Runner (GitHub Actions), CI/CD, CI/CD (signed artefacts), Card Management, Card Network, Carrier API, Carrier API (Yurtiçi, DHL), Cart & Checkout API, Civil Registry / Income Services, Connection Portal, Credit Bureau (KKB / Findeks), Credit Bureau (KKB), Customs Broker Portal, Dealer Portal, Dependency Scanner (Snyk), Device / Browser Grid, Device Cloud (MQTT), Enrichment API, Figma, Fraud Engine, Git (self-hosted), GitHub, Insurance Registry (SBM), Insurer Portal, Kubernetes / ArgoCD, Lab / Imaging, MASAK Reporting Portal, Merchant Checkout API, Merchant Portal, Network OSS, OSS Network, OTA Service, Observability (Grafana / Datadog), Open Banking API, Outage Management, Parts & Labour Catalogue (Audatex), Payer Portal, Product Catalogue / Offer Engine, Rating Engine, Repair Network, Repair Network Portal, Review Platform API, Sanctions Screening, Sanctions Screening (Dow Jones), Scheduling, SonarQube, SonarQube + Snyk, Static Analysis, Supplier Portal, Telematics, Terminal Management, Trade Registry (MERSİS), Visa / Mastercard Dispute API, Xcode Cloud, e-Devlet Gateway, e-Government Gateway, e-Invoice Portal, e-Procurement Platform, e-Sign | | Customer channels | 50 | Customer Portal, Hangar Tablet App, Microsoft Teams, Mobile App Chat, OTP / SMS Gateway, Push & SMS Gateway, SMS Gateway, Shift Channel (Teams), Slack, Slack / Teams, Technician App, Web Chat Widget, WhatsApp Business, WhatsApp Business API | | Core systems | 44 | Active Directory / Entra ID, BSS / Billing, Billing (Amdocs), Billing (CIS), CLM (Contract Lifecycle Management), Card Management, Claims System, Core Banking, Core Ledger, Device Management / OTA, EHR (HL7/FHIR), Guidewire ClaimCenter, HIS Scheduling, HLR / Provisioning, Hospital Information System, Hospital Information System (HIS), IoT Platform, IoT Platform (MQTT / Digital Twin), Loan Origination System, Policy Admin, Policy Admin System, SCADA / DMS, SCADA / Historian, Social Assistance Information System, TMS (Transport Management), Underwriting Workbench, WMS | | Databases and vectors | 44 | Audit Log, Collections Platform, Control & Sanctions Lists (vector), Eligibility Criteria (vector), Fleet Registry, GIS / Network Model, GIS / Zoning Plan, KYC / Customer 360, Knowledge Base · kart_sss_v3, Known Issues (vector), Legal Playbook Vector Store, Legislation Vector Store, Mailbox Rules & Checklists Vector Store, Manuals & Work Order Vector Store, Meter Data Management, Observability (metrics), PIM / Catalogue, Payer Policy Vector Store, Product Catalogue (PIM), QA Rubric Vector Store, Regulation Vector Store, Regulation Vector Store (EPDK), SCADA Historian, Security Requirements (vector), Smart Meter Data, Specification Vector Store, Stability Rules (vector), Tariff & Regulation Vector Store, Technical Publications (S1000D vector), Time-series DB, Tolerance Policy Vector Store, Triage Protocol Vector Store, Usage Mediation / CDR Store, Vector Store, Vendor Master / Procurement, Warranty Policy Vector Store, Workforce Work Log | | Ticketing and work orders | 25 | CMMS, CMMS (Maximo), Case Management, Field Service, ITSM / Incident, Jira, Jira (QA), Jira / Xray, Network OSS, QM Scorecard / Case Management, ServiceNow, Ticketing, Trade Compliance Case System, Zendesk | | ERP and MES | 25 | Billing / CIS (SAP IS-U), Billing / Settlement, ERP, ERP (SAP / Logo), ERP (SAP GTS), ERP Billing, ERP Spare Parts, Inventory, MES, MRO System, Maintenance Management System, Oracle EBS, SAP ERP, SAP S/4HANA, SAP S/4HANA MM, Shopify / OMS, Spares Inventory, Supplier Management (SRM), TMS, WMS / Lot Tracking | | E-mail | 20 | E-mail, Email, Email (Exchange), Exchange / M365 Mail, Procurement Inbox | | Document stores | 19 | Confluence, Contract Repository, Document Store, Document Store (S3), Tech Pub Repository (S1000D) | | CRM | 8 | CRM, Customer CRM / Billing, Customer Profile, HubSpot, Salesforce CRM | | Payment systems | 7 | Bank API, PSP / Acquirer API, Payment Gateway (iyzico), Payment Hub, Refund Service, Settlement Ledger | | Voice and switch | 6 | Call Centre SIP, Genesys SIP, IVR / SIP | | HR systems | 3 | LMS, SAP SuccessFactors, Workday | ## Operating model The platform’s standard operating model after go-live. Targets are settled with the organisation in the contract. | Topic | Practice | | --- | --- | | Versioning and rollback | Agent definition, rule set, prompt and model version recorded together; rollback to the previous version is one step. | | Test corpus gate | Every change passes the suite/scenario/turn test corpus before production; the coverage report is recorded. | | Continuity targets | Target RTO 4 hours and RPO 1 hour on the sovereign shared deployment; set with the organisation on dedicated and on-prem. Backups inside Türkiye. | | Model catalogue upkeep | When a new base release ships, the fine-tune is repeated and a new catalogue version opened; the previous one stays for the contract term. | | Incident management | Notification times by severity in the contract; kill-switch with the organisation, root-cause report with Aginies. | | Monthly review | Per-agent quality, cost and approval rates; threshold and rule updates together with the organisation’s administrator. | ## Product screens Platform Aginies · Dashboard · last 30 daysall workspaces · all tenants executions13,120,000 success97.4% p50 latency1,4 s platform healthhealthy busiest agents · each cell ≈ 1 day dialog_banking_assistant%97.4 card_debt_payment%99.1 kvkk_data_detection%96.2 kvkk_document_masking%98.8 motor_claim_end_to_end%98.1 merchant_operations%100 model usage aginies-tulpar-438% aginies-talay-229% aginies-sunkar-119% aginies-voice-tr-v39% cloud · proxy$2.960 cost · by scenario Dialogue banking26% Credit & onboarding24% Claims operations18% KVKK masking14% Merchant ops10% Other8% PII masking: onanonymisation: onaudit trail: full1,920+ active agents Platform Aginies · Reportsall workspaces · October 2026 monthly executions · last 12 monthslast 30 days: 13,120,000+31% YoY | report | breakdown | schedule | format | | --- | --- | --- | --- | | Monthly cost report | scenario · unit · channel · branch | 1st of month · 07:00 | PDF · XLSX | | Agent performance | success, latency, error classes | weekly · Monday | PDF | | Audit export | run traces, approvals, model versions | on demand | CSV · JSON | | Model usage | tokens, cost and proxy share per model | daily · 06:00 | XLSX | | SLA and capacity | p95 latency, queues, GPU occupancy, projection | weekly | PDF | delivered by email and shared folderinternal audit templatepersonal data masked Dialog Aginies · web widgetcard_debt_payment v2 I want to pay my full card balance from my current account. Your statement balance is 24,860, due 14 October. Current account ****7741 has enough balance. If you confirm I will send a one-time code. ✓ get_card_debt · get_accounts · policy: no approval needed Confirm. Code sent. Once entered, the payment executes instantly and you get a confirmation. session memoryhuman takeover: readyaudit trail Type a message…↑ Voice Aginies · SIP · call 0212-…-4471aginies-voice-tr-v3 · on-prem customerSomeone hit my car last night, plate 34 ABC 123, I want to report a claim. agentSorry to hear that. I found your policy by plate; can I take the location and time? STT180 ms intentclaim intake · 0.97 sentimentcalm TTS · aginies-tts-tr-v2ASR · stream-v1PII masking: on Vision Aginies · invoice_2026_1012.pdfaginies-konrul-1 · on-prem [Invoice with detected fields] extracted fields Invoice noFTR-2026-1012 Date12.10.2026 KDV%20 · 3.125,00 Total18.750,00 TL TCKNmasked (KVKK) source: page 1confidence ≥ 0.95 Code Aginies · billing-service/invoices.tscompletion · on-prem 14export async function exportInvoicePdf(id: string) { 15 const invoice = await repo.findById(id); 16 if (!invoice) throw new NotFound('invoice'); 17 const pdf = await renderer.render('invoice', maskPii(invoice)); 18 await audit.log('invoice.export', { id, user: ctx.user }); 19 return storage.put(`invoices/${id}.pdf`, pdf); 20} Tab: acceptPROJ-4821coverage 87%SAST ✓ IDE RegistryVS Code · JetBrains extensionDeveloper Agent · PR #2291 Hub Aginies · Model Huball workspaces · October 2026 | model | access | usage · last 30 days | cost | | --- | --- | --- | --- | | aginies-tulpar-4 | on-prem · GPU 4× | 58% | — | | aginies-talay-2 | on-prem · GPU 2× | 84% | — | | aginies-sunkar-1 | on-prem · GPU 1× | 91% | — | | aginies-konrul-1 | on-prem · vision | 22% | — | | qwen · current release | on-prem · open model | 31% | — | | aginies-voice-tr-v3 | on-prem · STT | 47% | — | | cloud · proxy | your key | 12% | $288 | keys encrypted · never shown to usersall requests maskedMCP Hub · 14 servers Workforce Aginies · Roster and work logworkspace: fintech-ops · this week · tenants separate | agent | role · team | hour-eq. | output | quality | mode | | --- | --- | --- | --- | --- | --- | | ios-dev-01 | iOS Developer · mobile team | 14 h | 9 PRs · 21 points | 4.7 | autonomous · 10% review | | qa-01 | QA Engineer · mobile team | 31 h | 412 tests · 6 bugs | 4.8 | autonomous · 10% review | | ba-01 | Business Analyst · payments product | 9 h | 11 stories · 2 processes | 4.5 | shadow · 100% review | | acct-01 | Accountant · finance | 27 h | 238 entries · 3 reconciliations | 4.9 | autonomous · threshold-approved | work log · today 09:12ios-dev-01PR #4821 opened: payment screen accessibility fixes · 2 tests addedreview pending 09:40acct-0138 supplier invoices processed, 2 out of tolerance → to m.ozkan for approvalapproval 10:05qa-01Regression round 3: 412/412 · 1 new bug (BUG-2291) filed with evidencedone 10:30ba-01Story PAY-118 with 6 acceptance criteria written → to the product ownerreview weekly timesheet exportedbilling: agent-monthteam lead: e.demir (Aginies) --- # Methodology and global patterns · Documentation URL: https://www.aginies.com/en/docs/processes Language: en Summary: The board-level framing, five questions, taker/shaper/maker, the five-layer assessment, the transformation journey, a banking example and the global patterns the agents mirror. Documentation On this page Board-level framing: work and judgmentFive questions boards must answerAdoption strategy: taker, shaper, makerThe five-layer assessmentThe transformation journeyBanking example: the core stays in placeGlobal patterns Processes and global patterns # Methodology and global patterns The board-level framing, five questions, taker/shaper/maker, the five-layer assessment, the transformation journey, a banking example and the global patterns the agents mirror. Updated: September 2026 ## Board-level framing: work and judgment AI is the first technology that automates judgment as well as work. Earlier waves accelerated data entry, calculation and communication and always left the decision to people. Agents classify, prioritise, recommend and, within defined limits, decide. That is why this is an operating-model redesign: which decisions are delegated to agents, which stay with people, how the limits are measured and audited. The answers belong at board level. ## Five questions boards must answer - Value pool: where and how large is the value AI can create in the enterprise? - Processes: which processes will be redesigned with agents, which merely accelerated? - Organisation and talent: how do roles, skills and structure change over the next 24 months? - Adoption strategy and platform: taker, shaper or maker; which platform choices follow? - Governance: how are decisions bounded, audited and stopped when necessary? ## Adoption strategy: taker, shaper, maker Every enterprise chooses what to build itself. The right question is which role to take in which area. Most enterprises mix the three strategies scenario by scenario. | Strategy | What it means | When it fits | | --- | --- | --- | | Taker | Use ready-made AI capabilities as they are | Standard, non-differentiating work; quick benefit | | Shaper | Build agents on a platform with the enterprise’s data and processes | Core processes that create advantage; sovereignty requirements | | Maker | Develop own models and infrastructure | Very large scale, unique data and strong engineering capacity | Aginies is the platform for the shaper strategy: the enterprise builds its agents with its own data, processes and limits, and makes model and infrastructure choices without lock-in. ## The five-layer assessment We decide with numbers which AI ideas are worth building and when to start. Every scenario passes through five layers, and each layer ends in a clear decision: proceed, fix or stop. | Layer | Question | Output | | --- | --- | --- | | 1 · Value | Why should this be done at all? | A business case in numbers and a prioritised scenario portfolio | | 2 · Readiness | Do we have the inputs we need? | A readiness scorecard with gaps closed before the project | | 3 · Feasibility | Can we actually build this? | A realistic MVP plan with clear scope and identified risks | | 4 · Operations | How will it live after go-live? | A living, measured, continuously improving AI product | | 5 · Governance | Is it secure, controlled and auditable? | An audit-ready, controlled, trustworthy AI operation | Value layer checks: - Problem: which business problem, how is it solved today, what does it cost? - Benefit: cost reduction, revenue, risk reduction, experience; which and how much? - Measurement: target KPIs, expected improvement, ROI and payback period - Strategic fit: aligned with company strategy, a management priority? Readiness layer checks: - Data: sources available and digital; quality, freshness and ownership clear? - Integration: systems and APIs ready; real-time access possible? - Organisation: does the business unit own it; sponsor and experts identified? - Technology: cloud/security infrastructure, AI platform and licences ready? Feasibility layer checks: - Technical fit: should AI solve this at all; classic software, classic ML, agents? - Complexity: how many systems, data sources and users are affected? - Model performance: expected accuracy, impact of a wrong output, human approval needed? - Project feasibility: MVP duration, team, external dependencies, technical risks Operations layer checks: - Operating model: product owner, process owner, operations and support defined - User management: training, adoption tracking, feedback loops - Monitoring: performance, metrics and error management in real time - Financial operations: monthly run cost, growth projection, budget ownership Governance layer checks: - Security: sensitive data map, data residency, control of data flowing to models, masking - Compliance: KVKK and GDPR impact, sector regulation, internal policy requirements - Risk: assessment of hallucination, wrong decisions, data leakage and prompt injection - Control: human approval points, automation limits, kill-switch, decision history Proceed: the foundation is in place on all five layers. Fix: the scenario is valuable, but one or two layers have gaps to close before the project. Stop: starting as-is would burn budget; saying so is part of the methodology. Duration, participants, tools and red flags of each layer: | Layer | Duration | Participants | Tools and outputs | Red flags | | --- | --- | --- | --- | --- | | Value | 1–2 weeks | Business owner, finance, sponsor | Scenario portfolio workshop and value card (problem, volume, cost, KPI) · Comparison with the closest scenario in the agent library | Benefit without numbers: “efficiency will rise” but no volume or cost · No sponsor; the request comes from IT and the business watches | | Readiness | 1–2 weeks | Data owner, integration team, process owner | Readiness scorecard: data, integration, process, organisation (each axis 0–2) · API and data-access test; sample data set | The target system has no API, only a screen · Process rules live only in people’s experience | | Feasibility | 2–3 weeks | Aginies engineering, enterprise architecture, information security | Approach decision: classic software, classic ML, agents, generative AI · Prototype and draft test corpus; wrong-output impact analysis | A wrong output is irreversible and no approval point is defined · Scope grows at every meeting | | Operations | 1 week (plan) + continuous after go-live | Product owner, process owner, operations and support | Operating-model template: roles, thresholds, review sample, budget owner · Dashboards and the monthly review calendar | Nobody owns operation after go-live · No budget owner for the monthly run cost | | Governance | 2 weeks, in parallel with the other layers | Information security, compliance, legal, internal audit | Data map and masking rule set; approval matrix · Control set and evidence list; sector regulation mapping | Personal data involved, no map · Governance postponed as “we will add it later” | Example: how the dialogue banking assistant passed the five layers: - Value: Simple transactions reaching the call centre and cost per transaction put in numbers; target 40% reduction. Sponsor: digital banking. - Readiness: Transaction APIs and a test environment for core banking and cards existed; OTP stack ready; policy thresholds written with information security. - Feasibility: Approach: supervisor + four specialist agents; per-transaction OTP and approval band against wrong-output risk; test corpus of 212 scenarios. - Operations: Product owner digital channels, process owner operations; monthly review and threshold updates on the calendar. - Governance: Personal-data map and masking rules; limit increase > 60%, dispute > 5,000 and new payee to approval; all systems in Türkiye. ## The transformation journey The assessment repeats inside a four-stage journey. Each stage’s output is the next stage’s input, and every stage progresses on the same platform. - Discover and portfolio: value pools, candidate scenarios, a prioritised portfolio and sponsors. - Assess: each candidate passes the five layers; a proceed, fix or stop decision. - MVP and validate: the agent is built, the test corpus run, KPIs validated on real data. - Production and scale: staged rights, monitoring, cost attribution; move to the next scenario in the portfolio. ## Banking example: the core stays in place In a bank the core system stays in place; the AI orchestration layer sits on top of it. Core services are agentised through MCP and REST; agents see account, card, credit and payment services as tools. Rights open in stages. - Read: balance, statement, limit and transaction queries; measurement starts with zero write risk. - Controlled write: limit increases, transfers and instructions written under threshold and approval rules. - Autonomous: proven scenarios run without approval within defined limits; the kill-switch stays ready. - Conversational banking: an assistant that transacts in the app and by voice - Core-banking AI transformation: staged agentisation of core services - Real-time task tracking in operations: request, SLA and bottleneck visibility - Customer satisfaction management: conversation analysis, complaint and vulnerability signals Every scenario is observable and costed: run-level traces, cost attribution by unit, channel and branch, quality and error timelines, capacity projection and audit reporting. The compliance frame: banking IT regulation (BDDK), primary and secondary systems in-country, kill-switch and mandatory human approval on critical decisions. ## Global patterns The agents in the library mirror practices proven around the world. The list below summarises those patterns; volume and cost figures are simulation definitions. - FNOL automation: large carriers opening claims at first contact without call-centre handling load. - AP automation: shared-service centres running three-way matching from invoice inbox to ERP posting. - KYB onboarding: payment institutions completing self-serve merchant onboarding in minutes. - Agentic SDLC pipelines: large software organisations delivering from request to production with agents. - WISMO deflection: e-commerce converting “where is my order” contacts to self-service through proactive notification. - AML alert triage: alerts enriched and narratives drafted, with the decision left to the analyst. - Outage communications: operators and distribution companies cutting inbound call volume with proactive updates. - Conversational banking: in-app assistants that complete the transaction rather than only inform. - 100% QA coverage: BPOs and bank contact centres scoring every call instead of a sample. - Prior-authorisation automation: health systems and payers automating document collection and rule checks. - Digital front door: patient triage and symptom assessment automated with clinical decisions kept with clinicians. - Zero-touch onboarding: large enterprises coordinating accounts, equipment and training automatically. --- # Security, compliance and sovereignty · Documentation URL: https://www.aginies.com/en/docs/security Language: en Summary: Deployment options, the compliance matrix, model governance, data residency and what is shared in a discovery session. Documentation On this page Deployment optionsCompliance matrixModel governanceData residency and sovereigntyThe code-assistant scenario: from IDE to modelWhat is shared in a discovery sessionSector regulatory frames Security and compliance # Security, compliance and sovereignty Deployment options, the compliance matrix, model governance, data residency and what is shared in a discovery session. Updated: September 2026 ## Deployment options AI must meet the same standards as every other critical system. Aginies is built to the requirements of regulated sectors such as finance, insurance and healthcare; data stays where it must. Four deployment models deliver the same platform. | Option | Where data lives | Where models run | When it fits | | --- | --- | --- | --- | | On-premise / air-gap | The enterprise’s own data centre; no external connection | oss and speech models served inside | The most sensitive data; regulatory obligation | | Hybrid | Sensitive workloads inside | oss inside; outside only with masked data through a secure gateway | Elastic capacity needs; a mixed scenario portfolio | | Private cloud / dedicated region | A single-tenant region reserved for the enterprise | Inside the region; data residency guaranteed | Enterprises that want residency guarantees without running a data centre | | GPU-as-a-Service | Inside national borders | Elastic GPU capacity that scales with usage | On-prem model serving without capital outlay | ## Compliance matrix Aginies is a SOC 2 Type II and ISO 27001 compliant platform; the control set is documented for internal audit. The matrix below summarises the compliance areas and how the platform addresses each. | Area | Status | What it covers | | --- | --- | --- | | SOC 2 Type II | Compliant platform | The control set is documented for internal audit | | ISO 27001 | Compliant platform | Information security management | | KVKK & GDPR | Compliance | Personal data map, data residency guarantee, deletion and access requests | | BDDK IT regulation | Compliant deployment | For banks, primary and secondary systems in-country; audit-ready deployment | | SSO / SAML & RBAC | Built in | Fine-grained roles, workspace isolation, per-agent access policies | | PII masking & anonymisation | Built in | Masking at ingestion, configurable redaction, anonymisation of all requests | | Audit trail | Built in | Input, output, model version, rule set and user for every agent action and model call | | Kill-switch | Built in | Emergency shutdown and automation limits on critical decisions; mandatory human approval | “Compliant platform” means the control set is designed and documented against the relevant standards; independent certification is a separate process and is documented separately. ## Model governance Model-agnostic, fine-tuned. Hub Aginies holds enterprise fine-tuned versions of proprietary models, open-source-based on-prem models and Turkish speech models in one catalogue. There is no vendor lock-in; switching models does not mean rewriting the agent. - Model proxy: cloud models reached with the enterprise’s own keys through one point - Keys stored encrypted; usage reported per agent, unit and channel - Every request to an external model is masked and anonymised - Fallback model strategy: operations continue through provider outages - Prompt versioning: which agent ran with which prompt and model version is recorded - On-prem oss models for sensitive scenarios; data never leaves the enterprise ## Data residency and sovereignty Sovereign design means control of data and models stays with the enterprise. In on-premise and air-gapped deployments no data leaves; in hybrid deployments only masked data goes outside, through a secure gateway. GPU-as-a-Service provides elastic capacity inside national borders. - Personal data map: which data, in which system, for what purpose, for how long - Data residency guarantee: contractual in dedicated-region and on-prem options - For banks, primary and secondary systems in-country - Deletion and access requests: recording and response aligned with KVKK and GDPR - Voice and code data: Voice and Code Aginies run on-prem ## The code-assistant scenario: from IDE to model For the information-security teams of organisations that want to work in a closed circuit. The organisation’s developers reach Hub Aginies through the Code Aginies extension in their IDEs, over the organisation’s own VPN, on our infrastructure in data centres located in Türkiye and aligned with banking IT regulation; the coding models run there, on in-country GPUs. Code context never leaves Türkiye, no external model is called, no model is trained on the organisation’s data, and every request leaves a per-user trace. [Request path from organisation IDEs over a VPN tunnel to Hub Aginies inside the Aginies sovereign perimeter, then to in-country GPU model servers; no external model calls and no training.] - Identity: The developer signs in from the IDE with the organisation’s identity provider (SAML / OIDC, MFA). The extension carries a short-lived per-user token; the licence key is held at organisation level and stays outside the extension. - Request: The extension sends only the relevant code context (open file, cursor surroundings, selected symbol definitions) and the prompt; the repository as a whole is never sent. The connection is encrypted with TLS 1.3 inside the organisation’s VPN tunnel; the Hub accepts traffic only from the IP addresses allocated to the organisation. - Licence and quota: Hub Aginies validates the token, binds it to the organisation’s workspace and checks seats and quota. When a user is disabled at the organisation, access drops at the same moment. - Masking and routing: The prompt is scanned with configurable rules for secrets (keys, passwords, connection strings) and personal data; matches are masked. The prompt is routed to the model selected in the contract. - Model: The model runs on GPUs in Aginies’ data centre in Türkiye. There is no external API call; prompt and response are not stored on the model side and are not used for training. - Response and trace: The suggestion returns to the IDE through the same tunnel. Platform Aginies records the request with user, time, model version, duration, tokens and cost; whether prompt and response content is retained, and for how long, is decided together with the organisation. - The organisation’s visibility: The organisation’s administrators see usage and cost dashboards; traces can be forwarded one-way to the organisation’s SIEM via syslog or API. The kill-switch cuts access instantly per user, team or organisation. | Layer | Control | Evidence / audit | | --- | --- | --- | | Identity | SSO with the organisation’s IdP (SAML / OIDC), MFA at the organisation; user- and group-level authorisation; instant revocation | Session and authorisation logs; seat list with the organisation’s administrator | | Network | Two access models: in the VPN model a site-to-site tunnel and a Hub unreachable from outside; in the internet model TLS 1.3 and the organisation’s egress IPs on an allowlist (rate-limit and auto-block exemption) | Tunnel or allowlist configuration; connectivity tests; network diagram with IP addresses | | Application | Signed extension, approved-version distribution from the IDE Registry; version pinning | Signature verification; release notes; SBOM on request | | Data | Secret and PII masking; context limiting; encryption (TLS in transit, AES-256 at rest); retention period by contract | Masking rule set; retention policy; deletion evidence | | Model | Served on in-country GPUs; no external calls; no training; versioned models, validated with a test corpus, fallback model | Model catalogue and version history; test reports | | Tenant | The organisation’s workspace is logically isolated; dedicated tenant or dedicated hardware on request | Isolation design; access matrix | | Observability | Per-request audit trail; usage and cost dashboards; SIEM forwarding; kill-switch | Dashboards; log samples; forwarding integration | | Operations | Incident notification, change management, backup and disaster recovery inside Türkiye | Incident response plan; RTO/RPO; drill records | | Compliance | SOC 2 Type II and ISO 27001 compliant control set; KVKK; deployment aligned with banking IT regulation | Control set document; penetration-test summary; audit right in the contract | Two access models: VPN tunnel or allowlist over the internet. The same request path opens in two ways. A VPN tunnel when the organisation wants a fully closed circuit; TLS over the internet with the organisation’s egress IPs on an allowlist when developers should start without installing anything extra. The second model was set up together with a bank’s information-security team; the diagram and flow table below are the anonymised version of that deployment. - VPN tunnel (fully closed circuit): Site-to-site IPsec between the organisation’s gateway and Aginies; the Hub is reached only through the tunnel from the organisation’s allocated IPs and is invisible from the internet. - Over the internet, with an allowlist (quick start): The endpoints are reachable from the public internet; the extension connects over TLS 1.3 from the organisation’s existing internet egress. The organisation’s egress IPs go on an allowlist so that rate-limit and auto-block rules do not cut its traffic; access itself is granted by identity. [Request path from organisation IDEs over the internet with TLS to the Aginies application server, then to the GPU model server on the internal network; no client VPN, organisation egress IPs on an allowlist, no external model calls.] | Flow | Source | Target | Port / protocol | Access and control | | --- | --- | --- | --- | --- | | 1 | Developer IDE · Code Aginies extension · organisation egress IPs | code-api.aginies.com → Aginies public IP | TCP 443 · HTTPS, TLS 1.3 | Internet; organisation IPs exempt from rate limit and auto-block. SSO token, seats and quota, masking, per-request trace. | | 2 | Aginies application server (internal interface) | GPU model server (internal network) | Internal network · model API | Internal network only; accepted only from the application server, never leaves the data centre. | | 3 | GPU model server | Internet | — | None: the server has no public IP and no internet egress; prompt and response are not stored or used for training. | | 4 | Aginies operations team | Aginies application server · administration | TCP 22 · SSH | Aginies operations VPN only; closed to the internet and to the organisation. No SSH or VPN needed on the organisation side. | | 5 | Aginies application server | External model providers | — | None: coding models run only on in-country GPUs; code context never leaves Türkiye. | | 6 | Organisation applications (optional, direct model access) | hub-api.aginies.com → Aginies public IP | TCP 443 · HTTPS, TLS 1.3 | Same allowlist, identity and quota; API access to the models in the Hub catalogue. | Addresses in the diagram are anonymised; the real IPs, hostnames and firewall rules are given to the organisation’s information-security team in the technical annex. ### Location and infrastructure **Where do the systems run? Does data leave Türkiye?** The whole platform runs in our data centres in Türkiye; primary and secondary systems are in-country. Code context, logs and backups never leave Türkiye. This is aligned with the in-country requirement of banking IT regulation. **Which data centre, and to which standards?** The systems run in data centres located in Türkiye that hold international Tier and national TS EN 50600 certifications and are aligned with financial-sector regulation (banking and payments regulators). The data centre’s name, documents and certificates are shared in the discovery session and in writing in the contract annex. **A dedicated environment or a shared one?** The default is a logically isolated workspace for the organisation on sovereign shared infrastructure. At the organisation’s choice a dedicated tenant, a dedicated GPU pool or an on-prem deployment in the organisation’s own data centre is possible; the architecture is the same in all three. **Who are the sub-processors? Do you use a cloud provider?** No cloud provider is used for model serving. Data-centre services come from providers located in Türkiye and approved by the regulators; the provider is named in the contract and the organisation is informed in advance of any change. ### Network and connectivity **How is the connection set up? Which ports and addresses?** There are two options. In the VPN model a site-to-site IPsec tunnel is established between the organisation’s gateway and Aginies, only HTTPS (443) is used inside the tunnel and the Hub accepts traffic only from the IP addresses allocated to the organisation. In the internet model the endpoints are reachable over TLS 1.3 from the public internet; the organisation opens a single outbound rule (Aginies public IP, TCP 443) and its egress IPs are added to an allowlist. The allowlist keeps rate-limit and auto-block rules from cutting the organisation’s traffic; access itself is granted by identity. In both models the IP addresses and parameters are given in the technical annex. **Is Hub Aginies exposed to the internet?** No. The Hub is reachable only through the tunnel interface; it has no public endpoint, so there is no DDoS or scanning surface. **How is encryption handled?** TLS 1.3 in transit inside the VPN tunnel (end-to-end, independent of the tunnel); AES-256 at rest. Key management is on the Aginies side; bring-your-own-key is available in the dedicated deployment. **Can the organisation’s proxy or SSL inspection sit in between?** The extension honours corporate proxy settings. TLS inspection with the organisation’s own certificate is technically possible; whether the inspection device sees code context is then the organisation’s own policy. Our recommendation is to keep end-to-end TLS inside the tunnel. ### Identity and access **How are users authenticated? What if the licence key is shared?** Identity lives in the organisation’s identity provider: SSO via SAML or OIDC, MFA by the organisation’s policy. The licence key is organisation-level and grants no access on its own; every request carries a short-lived per-user token. When a user is disabled at the organisation, Aginies access drops at the same moment. **Who can use which model, and how much?** Authorisation is group-based: which team reaches which model, with which quota and which features (completion, chat, agents); the organisation’s administrator manages it from the Platform Aginies dashboard. Seats and quotas are set by contract. **Can Aginies staff access the organisation’s data?** Production access is role-based, named and logged; access to prompt and response content is opened only on the organisation’s written request for support purposes, and the record is shared with the organisation. Under the default retention policy, content is not kept at all. **Do administrative actions leave a trace?** Yes. Administrative actions are written to the same audit trail: who changed what, and when. Policy changes are versioned. ### Data flow and retention **What exactly does the extension send?** Only the context the request needs: the relevant part of the open file, cursor surroundings, selected symbol definitions and the developer’s prompt. The repository as a whole, build outputs and environment files are never sent. The context limit and the exclusion list (.env, certificates, key files) are configured with the organisation and visible in the extension. **Do secrets (passwords, API keys) reach the model?** The masking layer in the Hub scans the prompt for keys, passwords, connection strings and personal-data patterns; matches are masked before they reach the model. The rule set can be extended with the organisation’s own patterns; masking events appear in the trace. **Are prompts and responses retained? For how long?** Metadata (user, time, model version, duration, tokens, cost) is always recorded. Retaining prompt and response content is the organisation’s decision: off (default), or encrypted retention for a period the organisation sets. Deletion at expiry is automatic and evidenced. **Is our data used to train models?** No. The organisation’s prompts and responses never enter any model’s training or fine-tuning data; this is an explicit clause in the contract. If organisation-specific fine-tuning is requested, a separate model version owned by the organisation is produced only from data the organisation approves. **Where are backups, and who can access them?** Backups are kept inside Türkiye, encrypted and under a separate access policy; if content retention is off, backups contain no content either. Backup retention and disaster-recovery targets are written in the contract annex. **What happens when the contract ends?** The organisation’s workspace, configuration and traces are exported in the requested format; all data is then deleted within the agreed period and deletion evidence is provided. An organisation-specific model version is handed over on request, otherwise destroyed. ### Model and AI **Which models are used? Does anything go to an outside provider?** Coding models from the Hub Aginies catalogue served on in-country GPUs: open-source-based on-prem models (aginies-talay-2, aginies-sunkar-1) and fine-tuned enterprise versions. No prompt goes to an outside provider; the cloud model proxy is off in the organisation’s workspace and can be locked at policy level. **Who selects the models, and how are they updated?** Models enter the catalogue after Aginies reviews them and validates them with a test corpus; every model is versioned. A model change in the organisation’s workspace happens with the organisation’s approval and prior notice; the previous version is kept for rollback for an agreed period. **Is model output safe? Can it produce harmful code?** Model output is a suggestion; the developer reviews the code and it passes the organisation’s existing CI, code review and static analysis. On the Aginies side, check families for prompt injection and off-policy output run in the test corpus; organisation-specific forbidden patterns can be added as policy. **Licensing and IP: who owns the generated code?** The generated code belongs to the organisation. The licences and commercial-use terms of the open-source models in use are listed in the contract annex. ### Extension and client side **Where is the extension installed from, and is it signed?** The extension is signed and distributed as an approved version through the organisation’s enterprise IDE Registry; no public marketplace is needed. Versions can be pinned; updates are released with the organisation’s approval. **Does the extension connect anywhere else? Is there telemetry?** The extension connects only to the Hub address the organisation configures; there is no other endpoint. Error-report telemetry is minimal, carries no content and can be switched off. **What if the connection drops? Does it work offline?** Without a connection the extension goes quiet; it keeps no local queue or cache, so no code context accumulates on the device. It resumes when the connection returns. **Can we review the source code or an SBOM?** The extension’s software bill of materials (SBOM) and dependency scan are shared on request; a source-code review can be arranged under an NDA. ### Compliance, audit and operations **Which certifications do you hold?** The platform is compliant with the SOC 2 Type II and ISO 27001 control sets; the control-set document is shared for internal audit. The KVKK data-processor agreement and privacy notices are ready. **For banks: is this an outsourced support service under banking regulation? Is there an audit right?** If the organisation is a bank, the service may fall under its outsourced-support classification; the required contract clauses (audit right, sub-processor notification, exit plan, information-security obligations, audit access for the regulator and the organisation) are then added. The in-country requirement is already met by the architecture. **Are penetration tests performed? Can we see the results?** Independent penetration tests are performed regularly; the executive summary is shared and finding-closure evidence is provided. The organisation may run its own test; scope and timing are agreed in advance. **What is the notification time for a security incident?** For an incident affecting the organisation, first notification is made within 24 hours, followed by a root-cause and corrective-action report. KVKK notification duties are coordinated with the organisation as data controller. **Service level and capacity?** Availability and latency targets are written in the contract; capacity is reserved by the organisation’s seat count and the GPU pool scales in-country. Maintenance windows are announced in advance. **Can we cut the service instantly in an emergency?** Yes. The kill-switch is in the organisation administrator’s hands: it cuts access instantly per user, team or organisation; the organisation can also close the VPN tunnel on its own side to end all access. **How does change management work?** Every change to the Hub, the extension or the models that affects the organisation is announced in advance with release notes; critical changes can be tied to the organisation’s approval. Policy and configuration changes are versioned in the audit trail. The same architecture applies to three deployments: sovereign shared infrastructure, dedicated tenant / GPU, on-prem in the organisation’s data centre. ## What is shared in a discovery session A discovery session is a working meeting to understand the enterprise’s scenarios and constraints. The following materials are shared so that security and compliance teams get their answers early. - Reference architecture and a detailed schema of the deployment options - Documentation of the SOC 2 Type II and ISO 27001 compliant control set for internal audit - Data flow and masking design: which data reaches which model, in what form - Model catalogue and proxy setup; key management and fallback strategy - A sample audit report and run trace - The five-layer assessment scorecard for the selected scenarios If the enterprise’s sector-specific requirements are shared before the session, the materials are prepared for those constraints. ## Sector regulatory frames The regulatory frame and recommended deployment model from each industry page. The first-90-days plan and measured targets are on the industry page. | Sector | Frame | Recommended deployment | First measured target | | --- | --- | --- | --- | | Banking | Banking IT and electronic banking regulation: primary and secondary systems in Türkiye, audit right in the outsourcing contract, cloud model subject to bank-specific permission · AML and personal-data law: suspicious-transaction reports decided by analysts, masking and purpose limitation in every scenario · Per-request audit trail for internal audit and regulator review: user, model version, data source, reasoning | Start with the sovereign shared deployment in our data centres in Türkiye (isolated bank workspace, VPN or allowlist access); on-prem in the bank’s data centre for scenarios classed as primary systems. | Simple transactions reaching the call centre: 40% fewer | | Fintech & Payments | Central-bank payment and e-money IT communiqué: community cloud only with an approved provider; primary systems in-country · AML obligations: transaction-monitoring alerts to the analyst with evidence, the reporting decision with people · Card-scheme rules and personal-data law: chargeback deadlines tracked, card data masked and never used beyond purpose | In the community cloud of a provider on the central bank’s approved list, or in the institution’s own data centre; GPU capacity in-country, models from the Hub catalogue. | Merchant request first-contact resolution: 70% and above | | Insurance | Insurance regulator rules and insurance IT requirements: claim and policy decisions reasoned, time-stamped, open to audit · Special-category personal data: health and damage photos processed inside the perimeter, masking and access logging mandatory · Payment and rejection decisions always with the authorised adjuster or underwriter; the agent prepares the file | Dedicated region in Türkiye or the sovereign shared deployment; vision models for damage photos and health data on in-country GPUs. | Simple claim closure time: From weeks to days | | Retail & E-commerce | Consumer law and distance-selling rules: withdrawal, return and delivery deadlines coded into agent rules · Personal-data law and marketing consent: contact only on consented channels, recommendations on purpose-limited data · Campaign and price rules: discount authority bounded by the matrix, out-of-band offers to customer service | Sovereign shared deployment; GPU capacity scales on usage at seasonal peaks, customer data stays in Türkiye. | First-contact resolution: +15–25 points | | Telecom | Telecom regulator rules and the electronic communications law: subscription, contract and number-porting operations recorded and traceable · Personal-data law and traffic data: subscriber data in Türkiye, recordings and transcripts masked · SIM swap and identity operations: extra verification and human approval; suspicious requests to the fraud team | Sovereign shared deployment or a dedicated tenant in the operator’s own data centre; speech models on in-country GPUs, connected to the existing switch over SIP. | Requests resolved without a representative: 60–70% | | Healthcare | Personal health-data regulation and special-category data: health data inside the hospital perimeter, access by role and purpose, every access logged · Ministry hospital-information and national health-record integration rules: the agent acts through the HIS and keeps no separate data copy · Clinical decisions with the physician: the agent prepares, queries and organises; physician approval card when in doubt | On-prem in the hospital data centre; models inside, no external connection. For hospital groups, a dedicated tenant in the group data centre. | No-show rate: 30–40% lower | | Manufacturing & Automotive | OT security (IEC 62443 approach): agents connect to the production network read-only and issue no commands; OT/IT separation kept · Product safety and warranty law: warranty decisions rule-based, out-of-tolerance files to people · Personal-data law and supplier contracts: supplier and customer data inside the organisation, sharing limited to the contract scope | Hybrid: read-only agents and historian access near the production network on site, decision and document agents in our data centre in Türkiye or the organisation’s own. | Unplanned downtime: 20–30% lower | | Logistics | Customs law: declaration and document preparation by the agent, declaration and signature with the authorised broker · GDP and cold-chain rules: temperature excursions documented, the decision with the quality officer · Personal-data law and customer contracts: shipment and consignee data inside the organisation, every customer message recorded | Sovereign shared deployment; TMS, telematics and carrier APIs connected through an allowlist, shipment data stays in Türkiye. | Delay detection lead: Hours ahead | | Energy & Utilities | Energy regulator consumer-service and connection regulations: dispute, correction and application deadlines coded into agent rules · Critical infrastructure and OT separation: read-only access to SCADA and OMS; grid decisions in the control room · Personal-data law and meter data: consumption data only for the customer’s own account, retention by policy | Sovereign shared deployment for customer operations; grid-adjacent read-only agents on-prem in the company data centre with critical-infrastructure separation kept. | Calls resolved without a representative: 55–65% | | Government & Public Sector | Presidential information and communication security guide: deployment in the institution’s data centre, air-gap included; open-weight models inside · Administrative procedure and public procurement law: every output with the regulation article and source document; the decision with the authorised officer or commission · Personal-data law and e-Government integration rules: citizen data purpose-limited, access by authorisation, full audit trail | On-prem in the institution’s own data centre; air-gapped where required. For municipalities and affiliated bodies, a dedicated tenant in the public cloud. | Documents asked of the citizen again: 70% fewer | | Defence & Aerospace | Defence-industry agency and export-control law: control-list mapping by the agent, licence and sharing decisions with the authorised compliance officer · Facility security clearance and classification levels: access separation, marking checks and an audit log of every access · ITAR/EAR-like restrictions in allied programmes: controlled technical data only to cleared users, no egress to external networks | Air-gapped deployment: models, platform and data inside the closed network; updates as approved packages, remote maintenance only in windows the organisation opens. | Party screening time: From days to minutes | | IoT & Connected Devices | Personal-data law and device data: telemetry inside the organisation, user data purpose-limited · Product safety and cyber-resilience requirements (incl. EU CRA for exporters): OTA packages signed, rollout recorded, rollback ready · Safety-critical devices: read-only; commands and bulk operations under product-owner approval | Hybrid: device cloud and time-series data on the existing infrastructure, agents and small models in our data centre in Türkiye or the manufacturer’s own. | Faults needing a field visit: 35% fewer | | Software Delivery | Source code and intellectual property: code never leaves the organisation; Code Aginies and completion models on-prem · Secure development requirements (incl. regulator expectations in regulated sectors): SAST, dependency and secret scanning on every PR, evidence stored per release · Production deployment and architecture decisions with people; the agent writes only in assigned repositories and branches | Code Aginies and coding models in the organisation’s data centre or in our data centre in Türkiye through the closed-circuit deployment; SDLC agents connected to Jira, Git and CI/CD with service accounts. | PRs accepted on first review: 60–75% | --- # Industries URL: https://www.aginies.com/en/industries Language: en Summary: Aginies for banking, fintech, insurance, retail, telecom, healthcare, manufacturing, logistics, energy, government, defence, IoT and software delivery: AI orchestration built around each sector’s regulation, systems and scenarios. Industries # Thirteen sectors, sixty agents: every sector has its own systems, rules and approval points. The same platform, with scenarios built for the sector. In regulated industries compliance is not bolted on; it is part of the architecture. Closed networks in defence and government, small on-prem models in IoT, read-only access to grid systems in energy: each sector deployed on its own terms. Finance and insurance The most tightly supervised sectors, with the clearest approval points. ## Banking 9 agents The core stays where it is; a deciding, acting AI layer goes on top. The PoC gapLegacy coreInvisible cost 4 compliance principles · 3 solutions Read more → ## Fintech & Payments 5 agents Onboarding, AML, credit decisions and reconciliation: agents where speed and compliance are needed at once. Onboarding bottleneckAlert fatigueSecond-level credit decisions 4 compliance principles · 3 solutions Read more → ## Insurance 6 agents From first notice of loss to recovery: processes with tracked steps and defined approval points. Storm-day capacityDocument chaosUnexplainable decisions 4 compliance principles · 3 solutions Read more → High-volume customer operations Millions of interactions, thousands of scenarios: agents between channel and core system. ## Retail & E-commerce 6 agents Behind every customer question there is a transaction; the agents complete it inside the order system. Seasonal peaksBots that answer but do not actCatalogue debt 3 compliance principles · 3 solutions Read more → ## Telecom 5 agents Millions of subscribers, thousands of scenarios: billing, outages, fraud and offers on one layer. Bill-dispute volumeSIM-swap fraudOutage communication 3 compliance principles · 3 solutions Read more → ## Energy & Utilities 5 agents Outages, meters, connection applications and demand response: agents connected to grid systems, talking to customers. Outage call surgeEstimated-bill disputesDistributed generation applications 3 compliance principles · 3 solutions Read more → ## Healthcare 4 agents The clinical decision stays with the physician; agents carry prior authorisation, intake and scheduling. Prior-authorisation delayIntake and triage loadNo-shows 3 compliance principles · 3 solutions Read more → Operations and infrastructure Agents that decide and dispatch on top of ERP, MES, TMS and IoT platforms. ## Manufacturing & Automotive 4 agents From supplier invoice to maintenance request: agents that decide and open work orders on top of ERP and MES. Disconnected systemsMaintenance backlogInvoice vs receipt gaps 3 compliance principles · 3 solutions Read more → ## Logistics 3 agents Shipment exceptions, customs documents and cold chain: agents that decide the moment an event happens. Exceptions surface lateCustoms document errorsCold-chain visibility 3 compliance principles · 3 solutions Read more → ## IoT & Connected Devices 4 agents Agents that reduce signals from thousands of devices to one incident and one decision: anomalies, OTA, cold chain. Alarm noiseRisky OTA rolloutsSensor-to-action gap 3 compliance principles · 3 solutions Read more → Government, defence and technology Closed networks, explainable decisions, source code that stays inside. ## Government & Public Sector 5 agents From citizen requests to tender files: agents bound to regulation, explainable, leaving the decision with the officer. Request routingFile pre-review loadExplainability 4 compliance principles · 3 solutions Read more → ## Defence & Aerospace 4 agents Export control, maintenance (MRO) and supply-chain security: agents that run on a fully closed network. Export-compliance loadTechnical publication sprawlSupplier security 4 compliance principles · 3 solutions Read more → ## Software Delivery 8 agents PM, analyst, architect, developer and test agents: the whole line from feature request to production. Requirement ambiguityReview bottleneckRelease risk 3 compliance principles · 3 solutions Read more → ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Banking URL: https://www.aginies.com/en/industries/banking Language: en Summary: The core stays where it is; a deciding, acting AI layer goes on top. Banking # The core stays where it is; a deciding, acting AI layer goes on top. Almost every bank runs AI pilots; very few operate them in core processes, at scale and audit-ready. Aginies establishes one orchestration layer for cards, lending, the call centre and digital channels: every agent, service, limit and approval is defined by central, auditable rules. In short In Banking, Aginies offers 9 live agent scenarios and 3 solution packages: Dialogue banking: voice and text transaction assistant, Credit card limit increase, Digital account opening (KYC), Card fraud alert triage, Chargeback dispute handling, Early-stage collections outreach, Mortgage document review, Voice banking money transfer; solutions: Conversational banking, Credit and onboarding operations, Fraud and dispute operations. The blockers on the way to production are The PoC gap, Legacy core, Invisible cost, Compliance arrives late; on the compliance side 4 requirements (including Deployment aligned with banking IT regulation; primary and secondary systems in-country) are part of the deployment. Why it does not reach production 01 ### The PoC gap Chatbot pilots impress, but no production path touches a core banking transaction. 02 ### Legacy core Core systems were not designed for AI to decide and act; every integration is its own project. 03 ### Invisible cost Which scenario burns what tokens, which decision used which data: unreported. 04 ### Compliance arrives late Regulatory, privacy and internal-audit expectations surface at the end and stop the project. Where to start: the first 90 days Three starting points; each measurable in a short pilot on existing systems. - 01 agent scenarioDialogue banking: voice and text transaction assistant → The most visible impact: call-centre load and customer experience at once; a 6-week pilot on the existing identity stack. - 02 agent scenarioCard fraud alert triage → Relieves the analyst queue immediately; the decision stays with people, so security sign-off comes fast. - 03 solution packageCredit and onboarding operations → Document review and KYC are the largest measurable workload; the auto-approval band is written together with the credit policy. Scenarios ## Agents you can watch live for this sector BankingHITL ### Dialogue banking: voice and text transaction assistant The customer talks or types; bill, limit, dispute and transfer are completed in core banking in one session. 186,000 runs / month 94.7% success 5 agents BankingHITL ### Credit card limit increase The customer types in the app or talks to the call centre; the decision takes seconds, and anything above threshold goes to a manager. 18,400 runs / month 96.4% success 4 agents BankingHITL ### Digital account opening (KYC) ID, liveness and sanctions checks in one flow; clean applications open an account in minutes, no branch visit. 42,000 runs / month 93.8% success 5 agents BankingHITL ### Card fraud alert triage The card is blocked in seconds and the customer asked on WhatsApp; the reply releases or cancels the card. 65,000 runs / month 97.1% success 4 agents BankingHITL ### Chargeback dispute handling Dispute captured, evidence assembled and filed with the card scheme; large provisional credits go to approval. 9,800 runs / month 95.2% success 4 agents BankingHITL ### Early-stage collections outreach Past-due accounts are segmented and reminded on the right channel; customers set up their own payment plan. 96,000 runs / month 91.2% success 4 agents BankingHITL ### Mortgage document review Payslips, social security records and appraisal read automatically; income, LTV and missing documents checked. 6,200 runs / month 94.7% success 4 agents BankingHITL ### Voice banking money transfer Customer calls and talks; voiceprint, security question and OTP verify them; transfer completes agent-free. 27,000 runs / month 93.1% success 5 agents Sector architecture Derived from the 9 scenarios in the sector: supervisors, specialist agents, Hub model families and systems touched. Supervisors Dialogue Banking SupervisorCard SupervisorOnboarding SupervisorFraud SupervisorDispute SupervisorCollections SupervisorLoan File SupervisorVoice Transaction Supervisor Specialist agents Dialogue AgentIdentity AgentTransaction AgentRisk AgentIntent AgentOperations AgentDocument AgentLiveness AgentScreening AgentAccount AgentTriage AgentOutreach AgentCard AgentIntake AgentEvidence AgentFiling AgentSegmentation AgentPayment Plan AgentExtraction AgentIncome AgentValuation AgentSpeech AgentTransfer AgentPayment AgentVoice Agent Models (Hub catalogue) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-konrul-1aginies-voice-tr-v3 Systems touched Mobile App ChatCore BankingCard ManagementFraud EngineSMS GatewayKYC / Customer 360Push & SMS GatewaySanctions Screening (Dow Jones)Document Store (S3)WhatsApp Business APISalesforce CRMVisa / Mastercard Dispute APIEmail (Exchange)Collections PlatformLoan Origination SystemCredit Bureau (KKB)Genesys SIPKnowledge Base · kart_sss_v3OTP / SMS Gateway Channels appvoicewhatsappwebsmsapiemailbranchbatch Simulation volumes 502,400runs / month 94.6%success rate 6.5 smedian duration $0.0028estimated cost / run Summed from the simulation definitions of the sector scenarios; recalculated for your scale in the discovery session. Compliance and sovereignty - Deployment aligned with banking IT regulation; primary and secondary systems in-country - Kill-switch and mandatory human approval on critical decisions - Model, platform and data in the bank’s own data centre, with no external dependency - A full audit trail for every agent action and model call Regulatory frame - Banking IT and electronic banking regulation: primary and secondary systems in Türkiye, audit right in the outsourcing contract, cloud model subject to bank-specific permission - AML and personal-data law: suspicious-transaction reports decided by analysts, masking and purpose limitation in every scenario - Per-request audit trail for internal audit and regulator review: user, model version, data source, reasoning Security → Recommended deployment Start with the sovereign shared deployment in our data centres in Türkiye (isolated bank workspace, VPN or allowlist access); on-prem in the bank’s data centre for scenarios classed as primary systems. Approval rules Human-approval thresholds of the sector scenarios. - Dialogue banking: voice and text transaction assistantLimit increase > 60%, dispute amount > 5,000 or transfer to a new payee → operations approval - Credit card limit increaseLimit > 100,000 → branch manager approval - Digital account opening (KYC)Sanctions/PEP fuzzy match score > 0.65 → compliance analyst approval - Card fraud alert triageConfirmed fraud loss > 2,500 → analyst approval before permanent cancellation - Chargeback dispute handlingAmount > 2,500 → dispute team lead approval before provisional credit - Early-stage collections outreachPlan balance > 25,000 or > 3 instalments → collections team lead approval - Mortgage document reviewLoan > 3,000,000 or DTI > 45% → underwriting officer approval - Voice banking money transferAmount > 50,000 or new payee → warm transfer to a call-centre agent - Credit card debt paymentAmount > 100,000 → branch manager approval Measured targets | Simple transactions reaching the call centre | 40% fewer | | Fraud alert review time | 70% shorter | | Accounts opened without a branch visit | 70–85% | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Related solution packages Conversational bankingComplete mobile banking operations by typing or talking.Credit and onboarding operationsDigital account opening, limit increases and loan documents under one decision logic, without a branch visit.Fraud and dispute operationsAlert triage, chargeback files, card blocks and early collections on one operations desk. ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Defence & Aerospace URL: https://www.aginies.com/en/industries/defence Language: en Summary: Export control, maintenance (MRO) and supply-chain security: agents that run on a fully closed network. Defence & Aerospace # Export control, maintenance (MRO) and supply-chain security: agents that run on a fully closed network. Defence and aerospace companies need speed and security at once: export-licence detection, platform maintenance planning, supplier qualification. Aginies runs these processes on the organisation’s closed network with open-source models and no external dependency. Every controlled decision, every work package affecting airworthiness and every qualification stays with people; agents classify, match and prepare. In short In Defence & Aerospace, Aginies offers 4 live agent scenarios and 3 solution packages: Technical publication assistant: maintenance manual, parts, procedure and revision control, Export control and end-user screening, Platform maintenance (MRO) work order preparation, Supplier security and compliance assessment, Contract clause extraction, Code review and security gate; solutions: Information management & decision support, Field service & dispatch automation, Defence export control & supply-chain security. The blockers on the way to production are Export-compliance load, Technical publication sprawl, Supplier security, No cloud; on the compliance side 4 requirements (including Air-gapped deployment; models, platform and data inside the closed network) are part of the deployment. Why it does not reach production 01 ### Export-compliance load Classification and party screening are manual per item; experts spend hours per file. 02 ### Technical publication sprawl Task cards and parts lists span thousands of pages; finding the right revision takes time. 03 ### Supplier security Questionnaires and documents are read by hand; gaps surface after qualification. 04 ### No cloud No external model can be used in classified environments. Where to start: the first 90 days Three starting points; each measurable in a short pilot on existing systems. - 01 agent scenarioTechnical publication assistant: maintenance manual, parts, procedure and revision control → Knowledge hangar and maintenance teams use every day; starts at low risk on the closed network. - 02 agent scenarioExport control and end-user screening → Party screening and control-list mapping are rule-based; the licence decision stays with the compliance manager. - 03 agent scenarioSupplier security and compliance assessment → Supplier questionnaires are document-heavy; assessment time is measured immediately. Scenarios ## Agents you can watch live for this sector Defence & AerospaceHITL ### Technical publication assistant: maintenance manual, parts, procedure and revision control The technician asks in the hangar; the right-revision procedure, part number and safety note arrive with clearance checks, every access traced. 9,800 runs / month 95.7% success 5 agents Defence & AerospaceHITL ### Export control and end-user screening Part, buyer and end use are classified in one flow; licence need is detected, the decision always stays with the compliance officer. 2,600 runs / month 96.1% success 4 agents Defence & AerospaceHITL ### Platform maintenance (MRO) work order preparation The fault report is matched to technical publications, task card and parts list are produced, the work order opens with planner approval. 4,100 runs / month 95.7% success 5 agents Defence & AerospaceHITL ### Supplier security and compliance assessment Supplier questionnaires and documents are read, compared against facility security requirements, gaps listed; the qualification decision belongs to the security manager. 900 runs / month 95.2% success 4 agents Enterprise FunctionsHITL ### Contract clause extraction Clauses are extracted and compared with the legal playbook; deviations reach counsel with redline suggestions. 1,400 runs / month 95.1% success 4 agents Software DeliveryHITL ### Code review and security gate Agents review, scan and coverage-check every PR; fixes land first, a human approves only clean PRs. 4,200 runs / month 93.6% success 5 agents Sector architecture Derived from the 4 scenarios in the sector: supervisors, specialist agents, Hub model families and systems touched. Supervisors Technical Publication SupervisorExport Compliance SupervisorMaintenance SupervisorSupplier Security Supervisor Specialist agents Clearance AgentPublication AgentProcedure AgentParts AgentClassification AgentParty Screening AgentLicence AgentFault AgentTech Pub AgentPlanning AgentDocument AgentAssessment AgentCase Agent Models (Hub catalogue) aginies-sunkar-1aginies-tulpar-4aginies-talay-2aginies-konrul-1 Systems touched Hangar Tablet AppAccess ControlTech Pub Repository (S1000D)MRO SystemAudit LogERP (SAP GTS)Control & Sanctions Lists (vector)Trade Compliance Case SystemE-mailMaintenance Management SystemTechnical Publications (S1000D vector)Spares InventoryTechnician AppSupplier Management (SRM)Security Requirements (vector)Document Store Channels appapiemailbatchweb Simulation volumes 17,400runs / month 95.7%success rate 7.6 smedian duration $0.0053estimated cost / run Summed from the simulation definitions of the sector scenarios; recalculated for your scale in the discovery session. Compliance and sovereignty - Air-gapped deployment; models, platform and data inside the closed network - Licence decisions on controlled items only with the export compliance officer - Access separation by classification level, full audit trail, kill-switch - Control set aligned with facility-security and supply-chain requirements Regulatory frame - Defence-industry agency and export-control law: control-list mapping by the agent, licence and sharing decisions with the authorised compliance officer - Facility security clearance and classification levels: access separation, marking checks and an audit log of every access - ITAR/EAR-like restrictions in allied programmes: controlled technical data only to cleared users, no egress to external networks Security → Recommended deployment Air-gapped deployment: models, platform and data inside the closed network; updates as approved packages, remote maintenance only in windows the organisation opens. Approval rules Human-approval thresholds of the sector scenarios. - Technical publication assistant: maintenance manual, parts, procedure and revision controlProposed deviation from the procedure → authorised engineer approval; export-controlled content → clearance required - Export control and end-user screeningEvery controlled item → export compliance officer approval; the agent cannot make licence decisions - Platform maintenance (MRO) work order preparationEvery work package affecting airworthiness → maintenance planner approval - Supplier security and compliance assessmentThe qualification decision always sits with the facility security manager; the agent only prepares Measured targets | Party screening time | From days to minutes | | Technical publication query response | Seconds, at the right revision | | Data leaving the network | Zero | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Related solution packages Information management & decision supportSource-linked access to regulations, policies and contracts in seconds.Field service & dispatch automationSensor alarm or customer report → the right technician, the right part, the right time.Defence export control & supply-chain securityClassification, party screening, licence file and supplier qualification: agents on a closed network that leave the decision with the expert. ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Energy & Utilities URL: https://www.aginies.com/en/industries/energy Language: en Summary: Outages, meters, connection applications and demand response: agents connected to grid systems, talking to customers. Energy & Utilities # Outages, meters, connection applications and demand response: agents connected to grid systems, talking to customers. Distribution and retail energy companies run two worlds at once: the grid with SCADA, GIS and meter data; millions of customers with CRM and billing. Aginies builds agents that bridge the two: clustering outage reports, solving meter disputes from photos, pre-assessing rooftop solar applications, running demand-response events. Engineering and regulatory decisions stay with people. In short In Energy & Utilities, Aginies offers 5 live agent scenarios and 3 solution packages: Utility customer assistant: bill, outage, consumption and instalments in one dialogue, Outage report handling, Meter reading dispute, Rooftop solar connection application, Demand response event, Bill dispute resolution; solutions: Invoice & document extraction, Field service & dispatch automation, Utility customer operations. The blockers on the way to production are Outage call surge, Estimated-bill disputes, Distributed generation applications, Idle flexibility; on the compliance side 3 requirements (including Billing correction, connection and notification processes aligned with energy regulation) are part of the deployment. Why it does not reach production 01 ### Outage call surge Hundreds of calls for one feeder; the call centre jams before anyone is informed. 02 ### Estimated-bill disputes Unread meters mean estimates; wrong estimates mean disputes; every case by hand. 03 ### Distributed generation applications Rooftop solar requests multiply; engineers spend time on document checks. 04 ### Idle flexibility Demand-response contracts exist; event management is manual and slow. Where to start: the first 90 days Three starting points; each measurable in a short pilot on existing systems. - 01 agent scenarioUtility customer assistant: bill, outage, consumption and instalments in one dialogue → Bill and outage calls are the largest volume; a measurable pilot with CIS and OMS connected. - 02 agent scenarioOutage report handling → Call load multiplies in outages; proactive information shows in the first storm. - 03 agent scenarioRooftop solar connection application → Applications are document-heavy and deadline-bound; compliance with the regulator calendar is measured. Scenarios ## Agents you can watch live for this sector Energy & UtilitiesHITL ### Utility customer assistant: bill, outage, consumption and instalments in one dialogue The customer calls or writes; the high bill is explained, outage ETA given, meter data shown, an instalment plan set up. 520,000 runs / month 94.6% success 5 agents Energy & UtilitiesHITL ### Outage report handling Reports are clustered by feeder, one OSS ticket is raised and everyone affected hears before the crew leaves. 38,000 runs / month 97.7% success 5 agents Energy & UtilitiesHITL ### Meter reading dispute Customer sends a meter photo; the index is read by vision, the estimated bill fixed, big credits approved. 9,200 runs / month 94.9% success 4 agents Energy & UtilitiesHITL ### Rooftop solar connection application Application documents are read, transformer capacity and regulation checked, gaps reported the same day; the connection opinion belongs to the engineer. 5,400 runs / month 95.3% success 5 agents Energy & UtilitiesHITL ### Demand response event Under grid stress eligible consumers are selected, the offer sent by channel, participation measured and settlement computed. 140 runs / month 96.2% success 4 agents TelecomHITL ### Bill dispute resolution Invoice lines and usage explained in seconds; goodwill credit applied by policy, larger amounts need a lead. 88,000 runs / month 96.3% success 4 agents Sector architecture Derived from the 5 scenarios in the sector: supervisors, specialist agents, Hub model families and systems touched. Supervisors Customer Service SupervisorOutage SupervisorBilling Dispute SupervisorConnection SupervisorDemand Response Supervisor Specialist agents Dialogue AgentBilling AgentGrid AgentCollections AgentVoice AgentReport AgentClustering AgentComms AgentVision AgentOperations AgentDocument AgentRegulation AgentCase AgentSelection AgentOffer AgentSettlement Agent Models (Hub catalogue) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-voice-tr-v3aginies-konrul-1 Systems touched IVR / SIPBilling (CIS)Outage ManagementSmart Meter DataSMS GatewayCustomer CRM / BillingSCADA / HistorianNetwork OSSWhatsApp Business APIBilling / CIS (SAP IS-U)Meter Data ManagementRegulation Vector Store (EPDK)Connection PortalGIS / Network ModelE-mailSCADA / DMSBilling / Settlement Channels voicewhatsappappwebemailapibatch Simulation volumes 572,740runs / month 94.8%success rate 6.4 smedian duration $0.0016estimated cost / run Summed from the simulation definitions of the sector scenarios; recalculated for your scale in the discovery session. Compliance and sovereignty - Billing correction, connection and notification processes aligned with energy regulation - Every decision affecting grid operation under control-room or engineer approval - Read-only access to SCADA and OT; critical-infrastructure separation preserved Regulatory frame - Energy regulator consumer-service and connection regulations: dispute, correction and application deadlines coded into agent rules - Critical infrastructure and OT separation: read-only access to SCADA and OMS; grid decisions in the control room - Personal-data law and meter data: consumption data only for the customer’s own account, retention by policy Security → Recommended deployment Sovereign shared deployment for customer operations; grid-adjacent read-only agents on-prem in the company data centre with critical-infrastructure separation kept. Approval rules Human-approval thresholds of the sector scenarios. - Utility customer assistant: bill, outage, consumption and instalments in one dialogueInstalments > 6 months, bill cancellation or meter suspicion → customer service approval - Outage report handlingMass notification > 1,000 customers → control room dispatcher approval - Meter reading disputeCredit / correction > 1,000 TL → billing team lead approval - Rooftop solar connection applicationThe connection opinion is always signed by a network planning engineer - Demand response eventEvent > 10 MW → grid operator approval; settlement with an audit trail Measured targets | Calls resolved without a representative | 55–65% | | Bill dispute resolution time | From days to minutes | | Call load during outages | 50% lower | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Related solution packages Invoice & document extractionOCR → validation → ERP: clean records without manual entry.Field service & dispatch automationSensor alarm or customer report → the right technician, the right part, the right time.Utility customer operationsOutages, meters, connections and demand response: customer agents that speak with grid data. ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Fintech & Payments URL: https://www.aginies.com/en/industries/fintech Language: en Summary: Onboarding, AML, credit decisions and reconciliation: agents where speed and compliance are needed at once. Fintech & Payments # Onboarding, AML, credit decisions and reconciliation: agents where speed and compliance are needed at once. Payment institutions, e-money companies and BNPL providers face bank-grade supervision at start-up growth speed. Aginies runs merchant onboarding, transaction-monitoring alerts, instant credit decisions and PSP reconciliation exceptions on one orchestration layer; regulatory filings and above-limit decisions stay with people. In short In Fintech & Payments, Aginies offers 5 live agent scenarios and 3 solution packages: Merchant operations assistant: POS, settlement, refund and limit, Merchant KYB onboarding, AML transaction monitoring alert triage, BNPL instant instalment decision, Payment reconciliation exceptions, Chargeback dispute handling, Card fraud alert triage; solutions: Conversational banking, Financial analysis & operational intelligence, Merchant and payment operations. The blockers on the way to production are Onboarding bottleneck, Alert fatigue, Second-level credit decisions, Reconciliation exceptions; on the compliance side 4 requirements (including AML/CFT and privacy requirements; suspicious-activity filings always by analyst decision) are part of the deployment. Why it does not reach production 01 ### Onboarding bottleneck KYB, UBO and sanctions checks take days; the merchant goes to a competitor. 02 ### Alert fatigue 95% of transaction-monitoring hits are false positives; analysts miss real risk. 03 ### Second-level credit decisions The decision is needed at checkout; bureau and open-banking data must merge in one flow. 04 ### Reconciliation exceptions Differences between PSP statements and the ledger are closed by hand at month end. Where to start: the first 90 days Three starting points; each measurable in a short pilot on existing systems. - 01 agent scenarioMerchant operations assistant: POS, settlement, refund and limit → The merchant desk is the busiest channel; a WhatsApp pilot is measurable within weeks. - 02 agent scenarioPayment reconciliation exceptions → Exceptions waiting for month-end lock the finance team; automatic matching shows a difference in the first month. - 03 agent scenarioMerchant KYB onboarding → KYB time is directly tied to revenue; the policy engine is set up with the risk team. Scenarios ## Agents you can watch live for this sector Fintech & PaymentsHITL ### Merchant operations assistant: POS, settlement, refund and limit The merchant writes on WhatsApp; POS fault, settlement query, refund and limit increase are resolved in one dialogue. 64,000 runs / month 93.8% success 5 agents Fintech & PaymentsHITL ### Merchant KYB onboarding Registry, UBO and sanctions checks in one flow; low-risk merchants start accepting payments the same day. 1,400 runs / month 93.4% success 5 agents Fintech & PaymentsHITL ### AML transaction monitoring alert triage Alerts arrive enriched with a draft narrative; the analyst always decides and the filing goes out in minutes. 8,500 runs / month 95.8% success 5 agents Fintech & PaymentsHITL ### BNPL instant instalment decision Checkout instalment decision in seconds from bureau and open-banking data; large tickets go to underwriting. 54,000 runs / month 97.3% success 5 agents Fintech & PaymentsHITL ### Payment reconciliation exceptions PSP settlement files matched to the ledger; small breaks fix themselves, large ones go to finance approval. 620 runs / month 95.1% success 4 agents BankingHITL ### Chargeback dispute handling Dispute captured, evidence assembled and filed with the card scheme; large provisional credits go to approval. 9,800 runs / month 95.2% success 4 agents BankingHITL ### Card fraud alert triage The card is blocked in seconds and the customer asked on WhatsApp; the reply releases or cancels the card. 65,000 runs / month 97.1% success 4 agents Sector architecture Derived from the 5 scenarios in the sector: supervisors, specialist agents, Hub model families and systems touched. Supervisors Merchant SupervisorMerchant Onboarding SupervisorAML SupervisorCredit Decision SupervisorReconciliation Supervisor Specialist agents Dialogue AgentPOS AgentOperations AgentRisk AgentBusiness Identity AgentUBO AgentContract AgentEnrichment AgentScreening AgentNarrative AgentFiling AgentIntake AgentBureau AgentDecision AgentBooking AgentMatching AgentInvestigation AgentCorrection Agent Models (Hub catalogue) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-konrul-1 Systems touched WhatsApp BusinessTerminal ManagementSettlement LedgerRefund ServiceTicketingMerchant PortalTrade Registry (MERSİS)Sanctions Screeninge-SignEmailCore LedgerCase ManagementMASAK Reporting PortalMerchant Checkout APICredit Bureau (KKB / Findeks)Open Banking APISMS GatewayPSP / Acquirer APICard Network Channels whatsappwebvoiceemailapibatchappsms Simulation volumes 128,520runs / month 95.4%success rate 7.8 smedian duration $0.0031estimated cost / run Summed from the simulation definitions of the sector scenarios; recalculated for your scale in the discovery session. Compliance and sovereignty - AML/CFT and privacy requirements; suspicious-activity filings always by analyst decision - Human approval for high-risk merchants and above-limit credit decisions - Data source, model version and rationale recorded for every decision - Data in-country; on-prem or dedicated region Regulatory frame - Central-bank payment and e-money IT communiqué: community cloud only with an approved provider; primary systems in-country - AML obligations: transaction-monitoring alerts to the analyst with evidence, the reporting decision with people - Card-scheme rules and personal-data law: chargeback deadlines tracked, card data masked and never used beyond purpose Security → Recommended deployment In the community cloud of a provider on the central bank’s approved list, or in the institution’s own data centre; GPU capacity in-country, models from the Hub catalogue. Approval rules Human-approval thresholds of the sector scenarios. - Merchant operations assistant: POS, settlement, refund and limitRefund > 25,000 or limit increase > 50% → risk team approval - Merchant KYB onboardingRisk tier high or UBO fuzzy match > 0.65 → compliance manager approval - AML transaction monitoring alert triageEvery alert → compliance analyst decision mandatory; the agent never files an STR or closes a case alone - BNPL instant instalment decisionBasket > 15,000 or bureau score < 1,200 → underwriter approval - Payment reconciliation exceptionsSingle break > 10,000 or total > 50,000 → finance manager approval before adjustment Measured targets | Merchant request first-contact resolution | 70% and above | | KYB completion time | From days to hours | | Reconciliation exception closure | 60% shorter | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Related solution packages Conversational bankingComplete mobile banking operations by typing or talking.Financial analysis & operational intelligenceAgents that read financial data and documents, interpret them and support decisions.Merchant and payment operationsMerchant onboarding, POS and settlement support, reconciliation exceptions and transaction monitoring on one desk. ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Government & Public Sector URL: https://www.aginies.com/en/industries/government Language: en Summary: From citizen requests to tender files: agents bound to regulation, explainable, leaving the decision with the officer. Government & Public Sector # From citizen requests to tender files: agents bound to regulation, explainable, leaving the decision with the officer. In public institutions every transaction is tied to a regulation clause, a deadline and a responsible officer. Aginies builds agents that route citizen requests, pre-assess permit and social-assistance files, and compare tender bids with specifications. Every output is justified with a clause number; the decision always belongs to the authorised officer or commission. Deployment can run in the institution’s own data centre with no external dependency. In short In Government & Public Sector, Aginies offers 5 live agent scenarios and 3 solution packages: Citizen services assistant: application, documents, appointment and status, Citizen request routing, Permit application check, Public procurement bid file pre-check, Social assistance eligibility pre-assessment, Export control and end-user screening; solutions: KVKK / GDPR personal data protection, Citizen services & file pre-review, Public procurement & tender compliance. The blockers on the way to production are Request routing, File pre-review load, Explainability, Sovereignty requirement; on the compliance side 4 requirements (including Deployment in the institution’s own data centre, air-gapped if required; open-source models served inside) are part of the deployment. Why it does not reach production 01 ### Request routing Petitions reach the wrong unit, the clock runs, the citizen applies again. 02 ### File pre-review load Missing-document checks in permit, assistance and tender files take most of officers’ time. 03 ### Explainability Every decision requires a rationale; black-box tools cannot be used. 04 ### Sovereignty requirement Data cannot leave the country; most cloud assistants are ruled out for that reason. Where to start: the first 90 days Three starting points; each measurable in a short pilot on existing systems. - 01 agent scenarioCitizen services assistant: application, documents, appointment and status → Applications and documents are what citizens feel most; e-Government integration removes repeated documents. - 02 agent scenarioCitizen request routing → Petition routing is rule-based and high-volume; the lowest-risk first pilot. - 03 solution packagePublic procurement & tender compliance → Tender-file pre-review frees commission time; the decision stays with the commission. Scenarios ## Agents you can watch live for this sector Government & Public SectorHITL ### Citizen services assistant: application, documents, appointment and status The citizen writes or calls; eligibility checked, documents read, application opened, appointment given, every step traceable. 230,000 runs / month 93.8% success 5 agents Government & Public SectorHITL ### Citizen request routing e-Government requests classified in seconds, given a legal deadline and routed to the right unit with an SLA. 54,000 runs / month 98.1% success 4 agents Government & Public SectorHITL ### Permit application check Missing documents and regulatory fit reported in minutes; the licensing officer always makes the decision. 3,100 runs / month 95.8% success 4 agents Government & Public SectorHITL ### Public procurement bid file pre-check Bid files are compared with the specification, missing and non-conforming documents listed; the evaluation decision belongs to the commission. 380 runs / month 94.8% success 4 agents Government & Public SectorHITL ### Social assistance eligibility pre-assessment The application is verified against agency records, eligibility scored, gaps reported to the citizen; the decision belongs to the case officer. 21,000 runs / month 96.6% success 4 agents Defence & AerospaceHITL ### Export control and end-user screening Part, buyer and end use are classified in one flow; licence need is detected, the decision always stays with the compliance officer. 2,600 runs / month 96.1% success 4 agents Sector architecture Derived from the 5 scenarios in the sector: supervisors, specialist agents, Hub model families and systems touched. Supervisors Citizen Services SupervisorRequest SupervisorPermit SupervisorTender SupervisorSocial Assistance Supervisor Specialist agents Dialogue AgentEligibility AgentDocument AgentCase AgentClassification AgentSLA AgentRegulation AgentFile AgentSpecification AgentMinutes AgentVerification AgentCitizen Comms Agent Models (Hub catalogue) aginies-talay-2aginies-tulpar-4aginies-konrul-1aginies-sunkar-1 Systems touched WhatsApp Businesse-Government GatewayCase ManagementDocument StoreAppointment Systeme-Devlet GatewayLegislation Vector StoreSMS GatewayGIS / Zoning PlanRegulation Vector StoreExchange / M365 Maile-Procurement PlatformSpecification Vector StoreSocial Assistance Information SystemCivil Registry / Income ServicesEligibility Criteria (vector) Channels whatsappvoicewebappemailbatch Simulation volumes 308,480runs / month 94.8%success rate 7 smedian duration $0.0021estimated cost / run Summed from the simulation definitions of the sector scenarios; recalculated for your scale in the discovery session. Compliance and sovereignty - Deployment in the institution’s own data centre, air-gapped if required; open-source models served inside - Every output justified with a regulation clause and source document - The administrative decision is always with the authorised officer or commission; the agent only prepares - Access control and audit trail aligned with privacy and public-sector security requirements Regulatory frame - Presidential information and communication security guide: deployment in the institution’s data centre, air-gap included; open-weight models inside - Administrative procedure and public procurement law: every output with the regulation article and source document; the decision with the authorised officer or commission - Personal-data law and e-Government integration rules: citizen data purpose-limited, access by authorisation, full audit trail Security → Recommended deployment On-prem in the institution’s own data centre; air-gapped where required. For municipalities and affiliated bodies, a dedicated tenant in the public cloud. Approval rules Human-approval thresholds of the sector scenarios. - Citizen services assistant: application, documents, appointment and statusBorderline eligibility, inconsistent document or another application in the household → officer review - Citizen request routingClassification confidence < 0.85 or staff complaint → registry officer approval - Permit application checkEvery permit decision is taken by the licensing officer; the system only pre-reviews - Public procurement bid file pre-checkConformity and elimination decisions always with the tender commission; the agent only prepares - Social assistance eligibility pre-assessmentThe assistance decision is always with the case officer; the agent pre-assesses Measured targets | Documents asked of the citizen again | 70% fewer | | Application pre-review time | From days to minutes | | Correct first routing | 95% and above | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Related solution packages KVKK / GDPR personal data protectionDetect personal data in documents and calls, mask it, and keep an audit-ready data map.Citizen services & file pre-reviewThe request to the right unit, the file complete, the decision with the officer: justified by clause.Public procurement & tender complianceBid files compared with the specification; the commission decides on a prepared findings list. ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Healthcare URL: https://www.aginies.com/en/industries/healthcare Language: en Summary: The clinical decision stays with the physician; agents carry prior authorisation, intake and scheduling. Healthcare # The clinical decision stays with the physician; agents carry prior authorisation, intake and scheduling. In hospitals and health groups most time goes to documents, forms and phone calls: payer prior authorisation, patient intake, appointment changes. Aginies takes that operational load with agents connected to the HIS, payer portals and communication channels; every clinical decision stays with physicians and nurses, and patient data never leaves the institution. In short In Healthcare, Aginies offers 4 live agent scenarios and 3 solution packages: Patient journey coordinator: appointment, preparation, approval and results, Prior authorization, Patient intake and triage, Appointment scheduling and no-show reduction, Call centre quality monitoring; solutions: Voice AI for claims intake, Healthcare operations & clinical guidance, KVKK / GDPR personal data protection. The blockers on the way to production are Prior-authorisation delay, Intake and triage load, No-shows, Data sensitivity; on the compliance side 3 requirements (including Special-category health data stays inside the institution; models on-prem, no external dependency) are part of the deployment. Why it does not reach production 01 ### Prior-authorisation delay Payer rules are scattered; collecting documents and writing rationale takes hours per case. 02 ### Intake and triage load Symptoms are taken by phone, transcribed into forms, then asked again. 03 ### No-shows No-show rates are high; freed slots are not refilled. 04 ### Data sensitivity Health data is special-category; cloud assistants do not pass the compliance team. Where to start: the first 90 days Three starting points; each measurable in a short pilot on existing systems. - 01 agent scenarioPatient journey coordinator: appointment, preparation, approval and results → Scheduling and preparation carry no clinical decision; a fast start with no-show rate measured immediately. - 02 agent scenarioPrior authorization → Insurer pre-approval is document-heavy and rule-based; it shortens reimbursement time. - 03 agent scenarioPatient intake and triage → Intake and pre-assessment run with nurse approval; it organises the emergency load. Scenarios ## Agents you can watch live for this sector HealthcareHITL ### Patient journey coordinator: appointment, preparation, approval and results The patient writes or calls; appointment, preparation instructions, insurer pre-approval and result notice come together in one coordinator. 58,000 runs / month 95.1% success 5 agents HealthcareHITL ### Prior authorization Payer rules searched, clinical evidence gathered and the packet ready to submit within minutes of the order. 6,200 runs / month 94.1% success 4 agents HealthcareHITL ### Patient intake and triage The patient types or talks; intake and urgency scoring take seconds, and a nurse always makes the final call. 24,500 runs / month 95.8% success 5 agents Healthcare ### Appointment scheduling and no-show reduction High no-show-risk patients are reached on WhatsApp in advance; freed slots are refilled from the waiting list. 62,000 runs / month 98.2% success 4 agents Enterprise FunctionsHITL ### Call centre quality monitoring Every call scored against the rubric, not a sample; compliance breaches and coaching flags ready by morning. 90,000 runs / month 98.8% success 4 agents Sector architecture Derived from the 4 scenarios in the sector: supervisors, specialist agents, Hub model families and systems touched. Supervisors Patient Journey SupervisorPrior Auth SupervisorTriage SupervisorAppointment Supervisor Specialist agents Dialogue AgentScheduling AgentClinical Preparation AgentInsurer AgentOrder AgentPayer Rules AgentSubmission AgentVoice AgentSymptom AgentUrgency AgentRouting AgentNo-show Risk AgentBooking Agent Models (Hub catalogue) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-konrul-1aginies-voice-tr-v3 Systems touched WhatsApp BusinessHospital Information SystemSchedulingInsurer PortalLab / ImagingEHR (HL7/FHIR)Payer Policy Vector StorePayer PortalExchange / M365 MailHIS SchedulingTriage Protocol Vector StoreWhatsApp Business APIHospital Information System (HIS)SMS Gateway Channels whatsappvoiceappapiemailwebsmsbatch Simulation volumes 150,700runs / month 96.4%success rate 7.3 smedian duration $0.0023estimated cost / run Summed from the simulation definitions of the sector scenarios; recalculated for your scale in the discovery session. Compliance and sovereignty - Special-category health data stays inside the institution; models on-prem, no external dependency - Every step of clinical nature under physician or nurse approval - Masking, access logging and audit trail aligned with privacy and health regulation Regulatory frame - Personal health-data regulation and special-category data: health data inside the hospital perimeter, access by role and purpose, every access logged - Ministry hospital-information and national health-record integration rules: the agent acts through the HIS and keeps no separate data copy - Clinical decisions with the physician: the agent prepares, queries and organises; physician approval card when in doubt Security → Recommended deployment On-prem in the hospital data centre; models inside, no external connection. For hospital groups, a dedicated tenant in the group data centre. Approval rules Human-approval thresholds of the sector scenarios. - Patient journey coordinator: appointment, preparation, approval and resultsDrug interaction, suspected contraindication or extra test need → physician approval - Prior authorizationSurgical procedure or cost > 25,000 TL → utilisation management nurse approval - Patient intake and triageA triage nurse confirms the urgency level on every case; "emergency" cases are directed to 112 immediately Measured targets | No-show rate | 30–40% lower | | Pre-approval completion time | From days to hours | | Call-centre average handling time | 35% shorter | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Related solution packages Voice AI for claims intakeA voice agent takes the loss notice and opens the claim in the system.Healthcare operations & clinical guidanceSmart triage and patient referral; clinical decisions always stay with clinicians.KVKK / GDPR personal data protectionDetect personal data in documents and calls, mask it, and keep an audit-ready data map. ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Insurance URL: https://www.aginies.com/en/industries/insurance Language: en Summary: From first notice of loss to recovery: processes with tracked steps and defined approval points. Insurance # From first notice of loss to recovery: processes with tracked steps and defined approval points. Insurance volumes swing, documents are scattered and every decision must be explainable to the regulator. Aginies runs claims intake, renewals, underwriting document processing and subrogation on one orchestration layer; adjuster assignment and payment approval stay with people. In short In Insurance, Aginies offers 6 live agent scenarios and 3 solution packages: End-to-end motor claim: from notice to payment, Voice claims intake (FNOL), Policy renewal offer, Underwriting document extraction, Motor claim photo assessment, Subrogation recovery; solutions: Voice AI for claims intake, Claims operations end to end, Underwriting and renewal operations. The blockers on the way to production are Storm-day capacity, Document chaos, Unexplainable decisions, Missed recovery; on the compliance side 4 requirements (including Privacy-compliant processing with sensitive data masking) are part of the deployment. Why it does not reach production 01 ### Storm-day capacity Loss notices pile up at the call centre; waiting erodes satisfaction. 02 ### Document chaos Underwriting and claim files arrive in every format; missing documents stall the process. 03 ### Unexplainable decisions The model produced an output with no rationale; indefensible to regulator and customer. 04 ### Missed recovery Third-party liability stays buried in files; recovery opportunities are lost. Where to start: the first 90 days Three starting points; each measurable in a short pilot on existing systems. - 01 agent scenarioEnd-to-end motor claim: from notice to payment → Motor claims are the highest-volume, most visible process; photo assessment frees adjuster capacity immediately. - 02 agent scenarioVoice claims intake (FNOL) → Call-centre notices taken without a representative; a short pilot on the existing SIP stack. - 03 solution packageUnderwriting and renewal operations → Renewal season comes with the calendar; automatic offers move the renewal rate measurably. Scenarios ## Agents you can watch live for this sector InsuranceHITL ### End-to-end motor claim: from notice to payment Voice notice, photo assessment, repair-shop assignment and payment order in one flow; the adjuster steps in only above the threshold. 21,000 runs / month 91.2% success 5 agents InsuranceHITL ### Voice claims intake (FNOL) The customer calls and describes it; policy found by plate, cover checked, claim opened and adjuster assigned. 34,000 runs / month 95.8% success 5 agents InsuranceHITL ### Policy renewal offer Policies re-priced, sharp increases go to approval; customer accepts with one tap and the policy is issued. 22,000 runs / month 94.4% success 4 agents InsuranceHITL ### Underwriting document extraction Medical reports and labs read and assessed against the manual; underwriters only see files above threshold. 4,800 runs / month 94.9% success 4 agents InsuranceHITL ### Motor claim photo assessment Damage detected from photos, estimate built from parts and labour, adjuster only approves; repair slot booked. 15,500 runs / month 93.3% success 4 agents InsuranceHITL ### Subrogation recovery Closed claims scanned, the at-fault party's insurer identified, and the demand sent after counsel approval. 3,400 runs / month 95.5% success 4 agents Sector architecture Derived from the 6 scenarios in the sector: supervisors, specialist agents, Hub model families and systems touched. Supervisors Claims SupervisorRenewal SupervisorUnderwriting SupervisorSubrogation Supervisor Specialist agents Intake AgentPhoto Assessment AgentAdjuster AgentCommunication AgentSpeech AgentPolicy AgentClaims Registration AgentPricing AgentOffer AgentIssuance AgentDocument AgentMedical Assessment AgentUnderwriting AgentVision AgentEstimate AgentRepair Routing AgentLiability Scan AgentEvidence AgentRecovery Agent Models (Hub catalogue) aginies-voice-tr-v3aginies-konrul-1aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Systems touched Call Centre SIPPolicy AdminClaims SystemRepair NetworkPayment HubGenesys SIPPolicy Admin SystemGuidewire ClaimCenterWhatsApp Business APIRating EngineEmail (Exchange)Underwriting WorkbenchDocument Store (S3)Vector StoreParts & Labour Catalogue (Audatex)Repair Network PortalInsurance Registry (SBM) Channels voiceappwhatsappbatchemailwebapi Simulation volumes 100,700runs / month 94.1%success rate 7.8 smedian duration $0.0042estimated cost / run Summed from the simulation definitions of the sector scenarios; recalculated for your scale in the discovery session. Compliance and sovereignty - Privacy-compliant processing with sensitive data masking - Mandatory human approval on payment and rejection decisions - A reasoned, timestamped trail for every claim decision - On-premise or in-country dedicated region options Regulatory frame - Insurance regulator rules and insurance IT requirements: claim and policy decisions reasoned, time-stamped, open to audit - Special-category personal data: health and damage photos processed inside the perimeter, masking and access logging mandatory - Payment and rejection decisions always with the authorised adjuster or underwriter; the agent prepares the file Security → Recommended deployment Dedicated region in Türkiye or the sovereign shared deployment; vision models for damage photos and health data on in-country GPUs. Approval rules Human-approval thresholds of the sector scenarios. - End-to-end motor claim: from notice to paymentEstimate > 150,000, unclear fault share or overlap with prior damage → adjuster approval - Voice claims intake (FNOL)Injury or reserve > 250,000 → warm transfer to a claims specialist - Policy renewal offerPremium increase > 20% or discount > 10% → renewal team lead approval - Underwriting document extractionLoading ≥ 50% or sum assured > 5,000,000 → senior underwriter approval - Motor claim photo assessmentEstimate > 15,000 or model confidence < 80% → adjuster approval - Subrogation recoveryRecovery demand > 50,000 → subrogation counsel approval Measured targets | Simple claim closure time | From weeks to days | | Files needing an adjuster | 40% fewer | | Policies renewed automatically | 60–75% | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Related solution packages Voice AI for claims intakeA voice agent takes the loss notice and opens the claim in the system.Claims operations end to endFrom notice to payment: voice intake, photo assessment, repair-shop assignment and subrogation in one flow.Underwriting and renewal operationsQuote documents read, risk variables extracted, renewal offers prepared within the rules. ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # IoT & Connected Devices URL: https://www.aginies.com/en/industries/iot Language: en Summary: Agents that reduce signals from thousands of devices to one incident and one decision: anomalies, OTA, cold chain. IoT & Connected Devices # Agents that reduce signals from thousands of devices to one incident and one decision: anomalies, OTA, cold chain. Connected device fleets produce large volumes of data; the burden of deciding still sits with the teams. Anomaly alarms reach teams unclustered, firmware rollouts are postponed out of fear, and nobody knows who does what in a sensor excursion. Aginies turns signal into incident and incident into decision with agents connected to the IoT platform, time-series database, device management and field service; bulk device commands and critical segments always pass human approval. In short In IoT & Connected Devices, Aginies offers 4 live agent scenarios and 3 solution packages: Field device support assistant: telemetry, remote diagnosis, OTA and technician, Device fleet anomaly triage, Staged OTA update rollout, Cold-chain excursion response, Maintenance ticket triage, Outage report handling; solutions: Operational automation & demand management, Field service & dispatch automation, IoT device fleet operations. The blockers on the way to production are Alarm noise, Risky OTA rollouts, Sensor-to-action gap, Data volume; on the compliance side 3 requirements (including Human approval on bulk device commands and critical fleet segments) are part of the deployment. Why it does not reach production 01 ### Alarm noise Hundreds of alarms from one root cause; the team misses the real incident. 02 ### Risky OTA rollouts Whole-fleet updates in one go; no rollback plan. 03 ### Sensor-to-action gap The breach is seen, but service order, customer notice and quality decision are done separately by hand. 04 ### Data volume Telemetry volume explodes cloud model cost; small on-prem models are needed. Where to start: the first 90 days Three starting points; each measurable in a short pilot on existing systems. - 01 agent scenarioField device support assistant: telemetry, remote diagnosis, OTA and technician → Faults fixed remotely cut field cost directly; a short pilot with the device cloud connected. - 02 agent scenarioDevice fleet anomaly triage → Alarm noise wears teams down; clustering is visible in the first week. - 03 agent scenarioStaged OTA update rollout → Staged OTA is the most feared process; it earns trust with stop and rollback criteria. Scenarios ## Agents you can watch live for this sector IoT & Connected DevicesHITL ### Field device support assistant: telemetry, remote diagnosis, OTA and technician The customer says "the device is dead"; telemetry read, remote diagnosis made, update pushed, technician dispatched with the right part if needed. 37,000 runs / month 94.1% success 5 agents IoT & Connected DevicesHITL ### Device fleet anomaly triage Telemetry anomalies from thousands of devices are clustered, root-cause candidates listed, one incident reaches the right team. 18,000 runs / month 96.9% success 4 agents IoT & Connected DevicesHITL ### Staged OTA update rollout Firmware rolls out in 1% → 10% → 100% stages; a health gate at every stage, human approval for critical fleets. 320 runs / month 98.1% success 4 agents IoT & Connected DevicesHITL ### Cold-chain excursion response A temperature excursion is assessed instantly, product risk computed, a field crew dispatched, the quality decision recorded. 2,700 runs / month 97.4% success 4 agents Manufacturing & AutomotiveHITL ### Maintenance ticket triage Sensor alarm and technician voice note become one work order, spare parts and a safe stop window included. 3,400 runs / month 95.2% success 5 agents Energy & UtilitiesHITL ### Outage report handling Reports are clustered by feeder, one OSS ticket is raised and everyone affected hears before the crew leaves. 38,000 runs / month 97.7% success 5 agents Sector architecture Derived from the 4 scenarios in the sector: supervisors, specialist agents, Hub model families and systems touched. Supervisors Device Support SupervisorFleet Health SupervisorRollout SupervisorCold-Chain Supervisor Specialist agents Dialogue AgentTelemetry AgentDiagnosis AgentDispatch AgentSignal AgentOperations AgentPlan AgentHealth AgentDeploy AgentSensor AgentQuality AgentField Agent Models (Hub catalogue) aginies-talay-2aginies-sunkar-1aginies-tulpar-4 Systems touched WhatsApp BusinessDevice Cloud (MQTT)Fleet RegistryOTA ServiceField ServiceIoT Platform (MQTT / Digital Twin)Time-series DBKnown Issues (vector)ITSM / IncidentSlackCI/CD (signed artefacts)Device Management / OTAObservability (metrics)IoT PlatformWMS / Lot TrackingStability Rules (vector)WhatsApp Business API Channels whatsappwebapibatchapp Simulation volumes 58,020runs / month 95.1%success rate 6.9 smedian duration $0.0025estimated cost / run Summed from the simulation definitions of the sector scenarios; recalculated for your scale in the discovery session. Compliance and sovereignty - Human approval on bulk device commands and critical fleet segments - Automatic rollback when a health gate fails; every stage recorded - Telemetry stays inside; low cost with small on-prem models Regulatory frame - Personal-data law and device data: telemetry inside the organisation, user data purpose-limited - Product safety and cyber-resilience requirements (incl. EU CRA for exporters): OTA packages signed, rollout recorded, rollback ready - Safety-critical devices: read-only; commands and bulk operations under product-owner approval Security → Recommended deployment Hybrid: device cloud and time-series data on the existing infrastructure, agents and small models in our data centre in Türkiye or the manufacturer’s own. Approval rules Human-approval thresholds of the sector scenarios. - Field device support assistant: telemetry, remote diagnosis, OTA and technicianBulk OTA (>100 devices) or out-of-warranty device replacement → product-owner approval - Device fleet anomaly triageBulk device command > 100 devices → fleet operations lead approval - Staged OTA update rolloutRollout to a critical fleet segment → product owner approval; automatic rollback if a health gate fails - Cold-chain excursion responseProduct transfer or disposal decision → quality manager approval Measured targets | Faults needing a field visit | 35% fewer | | Alarm-to-incident noise | 80% less | | OTA rollback rate | under 1% | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Related solution packages Operational automation & demand managementAgents that classify, route and answer demands; faster processes, lower cost.Field service & dispatch automationSensor alarm or customer report → the right technician, the right part, the right time.IoT device fleet operationsAnomaly clustering, root cause and health-gated OTA rollouts: one incident, one decision per fleet. ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Logistics URL: https://www.aginies.com/en/industries/logistics Language: en Summary: Shipment exceptions, customs documents and cold chain: agents that decide the moment an event happens. Logistics # Shipment exceptions, customs documents and cold chain: agents that decide the moment an event happens. In logistics value hides in how fast an exception is spotted and solved: a late truck, a missing customs document, a temperature excursion. Aginies detects the exception with agents connected to TMS, WMS, customs and IoT systems, computes options, informs the customer; costly decisions stay with the operations manager. In short In Logistics, Aginies offers 3 live agent scenarios and 3 solution packages: Control tower: delay prediction, rerouting, customer notice, SLA, Shipment exception handling, Customs document preparation, Cold-chain excursion response, Where is my order? (WISMO); solutions: Operational automation & demand management, Field service & dispatch automation, Logistics control tower. The blockers on the way to production are Exceptions surface late, Customs document errors, Cold-chain visibility, Customer communication; on the compliance side 3 requirements (including Human approval on customs declarations and high-cost re-routing) are part of the deployment. Why it does not reach production 01 ### Exceptions surface late The delay is learned when the customer calls; there is no alternative plan. 02 ### Customs document errors Documents are prepared by hand; one tariff-code error holds cargo for days. 03 ### Cold-chain visibility Sensors exist, the decision process does not; who does what in an excursion is unclear. 04 ### Customer communication The same information is asked through three channels; no proactive updates. Where to start: the first 90 days Three starting points; each measurable in a short pilot on existing systems. - 01 agent scenarioControl tower: delay prediction, rerouting, customer notice, SLA → Delay management sits at the centre of customer satisfaction and SLA cost; a fast pilot where telematics exists. - 02 agent scenarioShipment exception handling → The exception desk has high daily volume; the TMS integration alone is enough. - 03 agent scenarioCustoms document preparation → Document preparation is rule-based and time-critical; the declaration decision stays with the customs broker. Scenarios ## Agents you can watch live for this sector LogisticsHITL ### Control tower: delay prediction, rerouting, customer notice, SLA The telematics signal catches the delay early; alternative route and carrier chosen, customer informed, SLA compensation computed. 96,000 runs / month 93.4% success 5 agents LogisticsHITL ### Shipment exception handling When a delay or damage lands, impact is costed, the carrier rebooked and the customer hears before they ask. 12,800 runs / month 95.5% success 4 agents LogisticsHITL ### Customs document preparation HS codes, invoice and packing list prepared in minutes; only low-confidence lines reach a customs specialist. 4,300 runs / month 94.7% success 4 agents IoT & Connected DevicesHITL ### Cold-chain excursion response A temperature excursion is assessed instantly, product risk computed, a field crew dispatched, the quality decision recorded. 2,700 runs / month 97.4% success 4 agents Retail & E-commerce ### Where is my order? (WISMO) Customer asks, order and carrier events join instantly; on delay a new ETA is given and a trace opened. 110,000 runs / month 98.1% success 4 agents Sector architecture Derived from the 3 scenarios in the sector: supervisors, specialist agents, Hub model families and systems touched. Supervisors Control Tower SupervisorShipment SupervisorCustoms Supervisor Specialist agents Prediction AgentRouting AgentCustomer AgentFinance AgentException AgentImpact AgentAction AgentTariff AgentDocument AgentBroker Agent Models (Hub catalogue) aginies-tulpar-4aginies-talay-2aginies-sunkar-1 Systems touched TMSTelematicsCarrier APICustomer PortalERP BillingTMS (Transport Management)Salesforce CRMWhatsApp Business APISAP S/4HANAWMSTariff & Regulation Vector StoreCustoms Broker PortalDocument Store (S3) Channels apiemailwhatsappbatch Simulation volumes 113,100runs / month 93.7%success rate 6.9 smedian duration $0.0032estimated cost / run Summed from the simulation definitions of the sector scenarios; recalculated for your scale in the discovery session. Compliance and sovereignty - Human approval on customs declarations and high-cost re-routing - GDP-aligned cold-chain records and excursion documentation - Customer and shipment data inside the enterprise; every message recorded Regulatory frame - Customs law: declaration and document preparation by the agent, declaration and signature with the authorised broker - GDP and cold-chain rules: temperature excursions documented, the decision with the quality officer - Personal-data law and customer contracts: shipment and consignee data inside the organisation, every customer message recorded Security → Recommended deployment Sovereign shared deployment; TMS, telematics and carrier APIs connected through an allowlist, shipment data stays in Türkiye. Approval rules Human-approval thresholds of the sector scenarios. - Control tower: delay prediction, rerouting, customer notice, SLARoute cost increase > 15% or compensation > 10,000 → operations manager approval - Shipment exception handlingExtra cost > 1,000 € → dispatch lead approval - Customs document preparationHS confidence < 0.90 or duty difference > 2,000 € → customs specialist approval Measured targets | Delay detection lead | Hours ahead | | On-time delivery | +5–10 points | | “Where is my shipment” calls | 50% fewer | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Related solution packages Operational automation & demand managementAgents that classify, route and answer demands; faster processes, lower cost.Field service & dispatch automationSensor alarm or customer report → the right technician, the right part, the right time.Logistics control towerOperations agents that catch deviations early, reroute, inform the customer and compute the SLA. ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Manufacturing & Automotive URL: https://www.aginies.com/en/industries/manufacturing Language: en Summary: From supplier invoice to maintenance request: agents that decide and open work orders on top of ERP and MES. Manufacturing & Automotive # From supplier invoice to maintenance request: agents that decide and open work orders on top of ERP and MES. In plants, processes are split across ERP, MES, maintenance and quality systems; every hand-off is an e-mail, a form, a wait. Aginies runs three-way matching, maintenance triage and warranty claims end-to-end with agents connected to those systems; every out-of-tolerance decision goes to the responsible engineer or manager. In short In Manufacturing & Automotive, Aginies offers 4 live agent scenarios and 3 solution packages: Line stoppage response: from alarm to work order and shift report, Supplier invoice three-way match, Maintenance ticket triage, Warranty claim processing, Shipment exception handling, Device fleet anomaly triage; solutions: Invoice & document extraction, Agent outsourcing (agent workforce), Production and maintenance operations. The blockers on the way to production are Disconnected systems, Maintenance backlog, Invoice vs receipt gaps, Warranty cost; on the compliance side 3 requirements (including OT / IT separation preserved; agents connect to the production network read-only) are part of the deployment. Why it does not reach production 01 ### Disconnected systems ERP, MES and the maintenance system see the same event three times in three formats. 02 ### Maintenance backlog Requests queue without prioritisation; a critical fault sits behind routine ones. 03 ### Invoice vs receipt gaps Three-way matching is manual; tolerances vary by person. 04 ### Warranty cost Claims are assessed late and inconsistently; dealers and customers wait. Where to start: the first 90 days Three starting points; each measurable in a short pilot on existing systems. - 01 agent scenarioLine stoppage response: from alarm to work order and shift report → Unplanned stoppage is the costliest event; a pilot on one line with MES and historian connected. - 02 agent scenarioMaintenance ticket triage → Maintenance requests are high-volume and rule-based; the CMMS integration shows quick results. - 03 agent scenarioSupplier invoice three-way match → Measurable gain on the finance side; starts without touching the production network. Scenarios ## Agents you can watch live for this sector Manufacturing & AutomotiveHITL ### Line stoppage response: from alarm to work order and shift report An MES alarm lands; root cause found from history, spare part checked, work order and plan update ready within minutes. 3,100 runs / month 92.2% success 5 agents Manufacturing & AutomotiveHITL ### Supplier invoice three-way match Invoice, PO and goods receipt matched in SAP automatically; only out-of-tolerance lines reach a buyer. 21,000 runs / month 97.1% success 4 agents Manufacturing & AutomotiveHITL ### Maintenance ticket triage Sensor alarm and technician voice note become one work order, spare parts and a safe stop window included. 3,400 runs / month 95.2% success 5 agents Manufacturing & AutomotiveHITL ### Warranty claim processing Dealer claims validated for coverage, parts and labour in minutes; big amounts go to the regional manager. 8,600 runs / month 96.3% success 4 agents LogisticsHITL ### Shipment exception handling When a delay or damage lands, impact is costed, the carrier rebooked and the customer hears before they ask. 12,800 runs / month 95.5% success 4 agents IoT & Connected DevicesHITL ### Device fleet anomaly triage Telemetry anomalies from thousands of devices are clustered, root-cause candidates listed, one incident reaches the right team. 18,000 runs / month 96.9% success 4 agents Sector architecture Derived from the 4 scenarios in the sector: supervisors, specialist agents, Hub model families and systems touched. Supervisors Production SupervisorInvoice SupervisorMaintenance SupervisorWarranty Supervisor Specialist agents Diagnosis AgentParts AgentPlanning AgentReport AgentCapture AgentMatching AgentPosting AgentSignal AgentField Note AgentClaim AgentValidation AgentSettlement Agent Models (Hub catalogue) aginies-tulpar-4aginies-sunkar-1aginies-talay-2aginies-konrul-1aginies-voice-tr-v3 Systems touched MESSCADA HistorianCMMSERP Spare PartsShift Channel (Teams)SAP S/4HANAExchange / M365 MailTolerance Policy Vector StoreSCADA / HistorianCMMS (Maximo)SAP S/4HANA MMManuals & Work Order Vector StoreDealer PortalWarranty Policy Vector StoreDocument Store (S3) Channels apiappemailbatchvoiceweb Simulation volumes 36,100runs / month 96.3%success rate 7.1 smedian duration $0.005estimated cost / run Summed from the simulation definitions of the sector scenarios; recalculated for your scale in the discovery session. Compliance and sovereignty - OT / IT separation preserved; agents connect to the production network read-only - Human approval on out-of-tolerance invoice, warranty and maintenance decisions - Supplier and customer data stays inside; on-prem model option Regulatory frame - OT security (IEC 62443 approach): agents connect to the production network read-only and issue no commands; OT/IT separation kept - Product safety and warranty law: warranty decisions rule-based, out-of-tolerance files to people - Personal-data law and supplier contracts: supplier and customer data inside the organisation, sharing limited to the contract scope Security → Recommended deployment Hybrid: read-only agents and historian access near the production network on site, decision and document agents in our data centre in Türkiye or the organisation’s own. Approval rules Human-approval thresholds of the sector scenarios. - Line stoppage response: from alarm to work order and shift reportSchedule change > 2 hours or urgent purchase above 50,000 → shift lead approval - Supplier invoice three-way matchPrice deviation > 2% or > 1,000 TL → buyer approval - Maintenance ticket triagePlanned stop > 4 hours → shift supervisor approval - Warranty claim processingWarranty claim > 5,000 € → regional warranty manager approval Measured targets | Unplanned downtime | 20–30% lower | | Alarm-to-work-order time | Minutes | | Touchless supplier invoices | 70–85% | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Related solution packages Invoice & document extractionOCR → validation → ERP: clean records without manual entry.Agent outsourcing (agent workforce)Outsourced agents instead of outsourced people: iOS developer, business analyst, accountant, tester; with a work log.Production and maintenance operationsFrom line stoppage to work order, from maintenance request to warranty file: agents that decide on MES, CMMS and ERP. ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Retail & E-commerce URL: https://www.aginies.com/en/industries/retail Language: en Summary: Behind every customer question there is a transaction; the agents complete it inside the order system. Retail & E-commerce # Behind every customer question there is a transaction; the agents complete it inside the order system. Support volume multiplies at seasonal peaks; “where is my order”, returns, delivery changes and product questions overwhelm teams. Aginies answers with agents grounded in catalogue and policy knowledge, acts inside the order system, and runs back-office jobs like catalogue enrichment and review management on the same layer. In short In Retail & E-commerce, Aginies offers 6 live agent scenarios and 3 solution packages: Outfit stylist: a complete look by size, stock, budget and style, Where is my order? (WISMO), Returns and refunds, Product Q&A assistant, Catalogue enrichment (batch), Review response and quality escalation; solutions: E-commerce support & action, Email classification & attachment verification, Call-centre quality monitoring. The blockers on the way to production are Seasonal peaks, Bots that answer but do not act, Catalogue debt, Review and complaint noise; on the compliance side 3 requirements (including Privacy-compliant customer data processing) are part of the deployment. Why it does not reach production 01 ### Seasonal peaks November–December volume is three times the rest of the year; teams do not stretch. 02 ### Bots that answer but do not act The customer wants a cancellation; the bot says “let me transfer you”. 03 ### Catalogue debt Keying product attributes from supplier documents takes weeks. 04 ### Review and complaint noise Product-defect signals get lost among thousands of reviews. Where to start: the first 90 days Three starting points; each measurable in a short pilot on existing systems. - 01 agent scenarioOutfit stylist: a complete look by size, stock, budget and style → Direct revenue impact: basket size and return rate measured at the same time. - 02 agent scenarioWhere is my order? (WISMO) → “Where is my order” calls are the biggest support item; an agent wired to the order system shows results in week one. - 03 agent scenarioReturns and refunds → With a clear return policy automation is fast; exceptions stay with people. Scenarios ## Agents you can watch live for this sector Retail & E-commerceHITL ### Outfit stylist: a complete look by size, stock, budget and style The customer sends a photo and a sentence; the agent builds three looks that are in stock, fit and within budget, and adds the chosen one to the cart. 142,000 runs / month 96.1% success 5 agents Retail & E-commerce ### Where is my order? (WISMO) Customer asks, order and carrier events join instantly; on delay a new ETA is given and a trace opened. 110,000 runs / month 98.1% success 4 agents Retail & E-commerceHITL ### Returns and refunds Policy and history checked in seconds, label issued; the refund executes when the carrier scans the parcel. 48,000 runs / month 96.7% success 4 agents Retail & E-commerce ### Product Q&A assistant Size, compatibility and spec questions answered from catalogue and return data; stock and price verified. 120,000 runs / month 98.7% success 4 agents Retail & E-commerceHITL ### Catalogue enrichment (batch) Attributes pulled from supplier PDFs and images, TR/EN copy written; hundreds of SKUs ready in minutes. 900 runs / month 95.8% success 4 agents Retail & E-commerceHITL ### Review response and quality escalation Reviews classified and answered in brand tone; recurring defects joined with return data and escalated to QA. 26,000 runs / month 97.2% success 4 agents Sector architecture Derived from the 6 scenarios in the sector: supervisors, specialist agents, Hub model families and systems touched. Supervisors Style SupervisorOrder SupervisorReturns SupervisorShopping SupervisorCatalogue SupervisorVoice-of-Customer Supervisor Specialist agents Vision AgentStylist AgentCatalogue AgentCart AgentIntent AgentTracking AgentReply AgentIntake AgentPolicy AgentFulfilment AgentDocument AgentNormalisation AgentContent AgentClassification AgentResponse AgentQuality Agent Models (Hub catalogue) aginies-konrul-1aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Systems touched WhatsApp BusinessProduct Catalogue (PIM)InventoryCustomer ProfileCart & Checkout APIShopify / OMSCarrier API (Yurtiçi, DHL)WhatsApp Business APIZendeskPayment Gateway (iyzico)Email (Exchange)Vector StorePIM / CatalogueWeb Chat WidgetDocument Store (S3)SAP ERPReview Platform APIJira (QA) Channels appwebwhatsappvoiceemailbatchapi Simulation volumes 446,900runs / month 97.4%success rate 5.9 smedian duration $0.001estimated cost / run Summed from the simulation definitions of the sector scenarios; recalculated for your scale in the discovery session. Compliance and sovereignty - Privacy-compliant customer data processing - Human approval on out-of-policy refunds and campaign exceptions - A trace for every action written to the order system Regulatory frame - Consumer law and distance-selling rules: withdrawal, return and delivery deadlines coded into agent rules - Personal-data law and marketing consent: contact only on consented channels, recommendations on purpose-limited data - Campaign and price rules: discount authority bounded by the matrix, out-of-band offers to customer service Security → Recommended deployment Sovereign shared deployment; GPU capacity scales on usage at seasonal peaks, customer data stays in Türkiye. Approval rules Human-approval thresholds of the sector scenarios. - Outfit stylist: a complete look by size, stock, budget and styleDiscount outside campaign rules or a bespoke price request → customer service - Returns and refundsRefund > 2,500 or > 5 returns in 12 months → team lead approval - Catalogue enrichment (batch)New category or supplier price deviation > 15% → category manager approval - Review response and quality escalation1-star review replies and defect escalations → community manager approval Measured targets | First-contact resolution | +15–25 points | | Extra representatives at seasonal peak | 40–60% lower | | Basket size with outfit recommendations | +10–20% | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Related solution packages E-commerce support & actionWhere is my order, returns, delivery changes: the agent completes the transaction in the order system.Email classification & attachment verificationThe inbox reaches the right team; missing attachments are caught before work starts.Call-centre quality monitoring100% of conversations scored against objective criteria; sampling ends. ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Software Delivery URL: https://www.aginies.com/en/industries/software Language: en Summary: PM, analyst, architect, developer and test agents: the whole line from feature request to production. Software Delivery # PM, analyst, architect, developer and test agents: the whole line from feature request to production. In software teams the bottleneck is usually the rest of the line: requirement clarification, design decisions, review, release, incident response. Aginies runs that line with an SDLC agent team connected to Jira, Git and CI/CD; Code Aginies provides completion inside the IDE with the organisation’s own models. Production decisions and architectural approval always stay with people. In short In Software Delivery, Aginies offers 8 live agent scenarios and 3 solution packages: Legacy module modernisation: analysis, tests, rewrite, PR, Feature delivery pipeline, CI/CD release pipeline, Incident response and hotfix, Code review and security gate, Backlog grooming and estimation, Staged OTA update rollout; solutions: SDLC agent team, IoT device fleet operations, Agent outsourcing (agent workforce). The blockers on the way to production are Requirement ambiguity, Review bottleneck, Release risk, Code leakage concern; on the compliance side 3 requirements (including Code and completion models inside the organisation; Code Aginies on-prem) are part of the deployment. Why it does not reach production 01 ### Requirement ambiguity Three meetings before a request becomes a story; acceptance criteria written afterwards. 02 ### Review bottleneck PRs wait for days; security scanning is a separate queue. 03 ### Release risk Release notes, scope and rollback plans by hand; every release is a late night. 04 ### Code leakage concern Cloud code assistants carry the organisation’s source code outside. Where to start: the first 90 days Three starting points; each measurable in a short pilot on existing systems. - 01 agent scenarioFeature delivery pipeline → The requirement-to-PR line is the most visible bottleneck; a 4–6-week pilot with one team. - 02 agent scenarioCode review and security gate → A security and quality gate on every PR is low-risk and measured immediately. - 03 agent scenarioLegacy module modernisation: analysis, tests, rewrite, PR → Modernisation programmes run for years; characterisation tests make the steps safe. Scenarios ## Agents you can watch live for this sector Software DeliveryHITL ### Legacy module modernisation: analysis, tests, rewrite, PR Starts from a Jira epic; the old module is read, its behaviour pinned with tests, rewritten and opened as a PR with evidence. 1,900 runs / month 90.4% success 5 agents Software DeliveryHITL ### Feature delivery pipeline A Jira epic turns into criteria, design, code and tests via agents; humans only review and approve the PR. 1,200 runs / month 91.2% success 5 agents Software DeliveryHITL ### CI/CD release pipeline Build, tests, scans and staging run unattended; production ships only on release-manager approval, via ArgoCD. 950 runs / month 94.7% success 5 agents Software DeliveryHITL ### Incident response and hotfix Alert fires, bug reproduced, patch and regression run ready; on-call approves, post-mortem already drafted. 320 runs / month 90.5% success 5 agents Software DeliveryHITL ### Code review and security gate Agents review, scan and coverage-check every PR; fixes land first, a human approves only clean PRs. 4,200 runs / month 93.6% success 5 agents Software DeliveryHITL ### Backlog grooming and estimation Incoming requests are triaged, clarified and sized; a sprint proposal goes to the product owner for approval. 340 runs / month 92.8% success 4 agents IoT & Connected DevicesHITL ### Staged OTA update rollout Firmware rolls out in 1% → 10% → 100% stages; a health gate at every stage, human approval for critical fleets. 320 runs / month 98.1% success 4 agents Sector architecture Derived from the 8 scenarios in the sector: supervisors, specialist agents, Hub model families and systems touched. Supervisors Modernisation SupervisorDelivery SupervisorRelease SupervisorIncident SupervisorReview SupervisorPlanning SupervisoriOS Developer SupervisorQA Supervisor Specialist agents Analysis AgentTest AgentDeveloper AgentReview AgentAnalyst AgentArchitect AgentTester AgentBuild AgentSecurity AgentRelease AgentPost-mortem AgentReviewer AgentProject Manager AgentDevelopment AgentTest Plan AgentExecution AgentDefect Agent Models (Hub catalogue) aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Systems touched JiraGit (self-hosted)CI PipelineStatic AnalysisGitHubCI Runner (GitHub Actions)ConfluenceSonarQube + SnykKubernetes / ArgoCDSlackObservability (Grafana / Datadog)SonarQubeDependency Scanner (Snyk)Slack / TeamsXcode CloudFigmaWorkforce Work LogCI/CDDevice / Browser GridJira / Xray Channels apibatchwebapp Simulation volumes 9,550runs / month 92.7%success rate 15.8 smedian duration $0.0148estimated cost / run Summed from the simulation definitions of the sector scenarios; recalculated for your scale in the discovery session. Compliance and sovereignty - Code and completion models inside the organisation; Code Aginies on-prem - Human approval on production deployment and architectural decisions - SAST, dependency and secret scanning on every PR; evidence stored with the release Regulatory frame - Source code and intellectual property: code never leaves the organisation; Code Aginies and completion models on-prem - Secure development requirements (incl. regulator expectations in regulated sectors): SAST, dependency and secret scanning on every PR, evidence stored per release - Production deployment and architecture decisions with people; the agent writes only in assigned repositories and branches Security → Recommended deployment Code Aginies and coding models in the organisation’s data centre or in our data centre in Türkiye through the closed-circuit deployment; SDLC agents connected to Jira, Git and CI/CD with service accounts. Approval rules Human-approval thresholds of the sector scenarios. - Legacy module modernisation: analysis, tests, rewrite, PRProduction schema change or out-of-scope file → architect approval; PR merge always human - Feature delivery pipelineEvery PR → at least 1 human code review approval; agents never push to main directly - CI/CD release pipelineEvery production deploy → release-manager approval mandatory; a critical CVE closes the gate automatically - Incident response and hotfixEvery emergency production deploy → on-call engineer approval; the hotfix PR still gets a human review - Code review and security gateEvery PR → human CODEOWNERS approval; a high SAST finding or coverage < 80% blocks the gate - Backlog grooming and estimationSprint scope → product owner approval; agents only propose, never commit items to a sprint - iOS feature implementation (hired developer agent)Every PR is reviewed by a person; production deployment under the team lead’s approval - Regression test cycle (hired QA agent)The release “go / no-go” decision stays with the release manager Measured targets | PRs accepted on first review | 60–75% | | Requirement-to-PR lead time | 40–50% shorter | | Test coverage | +15–25 points | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Related solution packages SDLC agent teamPM, analyst, architect, developer and tester agents working together in the CI/CD pipeline.IoT device fleet operationsAnomaly clustering, root cause and health-gated OTA rollouts: one incident, one decision per fleet.Agent outsourcing (agent workforce)Outsourced agents instead of outsourced people: iOS developer, business analyst, accountant, tester; with a work log. ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Telecom URL: https://www.aginies.com/en/industries/telecom Language: en Summary: Millions of subscribers, thousands of scenarios: billing, outages, fraud and offers on one layer. Telecom # Millions of subscribers, thousands of scenarios: billing, outages, fraud and offers on one layer. Telecom operations mean volume and variety: bill disputes, SIM changes, outage reports, plan upgrades. Aginies runs these end-to-end with agents connected to billing, provisioning and network systems; transactions carrying fraud risk route to people by risk score. In short In Telecom, Aginies offers 5 live agent scenarios and 3 solution packages: Subscriber care assistant: bill, plan, fault and contract, Bill dispute resolution, SIM swap fraud check, Network outage communications, Plan upgrade advisor; solutions: Call-centre quality monitoring, Voice analysis & quality, Subscriber care and network support. The blockers on the way to production are Bill-dispute volume, SIM-swap fraud, Outage communication, Offer timing; on the compliance side 3 requirements (including Subscriber data in-country; personal data masking) are part of the deployment. Why it does not reach production 01 ### Bill-dispute volume The call centre jams every billing cycle; most disputes fit a handful of patterns. 02 ### SIM-swap fraud A process squeezed between speed and security; costs in both directions. 03 ### Outage communication Affected subscribers call in before anyone informs them. 04 ### Offer timing Usage data exists; the right offer at the right moment does not. Where to start: the first 90 days Three starting points; each measurable in a short pilot on existing systems. - 01 agent scenarioSubscriber care assistant: bill, plan, fault and contract → Multi-topic calls are the costliest; one-session resolution cuts average handling time directly. - 02 agent scenarioBill dispute resolution → Bill disputes are high-volume and rule-based; the BSS integration is enough for the first pilot. - 03 agent scenarioSIM swap fraud check → High fraud risk, clear rule set; goes live quickly with human approval retained. Scenarios ## Agents you can watch live for this sector TelecomHITL ### Subscriber care assistant: bill, plan, fault and contract The subscriber calls or writes; bill dispute, plan change and line fault are resolved in BSS and network systems in one session. 410,000 runs / month 94.3% success 5 agents TelecomHITL ### Bill dispute resolution Invoice lines and usage explained in seconds; goodwill credit applied by policy, larger amounts need a lead. 88,000 runs / month 96.3% success 4 agents TelecomHITL ### SIM swap fraud check Every SIM swap is risk-scored before provisioning; suspicious requests are held and banks are signalled. 41,000 runs / month 97.7% success 4 agents TelecomHITL ### Network outage communications When an outage ticket opens, affected subscribers are found and the approved message sent before they call. 1,400 runs / month 98.5% success 4 agents TelecomHITL ### Plan upgrade advisor Usage analysed, the right plan offered with a comparison; on acceptance the change provisions instantly. 52,000 runs / month 95.8% success 4 agents Sector architecture Derived from the 5 scenarios in the sector: supervisors, specialist agents, Hub model families and systems touched. Supervisors Subscriber Care SupervisorBilling SupervisorSIM Security SupervisorOutage Communications SupervisorPlan Supervisor Specialist agents Dialogue AgentAccount AgentNetwork AgentRetention AgentIntent AgentBill Analysis AgentAdjustment AgentRisk AgentVerification AgentProvisioning AgentDetection AgentImpact AgentCommunications AgentUsage AgentOffer Agent Models (Hub catalogue) aginies-talay-2aginies-sunkar-1aginies-tulpar-4 Systems touched IVR / SIPBSS / BillingOSS NetworkCRMSMS GatewayBilling (Amdocs)Usage Mediation / CDR StoreSalesforce CRMHLR / ProvisioningBank SIM-Swap Signal APIServiceNowNetwork OSSProduct Catalogue / Offer EngineWhatsApp Business API Channels voiceappwhatsappbranchapismsbatch Simulation volumes 592,400runs / month 95.0%success rate 7.3 smedian duration $0.0016estimated cost / run Summed from the simulation definitions of the sector scenarios; recalculated for your scale in the discovery session. Compliance and sovereignty - Subscriber data in-country; personal data masking - Human approval on SIM changes and high-value adjustments - A trace for every write to billing and provisioning Regulatory frame - Telecom regulator rules and the electronic communications law: subscription, contract and number-porting operations recorded and traceable - Personal-data law and traffic data: subscriber data in Türkiye, recordings and transcripts masked - SIM swap and identity operations: extra verification and human approval; suspicious requests to the fraud team Security → Recommended deployment Sovereign shared deployment or a dedicated tenant in the operator’s own data centre; speech models on in-country GPUs, connected to the existing switch over SIP. Approval rules Human-approval thresholds of the sector scenarios. - Subscriber care assistant: bill, plan, fault and contractContract cancellation, bill correction above 500 or technician priority → representative approval - Bill dispute resolutionGoodwill credit > 75 → team lead approval - SIM swap fraud checkRisk score ≥ 70 or credential change within 72 hours → fraud desk approval - Network outage communicationsAffected subscribers > 10,000 or P1 outage → network operations manager approval - Plan upgrade advisorCommitment-based or discounted bespoke offer → campaign manager approval Measured targets | Requests resolved without a representative | 60–70% | | Unnecessary technician visits | 35–45% fewer | | Average handling time | 40% shorter | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Related solution packages Call-centre quality monitoring100% of conversations scored against objective criteria; sampling ends.Voice analysis & quality100% of calls transcribed with Turkish STT, scored and coached; models run on-prem.Subscriber care and network supportBill, plan, contract and fault operations on voice and text channels, inside BSS and network systems. ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Methodology URL: https://www.aginies.com/en/methodology Language: en Summary: The five-layer assessment framework: Value, Readiness, Feasibility, Operations, Governance. Every stage ends at a clear decision point: proceed, fix or stop. Methodology # Not every AI idea deserves to be built. Those that do should not start unprepared. Every use case passes our five-layer assessment framework. It is the shared language for every decision, from portfolio prioritisation to live operation. Each stage ends at a clear decision point: proceed, fix or stop. 01 ## Value Why should this be done at all? We start from the business case: problem, benefit, measurement and strategic fit are defined in numbers. - Problem: which business problem, how is it solved today, what does it cost? - Benefit: cost reduction, revenue, risk reduction, experience; which and how much? - Measurement: target KPIs, expected improvement, ROI and payback period - Strategic fit: aligned with company strategy, a management priority? Duration1–2 weeks ParticipantsBusiness owner, finance, sponsor Tools and outputs - Scenario portfolio workshop and value card (problem, volume, cost, KPI) - Comparison with the closest scenario in the agent library Red flags - Benefit without numbers: “efficiency will rise” but no volume or cost - No sponsor; the request comes from IT and the business watches Output A business case that speaks in numbers and a prioritised scenario portfolio 02 ## Readiness Do we have the inputs we need? Even the best idea stalls without data and organisational readiness. We check four axes before starting. - Data: sources available and digital; quality, freshness and ownership clear? - Integration: systems and APIs ready; real-time access possible? - Organisation: does the business unit own it; sponsor and experts identified? - Technology: cloud/security infrastructure, AI platform and licences ready? Duration1–2 weeks ParticipantsData owner, integration team, process owner Tools and outputs - Readiness scorecard: data, integration, process, organisation (each axis 0–2) - API and data-access test; sample data set Red flags - The target system has no API, only a screen - Process rules live only in people’s experience Output A readiness scorecard with gaps closed before the project; no surprises 03 ## Feasibility Can we actually build this? An honest engineering assessment: AI is not always the right answer. Classic software, classic ML, agents or generative AI; which one is needed? - Technical fit: should AI solve this at all; which approach? - Complexity: how many systems, data sources and users are affected? - Model performance: expected accuracy, impact of a wrong output, is human approval needed? - Project feasibility: MVP duration, team, external dependencies, technical risks Duration2–3 weeks ParticipantsAginies engineering, enterprise architecture, information security Tools and outputs - Approach decision: classic software, classic ML, agents, generative AI - Prototype and draft test corpus; wrong-output impact analysis Red flags - A wrong output is irreversible and no approval point is defined - Scope grows at every meeting Output A realistic MVP plan with clear scope and identified risks 04 ## Operations How will it live after go-live? AI systems are run like products; after go-live, monitoring, maintenance and release management are part of the job. - Operating model: product owner, process owner, operations and support defined - User management: training, adoption tracking, feedback loops - Monitoring: performance, metrics and error management in real time - Financial operations: monthly run cost, cost projection with growth, budget ownership Duration1 week (plan) + continuous after go-live ParticipantsProduct owner, process owner, operations and support Tools and outputs - Operating-model template: roles, thresholds, review sample, budget owner - Dashboards and the monthly review calendar Red flags - Nobody owns operation after go-live - No budget owner for the monthly run cost Output A living, measured, continuously improving AI product 05 ## Governance Is it secure, controlled and auditable? AI must meet the same standards as every other critical system. Governance is not bolted on; it is part of the architecture. - Security: sensitive/personal data map, data residency, control of data flowing to models, masking - Compliance: privacy impact, sector regulation, internal policy requirements - Risk: assessment of hallucination, wrong decisions, data leakage and prompt injection - Control and auditability: human approval points, automation limits, kill-switch, decision history Duration2 weeks, in parallel with the other layers ParticipantsInformation security, compliance, legal, internal audit Tools and outputs - Data map and masking rule set; approval matrix - Control set and evidence list; sector regulation mapping Red flags - Personal data involved, no map - Governance postponed as “we will add it later” Output An audit-ready, controlled, trustworthy AI operation Transformation journey ## How we progress - 1 ### Discovery & portfolio Collect scenarios with business units, prioritise with the Value layer - 2 ### Assessment Readiness + Feasibility analysis of selected scenarios, MVP design - 3 ### MVP & validation Build on Aginies, pilot with real data, validate KPIs - 4 ### Production & scale Go live with the Operations + Governance layers, grow the portfolio Worked example ## Example: how the dialogue banking assistant passed the five layers - 01 Value · Why should this be done at all? Simple transactions reaching the call centre and cost per transaction put in numbers; target 40% reduction. Sponsor: digital banking. - 02 Readiness · Do we have the inputs we need? Transaction APIs and a test environment for core banking and cards existed; OTP stack ready; policy thresholds written with information security. - 03 Feasibility · Can we actually build this? Approach: supervisor + four specialist agents; per-transaction OTP and approval band against wrong-output risk; test corpus of 212 scenarios. - 04 Operations · How will it live after go-live? Product owner digital channels, process owner operations; monthly review and threshold updates on the calendar. - 05 Governance · Is it secure, controlled and auditable? Personal-data map and masking rules; limit increase > 60%, dispute > 5,000 and new payee to approval; all systems in Türkiye. Watch the scenario live: Dialogue banking: voice and text transaction assistant → Self-assessment ## Would your scenario pass all five layers? Apply the same framework to your own scenario. The result becomes the first thing we discuss in a discovery session. ### Would your scenario pass all five layers? 10 questions, 3 minutes. Answers stay in your browser; bring the result to a discovery session if you like. ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Platform URL: https://www.aginies.com/en/platform Language: en Summary: Aginies: enterprise agentic AI orchestration platform with eight integrated capability layers. Visual agent builder, model gateway, RAG, governance, observability, MCP integration and sovereign deployment. Platform # One platform. Every layer of AI execution. Build agents, manage knowledge, observe executions, connect systems. All from a single, production-grade control plane. Aginies is not a chatbot; it is the execution layer where all of the enterprise’s AI scenarios run. Platform facts derived from the library ## The numbers come from scenario definitions, not from copy. The 77 live scenarios in the agent library are read at build time; the figures and distributions below are derived from those definitions and change as scenarios are added. 77live scenarios 14sectors 261specialist agents 657distinct tools 200systems connected 73approval rules 127write steps in systems 3,320,970simulated runs / month Channel distribution (scenarios) api37 app34 web29 whatsapp27 email27 batch25 voice23 sms5 branch3 Step kinds (all steps) write127 tool call110 reason109 verify89 notify80 ingest77 human approval61 plan48 retrieve38 Model families (agent assignments) aginies-tulpar83 aginies-sunkar78 aginies-talay71 aginies-konrul21 aginies-voice-tr8 Autonomous Agent ## Build the agent by dialogue; test it, approve it, deploy it. Describe what you need: the Autonomous Agent sets up the supervisor and specialist agents, connects tools and the knowledge base, picks the models, writes and runs the test corpus, and deploys with your approval. Everything in the agent library can be produced through this dialogue; runs are observed in logs and the dashboard. Autonomous AgentAginies Workspace · Dialogue banking: voice and text transaction assistant Message the Autonomous Agent… AgentTestsLive The agent will appear here - Test corpus: by suite, scenario and turn; golden path, branches, robustness injection - Check families: customer resolved, PII before verification, state machine, memory round-trip, judge dimensions - Coverage report: which checks score, which are blind - Changes are versioned; deployment always stops at a human approval - Logs, conversations and the dashboard are queried from the same dialogue Product family ## 7 products, one platform. Platform Aginies is the umbrella; Voice, Code, Hub, Vision, Workforce and Dialog are its modules. Same governance, same logs, same licence. ### Platform Aginies Everything comes together here Orchestration core, visual agent builder, Autonomous Agent, knowledge base, logs and dashboards. The single control plane where every agent is built, observed and governed. - Visual agent builder: agent, router, memory, guardrail and human-approval blocks - Autonomous Agent: build agents by dialogue, test corpora, deployment - Knowledge base: document pipeline, versioning, chunk editor - Logs and dashboards: live runs, conversations, per-agent success heatmap - Workspace and tenant separation: teams share agents inside the organisation and build together - Monitoring: what happens per agent, platform health, model usage and cost - Reports: scheduled cost, performance, SLA and audit reports by unit, channel and branch - Templates, users, licences What is included - Visual agent builder, workspace and tenant management, roles via SSO - Execution engine, approval cards, kill-switch, dashboards and reports - Knowledge-base management and the Autonomous Agent Integration With the organisation’s identity provider (SAML/OIDC), SIEM, data warehouse and ticketing. Governance A separate policy set per workspace; four-eyes rule and versioning platform-wide. 77 / 77 scenarios use it Platform Aginies · Dashboard · last 30 daysall workspaces · all tenants executions13,120,000 success97.4% p50 latency1,4 s platform healthhealthy busiest agents · each cell ≈ 1 day dialog_banking_assistant%97.4 card_debt_payment%99.1 kvkk_data_detection%96.2 kvkk_document_masking%98.8 motor_claim_end_to_end%98.1 merchant_operations%100 model usage aginies-tulpar-438% aginies-talay-229% aginies-sunkar-119% aginies-voice-tr-v39% cloud · proxy$2.960 cost · by scenario Dialogue banking26% Credit & onboarding24% Claims operations18% KVKK masking14% Merchant ops10% Other8% PII masking: onanonymisation: onaudit trail: full1,920+ active agents ### Voice Aginies STT · TTS · ASR Speech models trained for Turkish: streaming speech-to-text, natural text-to-speech and real-time ASR. Voice agents through call-centre SIP integration. - aginies-voice-tr-v3 (STT), aginies-tts-tr-v2, aginies-asr-stream-v1 - On-prem serving; audio never leaves the enterprise - Voice analysis: quality scoring, sentiment and intent What is included - Streaming Turkish ASR and STT (aginies-voice-tr-v3), natural TTS (tts-tr-v2) - Voiceprint verification, spoken OTP, warm hand-off - Call quality analysis: transcript, sentiment, compliance checks Integration SIP with Genesys, Avaya and open-source switches; API with recording systems. Governance Recordings in-country; identity level and transaction threshold per channel. 23 / 77 scenarios use it Voice Aginies · SIP · call 0212-…-4471aginies-voice-tr-v3 · on-prem customerSomeone hit my car last night, plate 34 ABC 123, I want to report a claim. agentSorry to hear that. I found your policy by plate; can I take the location and time? STT180 ms intentclaim intake · 0.97 sentimentcalm TTS · aginies-tts-tr-v2ASR · stream-v1PII masking: on ### Code Aginies IDE Registry · IDE extensions · code completion For engineering teams: IDE extensions and code completion, an enterprise IDE registry, and the SDLC agent team (PM, analyst, architect, developer, tester) in the CI/CD pipeline. - In-house code completion; code never leaves - IDE Registry: approved extension and model distribution - SDLC agents work with Jira, Git and CI What is included - IDE extensions (VS Code, JetBrains) and in-house code completion - IDE Registry: approved extension and model distribution, version pinning - SDLC agent team: analysis, design, development, test, review Integration With Jira, Git (GitHub, GitLab, Bitbucket, self-hosted), CI/CD and static analysis tools. Governance Code never leaves the organisation; PR merge and production deployment always with people. 8 / 77 scenarios use it Code Aginies · billing-service/invoices.tscompletion · on-prem 14export async function exportInvoicePdf(id: string) { 15 const invoice = await repo.findById(id); 16 if (!invoice) throw new NotFound('invoice'); 17 const pdf = await renderer.render('invoice', maskPii(invoice)); 18 await audit.log('invoice.export', { id, user: ctx.user }); 19 return storage.put(`invoices/${id}.pdf`, pdf); 20} Tab: acceptPROJ-4821coverage 87%SAST ✓ IDE RegistryVS Code · JetBrains extensionDeveloper Agent · PR #2291 ### Hub Aginies Model Hub · MCP Hub · Model proxy Model catalogue: Aginies model families fine-tuned on enterprise data from Qwen, GLM and Kimi bases (tulpar, talay, sunkar, konrul) plus the current open releases of the same bases, on in-country GPUs. The catalogue is kept current: a new release is added to the Hub when it ships, the previous one stays for the contract term. Enter your own keys and let the platform reach cloud models as a proxy: single-point management, usage and cost tracking, masking. MCP Hub: approved tool servers. - aginies-tulpar-4 · talay-2 · sunkar-1 · konrul-1 (fine-tuned families, on-prem) - Current open releases of Qwen, GLM and Kimi in the same catalogue; aginies-voice-tr-v3 and tts-tr-v2 for speech - Model proxy: cloud models with your own keys, observed and masked from one point - Switch models with one selection; fallback strategy What is included - Model catalogue: tulpar, talay, sunkar, konrul families and current open releases - Model proxy and key management; usage and cost tracking - MCP Hub: approved tool servers, invocation policies, connection tests Integration With in-country GPU pools; with the organisation’s chosen cloud model providers through the proxy. Governance Model and tool changes versioned and gated by tests; keys in an encrypted vault. 77 / 77 scenarios use it Hub Aginies · Model Huball workspaces · October 2026 | model | access | usage · last 30 days | cost | | --- | --- | --- | --- | | aginies-tulpar-4 | on-prem · GPU 4× | 58% | — | | aginies-talay-2 | on-prem · GPU 2× | 84% | — | | aginies-sunkar-1 | on-prem · GPU 1× | 91% | — | | aginies-konrul-1 | on-prem · vision | 22% | — | | qwen · current release | on-prem · open model | 31% | — | | aginies-voice-tr-v3 | on-prem · STT | 47% | — | | cloud · proxy | your key | 12% | $288 | keys encrypted · never shown to usersall requests maskedMCP Hub · 14 servers ### Vision Aginies Documents · images · masking Image and document understanding: invoice, contract and form extraction, damage-photo assessment, visual quality control. Personal-data detection and document masking run on the same models. - aginies-konrul-1 · images and documents · on-prem - Field extraction with source document and confidence score - Personal-data detection and document masking What is included - Invoice, contract, form and ID document extraction; per-field confidence - Damage-photo assessment and visual quality control - Personal-data detection and document masking Integration With document stores (S3, SharePoint), scanners and ECM; results written to ERP and claims systems through tools. Governance Documents processed inside the perimeter; source document and extracted value stored together. 21 / 77 scenarios use it Vision Aginies · invoice_2026_1012.pdfaginies-konrul-1 · on-prem [Invoice with detected fields] extracted fields Invoice noFTR-2026-1012 Date12.10.2026 KDV%20 · 3.125,00 Total18.750,00 TL TCKNmasked (KVKK) source: page 1confidence ≥ 0.95 ### Workforce Aginies Agent workforce, on demand Outsourced agents instead of outsourced people: roles such as iOS developer, business analyst, accountant and test engineer are hired as agents that work inside your tools. Every task, output and hour-equivalent in the work log; a named team lead on the Aginies side. - Role catalogue: engineering, business analysis, finance, support, HR, legal - Work log: task, output, hour-equivalent, quality score, review status - Shadow period → sampled review → autonomous; thresholds and kill-switch - Agent-month or output-based billing; you own the outputs What is included - Role catalogue: engineering, business analysis, finance, support, HR, legal - Work log: task, output, hour-equivalent, quality score, review status - Shadow period → sampled review → autonomous; a named team lead Integration With the organisation’s own tools: Jira, Git, ERP, ticketing, e-mail. Governance Production and financial record changes under the organisation’s approval; the organisation owns the outputs. 17 / 77 scenarios use it Workforce Aginies· roster and work log · fintech-opsMon 09:00 hired agents · 4 ios-dev-01iOS Developer ready mobile0 h this week qa-01QA Engineer ready mobile0 h this week ba-01Business Analyst ready payments0 h this week acct-01Accountant ready finance0 h this week ### Dialog Aginies Dialogue UI The dialogue surface for customers and employees: web widget, mobile, WhatsApp and messaging. Session memory, human takeover and an audit trail for every message. - The same agent in text and voice channels - Human takeover and hand-back - Conversation dashboard and quality scoring What is included - Web and mobile dialogue UI; embeddable widget - WhatsApp, SMS and e-mail channels with the same agent - Approval cards, file and image sharing, warm hand-off Integration With the organisation’s identity and OTP stack; with the WhatsApp Business API and SMS gateways. Governance Per-transaction identity threshold; every conversation and transaction in the trace. 60 / 77 scenarios use it Dialog Aginies · web widgetcard_debt_payment v2 I want to pay my full card balance from my current account. Your statement balance is 24,860, due 14 October. Current account ****7741 has enough balance. If you confirm I will send a one-time code. ✓ get_card_debt · get_accounts · policy: no approval needed Confirm. Code sent. Once entered, the payment executes instantly and you get a confirmation. session memoryhuman takeover: readyaudit trail Type a message…↑ Architecture ## Wherever it needs to run. Same platform, same governance; different perimeter. Switch the deployment mode to see where the sovereign boundary falls. Platform, models and data inside the enterprise data centre. No external dependency; the Aginies model families and open releases are served in-house. [AGINIES ORCHESTRATION CORE] Conceptual view. Component names and responsibilities are shown; implementation details are shared in a discovery session. Capability layers ## Eight layers, one execution model. A ### Orchestration & Execution Intelligence the engine layer Runs complex work end-to-end; routes requests intelligently and keeps multi-step processes stable even when dependencies fail. - Supervisor → sub-agent → supervisor architecture - Dynamic agent routing - Stateful multi-step execution - Agent-to-agent communication (A2A) - Scheduled and webhook triggers - Agent templates - Autonomous Agent: build agents by dialogue with test corpora Scope - Supervisor plan, agent assignment and result merge; retry and fallback path on a failed step - Stateful execution: session and long-term memory, pending approvals, timeout rules - Parallel branches and loops; condition-based termination Governance Each agent’s authority is defined per tool; the supervisor cannot call out-of-scope tools. Evidence The run plan and every step’s input/output in the trace; replayable. B ### AI / Model Orchestration brain selection and resilience Uses the right model for each task; reduces vendor lock-in and keeps operations running when a provider is unavailable. - Multi-model gateway - Model proxy: single-point, observable access to cloud models with your own keys - Model switching without redesign - Fallback model strategy - On-prem serving for open-source models - Prompt testing environment and versioning - Hub Aginies: enterprise fine-tuned models (c5, g5, g4, oss) and speech models Scope - Hub catalogue: fine-tuned families on Qwen, GLM and Kimi bases (tulpar, talay, sunkar, konrul) and current open releases - Model proxy: cloud models with the organisation’s own keys, single-point observation and masking - Per-scenario model choice, fallback model and prompt versioning Governance Model changes are versioned and never reach production without passing the test corpus. Evidence Model version, tokens and cost recorded per request; catalogue history. C ### Knowledge & Retrieval enterprise memory Turns fragmented enterprise information into governed, searchable, trustworthy knowledge; answers link back to sources. - Document embedding pipeline - Versioned knowledge base management - Tag-based document segmentation - Hybrid search and multi-source RAG - Chunk-level inspection and editing - Hallucination-reduction architecture - Knowledge base management: sources, tags and access policy Scope - Document embedding pipeline, chunk inspection and tagging; versioned knowledge base - Access policy at chunk level: who may see which document through which agent - Cited answers; “I don’t know” when there is no source Governance Knowledge-base changes are approved by the owner; the previous version can be restored. Evidence Chunks and their versions used in every answer; coverage and quality report. D ### Governance, Risk & Compliance the control layer Makes AI auditable, safe and fit for regulated environments; every decision is traceable. - Guardrails and policy engine - Human-in-the-loop checkpoints - Agent decision traceability - Sensitive data masking and PII redaction - Masking and anonymisation of all requests - Full input/output logging and audit trail - Kill-switch and automation limits Scope - Guardrails and policy engine: input/output checks, masking, forbidden actions - Human-approval checkpoints: threshold, approver, timeout; approval card on mobile and web - Kill-switch per user, team and organisation; role-based authorisation Governance Policy changes under the four-eyes rule; every rule versioned with a reason. Evidence Approval records (who, when, what), guardrail interventions and kill-switch events. E ### Observability & Optimisation performance management Provides visibility into accuracy, speed, cost and health; produces the data for continuous improvement. - Per-run traces and timelines - Latency and token tracking - Cost attribution by scenario, unit and channel - Usage and cost per model; platform health - Routing accuracy measurement - Dashboard and analytics layer - Live logs, conversation records, per-agent success heatmap - Reports: scheduled, exportable (PDF, XLSX, CSV); internal audit template Scope - Per-run trace: step, duration, tokens, tool call, error - Dashboards: agent success heatmap, platform health, model usage, cost attribution - Forwarding to SIEM and data warehouse; alert rules Governance Traces are immutable and retained according to policy. Evidence One-click audit sample; monthly quality report per agent. F ### Integration & Connectivity enterprise reach Connects to enterprise systems and customer channels without rebuilding the existing technology landscape. - REST API and database integrations - Model Context Protocol (MCP) servers - Custom tool definitions and invocation policies - Native connectors: email, calendar, files, collaboration tools - Web widget and messaging integrations - MCP Hub: catalogue of approved tool servers Scope - REST, databases and MCP servers; invocation policy per tool - Ready connectors: core banking, ERP, CRM, ticketing, payments, switch - Read/write separation; write tools require approval by default Governance Each tool defined with a service account and scope; secrets kept in the organisation’s vault. Evidence Tool-call log: parameters (masked), response code, duration. G ### Multichannel Experience & Voice customer interaction layer Delivers AI across digital, messaging and voice channels on one consistent orchestration underneath. - Real-time speech-to-text and streaming voice processing - Call-centre SIP integration - WhatsApp, SMS and email services - Intent detection and entity extraction blocks - Session memory - Dialog Aginies: dialogue UI, human takeover - Voice Aginies: Turkish STT, TTS and ASR models Scope - Dialog Aginies on web and mobile; WhatsApp, SMS and e-mail channels - Voice Aginies: streaming ASR, Turkish STT/TTS, call centre over SIP - The same agent on every channel; warm hand-off and conversation summary Governance Identity level and transaction threshold defined per channel. Evidence Transcripts, recordings and summaries under the retention policy; quality score per channel. H ### Deployment & Enterprise Architecture fit to enterprise constraints Supports the on-premise, hybrid and sovereignty requirements of large enterprises. - On-premise and air-gapped deployment - Hybrid architecture: sensitive workloads inside, elastic capacity outside - Private cloud / dedicated region - SSO / SAML, fine-grained RBAC, workspace isolation - Per-tenant agent sharing and collaborative development - GPU-as-a-Service for in-country elastic capacity Scope - Sovereign shared (our data centres in Türkiye), dedicated tenant/GPU, on-prem or air-gapped in the organisation’s data centre - Access: VPN tunnel or allowlist over the internet - In-country elastic capacity with GPU-as-a-Service Governance Change management and incident notification in the contract; remote maintenance in windows the organisation opens. Evidence Network diagram, tunnel/allowlist configuration, backup and drill records. Visual agent builder ## Design agent pipelines without writing glue code. Drag, connect, configure. Every block, from agents and routers to memory, guardrails and human approval, is a first-class primitive you compose into any workflow. Business and IT look at the same screen. ### Agent A specialist with its own model, tools and memory. ### Supervisor Breaks the objective into steps, delegates to agents, merges the result. ### Router Branches the flow by intent, rule or score. ### Memory Shares session and long-term context across agents. ### Guardrail Checks input and output against policy; blocks or masks. ### Human approval Pauses on critical decisions, presents to the authorised person, records the decision. ### Tool / MCP Read or write to an enterprise system; invocation policy per tool. ### Parallel & loop Runs independent steps simultaneously and repeated work until a condition is met. Monitoring and reports ## Every run is visible, step by step. What happens per agent, the health of the platform, which model is used how much and what it costs, on one dashboard. A step-by-step record of every run: input, output, model, duration, tokens, approver. Every request can be masked and anonymised; everything is separated by workspace and tenant. - Per-agent executions, success and error heatmap - Platform health: latency, queues, GPU and model availability - Usage and cost per model, including cloud models through the proxy - Cost attribution by scenario, unit and channel - Masking and anonymisation on every request - Workspace / tenant separation; teams share and co-develop agents - Reports: scheduled cost, performance, SLA and audit reports; PDF, XLSX, CSV; delivered by email and shared folder - One report for internal audit: model version, rule set, user Platform Aginies · Dashboard · last 30 daysall workspaces · all tenants executions13,120,000 success97.4% p50 latency1,4 s platform healthhealthy busiest agents · each cell ≈ 1 day dialog_banking_assistant%97.4 card_debt_payment%99.1 kvkk_data_detection%96.2 kvkk_document_masking%98.8 motor_claim_end_to_end%98.1 merchant_operations%100 model usage aginies-tulpar-438% aginies-talay-229% aginies-sunkar-119% aginies-voice-tr-v39% cloud · proxy$2.960 cost · by scenario Dialogue banking26% Credit & onboarding24% Claims operations18% KVKK masking14% Merchant ops10% Other8% PII masking: onanonymisation: onaudit trail: full1,920+ active agents Platform Aginies · Reportsall workspaces · October 2026 monthly executions · last 12 monthslast 30 days: 13,120,000+31% YoY | report | breakdown | schedule | format | | --- | --- | --- | --- | | Monthly cost report | scenario · unit · channel · branch | 1st of month · 07:00 | PDF · XLSX | | Agent performance | success, latency, error classes | weekly · Monday | PDF | | Audit export | run traces, approvals, model versions | on demand | CSV · JSON | | Model usage | tokens, cost and proxy share per model | daily · 06:00 | XLSX | | SLA and capacity | p95 latency, queues, GPU occupancy, projection | weekly | PDF | delivered by email and shared folderinternal audit templatepersonal data masked Workforce Aginies · Roster and work logworkspace: fintech-ops · this week · tenants separate | agent | role · team | hour-eq. | output | quality | mode | | --- | --- | --- | --- | --- | --- | | ios-dev-01 | iOS Developer · mobile team | 14 h | 9 PRs · 21 points | 4.7 | autonomous · 10% review | | qa-01 | QA Engineer · mobile team | 31 h | 412 tests · 6 bugs | 4.8 | autonomous · 10% review | | ba-01 | Business Analyst · payments product | 9 h | 11 stories · 2 processes | 4.5 | shadow · 100% review | | acct-01 | Accountant · finance | 27 h | 238 entries · 3 reconciliations | 4.9 | autonomous · threshold-approved | work log · today 09:12ios-dev-01PR #4821 opened: payment screen accessibility fixes · 2 tests addedreview pending 09:40acct-0138 supplier invoices processed, 2 out of tolerance → to m.ozkan for approvalapproval 10:05qa-01Regression round 3: 412/412 · 1 new bug (BUG-2291) filed with evidencedone 10:30ba-01Story PAY-118 with 6 acceptance criteria written → to the product ownerreview weekly timesheet exportedbilling: agent-monthteam lead: e.demir (Aginies) Integration ## Plug in any tool with MCP. Add a server URL, configure headers, test the connection: agents can invoke its tools securely. With chunk-level control, every piece of knowledge is inspectable, editable and tagged. serverhttps://mcp.erp.internal/v1connected toolsget_purchase_order · post_invoice · get_vendor policypost_invoice → human approval > 50,000 authSSO · workspace: finance-ops chunks4.212 · tag: policy/v3 Integration catalogue ## 200 systems in 12 kinds: the enterprise systems the scenarios touch. Each row is a system kind with its examples from the library. Connecting a new system means one tool definition and one invocation policy. Enterprise APIs76 connections Access ControlAppointment SystemBank SIM-Swap Signal APICI PipelineCI Runner (GitHub Actions)CI/CDCI/CD (signed artefacts)Card Management+55 Customer channels50 connections Customer PortalHangar Tablet AppMicrosoft TeamsMobile App ChatOTP / SMS GatewayPush & SMS GatewaySMS GatewayShift Channel (Teams)+6 Core systems44 connections Active Directory / Entra IDBSS / BillingBilling (Amdocs)Billing (CIS)CLM (Contract Lifecycle Management)Card ManagementClaims SystemCore Banking+19 Databases and vectors44 connections Audit LogCollections PlatformControl & Sanctions Lists (vector)Eligibility Criteria (vector)Fleet RegistryGIS / Network ModelGIS / Zoning PlanKYC / Customer 360+29 Ticketing and work orders25 connections CMMSCMMS (Maximo)Case ManagementField ServiceITSM / IncidentJiraJira (QA)Jira / Xray+6 ERP and MES25 connections Billing / CIS (SAP IS-U)Billing / SettlementERPERP (SAP / Logo)ERP (SAP GTS)ERP BillingERP Spare PartsInventory+12 E-mail20 connections E-mailEmailEmail (Exchange)Exchange / M365 MailProcurement Inbox Document stores19 connections ConfluenceContract RepositoryDocument StoreDocument Store (S3)Tech Pub Repository (S1000D) CRM8 connections CRMCustomer CRM / BillingCustomer ProfileHubSpotSalesforce CRM Payment systems7 connections Bank APIPSP / Acquirer APIPayment Gateway (iyzico)Payment HubRefund ServiceSettlement Ledger Voice and switch6 connections Call Centre SIPGenesys SIPIVR / SIP HR systems3 connections LMSSAP SuccessFactorsWorkday Operations ## How it is run after go-live. Targets are settled with the organisation in the contract; the items below are the platform’s standard operating model. ### Versioning and rollback Agent definition, rule set, prompt and model version recorded together; rollback to the previous version is one step. ### Test corpus gate Every change passes the suite/scenario/turn test corpus before production; the coverage report is recorded. ### Continuity targets Target RTO 4 hours and RPO 1 hour on the sovereign shared deployment; set with the organisation on dedicated and on-prem. Backups inside Türkiye. ### Model catalogue upkeep When a new base release ships, the fine-tune is repeated and a new catalogue version opened; the previous one stays for the contract term. ### Incident management Notification times by severity in the contract; kill-switch with the organisation, root-cause report with Aginies. ### Monthly review Per-agent quality, cost and approval rates; threshold and rule updates together with the organisation’s administrator. Deployment models ## One platform, three deployments. From sovereign shared to air-gapped; access over a VPN tunnel or an allowlist. Details on the security page. Security → recommended startSovereign shared Our data centres in Türkiye, an isolated workspace for the organisation, access over VPN. - Starts within days - Model updates and capacity handled by Aginies - Fixed IPs, isolated tenant, SIEM forwarding higher isolationDedicated tenant / GPU A tenant and GPU pool reserved for the organisation in the same data centre; encryption with the organisation’s own key on request. - No resource sharing - Organisation-specific model versions - Capacity commitment by contract full controlOn-prem in the organisation’s data centre Hub, model servers and Platform in the organisation’s own data centre; remote maintenance only in windows the organisation opens. - No data leaves the organisation’s perimeter - Requires the organisation’s GPU investment - Installation and updates follow the organisation’s change calendar ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Security and Sovereignty URL: https://www.aginies.com/en/security Language: en Summary: SOC 2 Type II and ISO 27001 compliant platform; GDPR and KVKK compliance; on-prem, air-gap, hybrid and dedicated-region deployment; in-country elastic capacity with GPU-as-a-Service; full audit trail for every agent action. Security, compliance and sovereignty # Your data stays where it must. AI must meet the same standards as every other critical system. Aginies is built to the requirements of regulated sectors such as finance, insurance and healthcare. Security and auditability are not bolted on; they are part of the architecture. Platform, models and data inside the enterprise data centre. No external dependency; the Aginies model families and open releases are served in-house. [AGINIES ORCHESTRATION CORE] Conceptual view. Component names and responsibilities are shown; implementation details are shared in a discovery session. Closed-circuit deployment ## The code-assistant scenario: from IDE to model For the information-security teams of organisations that want to work in a closed circuit. The organisation’s developers reach Hub Aginies through the Code Aginies extension in their IDEs, over the organisation’s own VPN, on our infrastructure in data centres located in Türkiye and aligned with banking IT regulation; the coding models run there, on in-country GPUs. Code context never leaves Türkiye, no external model is called, no model is trained on the organisation’s data, and every request leaves a per-user trace. TürkiyePrimary and secondary systems in-country; data residency guaranteed 0External model calls; code context never leaves the sovereign perimeter 0Models trained on the organisation’s data; prompts are not learning data %100Per-request audit trail: user, model version, duration, cost [Request path from organisation IDEs over a VPN tunnel to Hub Aginies inside the Aginies sovereign perimeter, then to in-country GPU model servers; no external model calls and no training.] Request path: (1) the developer signs in with the organisation’s SSO; (2) the extension sends code context through the organisation’s VPN gateway and the tunnel to Hub Aginies; (3) licence and quota are verified; (4) the masked prompt goes to the model on in-country GPUs; (5) the suggestion returns to the IDE the same way; (6) every request leaves a trace in Platform Aginies; (7) traces are optionally forwarded to the organisation’s SIEM. ### The seven steps of one completion request - 1 IdentityThe developer signs in from the IDE with the organisation’s identity provider (SAML / OIDC, MFA). The extension carries a short-lived per-user token; the licence key is held at organisation level and stays outside the extension. - 2 RequestThe extension sends only the relevant code context (open file, cursor surroundings, selected symbol definitions) and the prompt; the repository as a whole is never sent. The connection is encrypted with TLS 1.3 inside the organisation’s VPN tunnel; the Hub accepts traffic only from the IP addresses allocated to the organisation. - 3 Licence and quotaHub Aginies validates the token, binds it to the organisation’s workspace and checks seats and quota. When a user is disabled at the organisation, access drops at the same moment. - 4 Masking and routingThe prompt is scanned with configurable rules for secrets (keys, passwords, connection strings) and personal data; matches are masked. The prompt is routed to the model selected in the contract. - 5 ModelThe model runs on GPUs in Aginies’ data centre in Türkiye. There is no external API call; prompt and response are not stored on the model side and are not used for training. - 6 Response and traceThe suggestion returns to the IDE through the same tunnel. Platform Aginies records the request with user, time, model version, duration, tokens and cost; whether prompt and response content is retained, and for how long, is decided together with the organisation. - 7 The organisation’s visibilityThe organisation’s administrators see usage and cost dashboards; traces can be forwarded one-way to the organisation’s SIEM via syslog or API. The kill-switch cuts access instantly per user, team or organisation. ### Two access models: VPN tunnel or allowlist over the internet The same request path opens in two ways. A VPN tunnel when the organisation wants a fully closed circuit; TLS over the internet with the organisation’s egress IPs on an allowlist when developers should start without installing anything extra. The second model was set up together with a bank’s information-security team; the diagram and flow table below are the anonymised version of that deployment. fully closed circuitVPN tunnel Site-to-site IPsec between the organisation’s gateway and Aginies; the Hub is reached only through the tunnel from the organisation’s allocated IPs and is invisible from the internet. - VPN gateway configuration on the organisation side - Hub unreachable from outside - Tunnel parameters in the technical annex quick startOver the internet, with an allowlist The endpoints are reachable from the public internet; the extension connects over TLS 1.3 from the organisation’s existing internet egress. The organisation’s egress IPs go on an allowlist so that rate-limit and auto-block rules do not cut its traffic; access itself is granted by identity. - On the organisation side only an outbound 443 rule and the egress IP list - No VPN client or server to install - Identity, quota, masking and trace are the same [Request path from organisation IDEs over the internet with TLS to the Aginies application server, then to the GPU model server on the internal network; no client VPN, organisation egress IPs on an allowlist, no external model calls.] | Flow | Source | Target | Port / protocol | Access and control | | --- | --- | --- | --- | --- | | 1 | Developer IDE · Code Aginies extension · organisation egress IPs | code-api.aginies.com → Aginies public IP | TCP 443 · HTTPS, TLS 1.3 | Internet; organisation IPs exempt from rate limit and auto-block. SSO token, seats and quota, masking, per-request trace. | | 2 | Aginies application server (internal interface) | GPU model server (internal network) | Internal network · model API | Internal network only; accepted only from the application server, never leaves the data centre. | | 3 | GPU model server | Internet | — | None: the server has no public IP and no internet egress; prompt and response are not stored or used for training. | | 4 | Aginies operations team | Aginies application server · administration | TCP 22 · SSH | Aginies operations VPN only; closed to the internet and to the organisation. No SSH or VPN needed on the organisation side. | | 5 | Aginies application server | External model providers | — | None: coding models run only on in-country GPUs; code context never leaves Türkiye. | | 6 | Organisation applications (optional, direct model access) | hub-api.aginies.com → Aginies public IP | TCP 443 · HTTPS, TLS 1.3 | Same allowlist, identity and quota; API access to the models in the Hub catalogue. | Addresses in the diagram are anonymised; the real IPs, hostnames and firewall rules are given to the organisation’s information-security team in the technical annex. ### Goes from the organisation to Aginies - The relevant part of the open file, cursor surroundings and selected context (code context) - The developer’s prompt (question, command) - User identity (SSO token), IDE and extension version - Minimal technical telemetry for error reports (optional, can be switched off) ### Goes nowhere - The repository as a whole, build outputs, environment files, secret keys (masked) - No prompt from the Aginies perimeter to the public internet or cloud model providers - No log, backup or copy outside Türkiye - No organisation data into model training or another customer’s workspace ### The first questions a security team asks **Where do the systems run? Does data leave Türkiye?** The whole platform runs in our data centres in Türkiye; primary and secondary systems are in-country. Code context, logs and backups never leave Türkiye. This is aligned with the in-country requirement of banking IT regulation. **Is Hub Aginies exposed to the internet?** No. The Hub is reachable only through the tunnel interface; it has no public endpoint, so there is no DDoS or scanning surface. **How are users authenticated? What if the licence key is shared?** Identity lives in the organisation’s identity provider: SSO via SAML or OIDC, MFA by the organisation’s policy. The licence key is organisation-level and grants no access on its own; every request carries a short-lived per-user token. When a user is disabled at the organisation, Aginies access drops at the same moment. **Can Aginies staff access the organisation’s data?** Production access is role-based, named and logged; access to prompt and response content is opened only on the organisation’s written request for support purposes, and the record is shared with the organisation. Under the default retention policy, content is not kept at all. **What exactly does the extension send?** Only the context the request needs: the relevant part of the open file, cursor surroundings, selected symbol definitions and the developer’s prompt. The repository as a whole, build outputs and environment files are never sent. The context limit and the exclusion list (.env, certificates, key files) are configured with the organisation and visible in the extension. **Do secrets (passwords, API keys) reach the model?** The masking layer in the Hub scans the prompt for keys, passwords, connection strings and personal-data patterns; matches are masked before they reach the model. The rule set can be extended with the organisation’s own patterns; masking events appear in the trace. **Are prompts and responses retained? For how long?** Metadata (user, time, model version, duration, tokens, cost) is always recorded. Retaining prompt and response content is the organisation’s decision: off (default), or encrypted retention for a period the organisation sets. Deletion at expiry is automatic and evidenced. **Is our data used to train models?** No. The organisation’s prompts and responses never enter any model’s training or fine-tuning data; this is an explicit clause in the contract. If organisation-specific fine-tuning is requested, a separate model version owned by the organisation is produced only from data the organisation approves. **Which models are used? Does anything go to an outside provider?** Coding models from the Hub Aginies catalogue served on in-country GPUs: open-source-based on-prem models (aginies-talay-2, aginies-sunkar-1) and fine-tuned enterprise versions. No prompt goes to an outside provider; the cloud model proxy is off in the organisation’s workspace and can be locked at policy level. **What is the notification time for a security incident?** For an incident affecting the organisation, first notification is made within 24 hours, followed by a root-cause and corrective-action report. KVKK notification duties are coordinated with the organisation as data controller. **Can we cut the service instantly in an emergency?** Yes. The kill-switch is in the organisation administrator’s hands: it cuts access instantly per user, team or organisation; the organisation can also close the VPN tunnel on its own side to end all access. The full question list and the control matrix are in the docs. ### Same architecture, three deployments recommended startSovereign shared Our data centres in Türkiye, an isolated workspace for the organisation, access over VPN. - Starts within days - Model updates and capacity handled by Aginies - Fixed IPs, isolated tenant, SIEM forwarding higher isolationDedicated tenant / GPU A tenant and GPU pool reserved for the organisation in the same data centre; encryption with the organisation’s own key on request. - No resource sharing - Organisation-specific model versions - Capacity commitment by contract full controlOn-prem in the organisation’s data centre Hub, model servers and Platform in the organisation’s own data centre; remote maintenance only in windows the organisation opens. - No data leaves the organisation’s perimeter - Requires the organisation’s GPU investment - Installation and updates follow the organisation’s change calendar ## Deployment options ### On-premise / air-gap Sensitive data never leaves; runs in your own data centre. Models are served inside; there is no external dependency. ### Hybrid Sensitive workloads inside, elastic capacity outside. External models only with masked data and through a secure gateway. ### Private cloud / dedicated region A single-tenant region reserved for the enterprise; data residency guaranteed. ### GPU-as-a-Service Elastic in-country GPU capacity that scales with usage and needs no capital outlay. ## Compliance matrix | SOC 2 Type II | Compliant platform; the control set is documented for internal audit | | ISO 27001 | Compliant information security management | | GDPR & KVKK | Personal data map, data residency guarantee, deletion and access requests | | Banking IT regulation | For banks, primary and secondary systems in-country; audit-ready deployment | | SSO / SAML & RBAC | Fine-grained roles, workspace isolation, per-workflow access policies | | PII masking | Sensitive data masking at ingestion, configurable redaction | | Audit trail | Input, output, model version, rule set and user for every agent action and model call | | Kill-switch | Emergency shutdown and automation limits on critical decisions; mandatory human approval | ## Model-agnostic, fine-tuned Hub Aginies holds the Aginies families fine-tuned on Qwen, GLM and Kimi bases (tulpar, talay, sunkar, konrul), the current open releases of the same bases and Turkish speech models in one catalogue, all on in-country GPUs. Cloud models can be added through the proxy with the organisation’s own key. No vendor lock-in. Switching models does not mean rewriting the pipeline. Each scenario gets the model that balances accuracy, latency and cost; a fallback strategy keeps operations running through provider outages. aginies-tulpar-4 · on-premaginies-talay-2 · on-premaginies-sunkar-1 · on-premaginies-konrul-1 · on-premqwen · güncel sürümglm · güncel sürümkimi · güncel sürümaginies-voice-tr-v3 · on-premaginies-tts-tr-v2 · on-prem Hub Aginies model catalogue: Aginies families fine-tuned on Qwen, GLM and Kimi bases (tulpar, talay, sunkar, konrul), the current open releases of the same bases and Turkish speech models; all on in-country GPUs. Cloud models can be added through the proxy with the organisation’s own key. Sector regulatory frames ## Which frame each sector answers to, and with which deployment. The regulatory frame and recommended deployment from the industry pages, in one table. Details and the first-90-days plan on each industry page. | Sector | Frame | Recommended deployment | | --- | --- | --- | | Banking → | Banking IT and electronic banking regulation: primary and secondary systems in Türkiye, audit right in the outsourcing contract, cloud model subject to bank-specific permission | Start with the sovereign shared deployment in our data centres in Türkiye (isolated bank workspace, VPN or allowlist access); on-prem in the bank’s data centre for scenarios classed as primary systems. | | Fintech & Payments → | Central-bank payment and e-money IT communiqué: community cloud only with an approved provider; primary systems in-country | In the community cloud of a provider on the central bank’s approved list, or in the institution’s own data centre; GPU capacity in-country, models from the Hub catalogue. | | Insurance → | Insurance regulator rules and insurance IT requirements: claim and policy decisions reasoned, time-stamped, open to audit | Dedicated region in Türkiye or the sovereign shared deployment; vision models for damage photos and health data on in-country GPUs. | | Retail & E-commerce → | Consumer law and distance-selling rules: withdrawal, return and delivery deadlines coded into agent rules | Sovereign shared deployment; GPU capacity scales on usage at seasonal peaks, customer data stays in Türkiye. | | Telecom → | Telecom regulator rules and the electronic communications law: subscription, contract and number-porting operations recorded and traceable | Sovereign shared deployment or a dedicated tenant in the operator’s own data centre; speech models on in-country GPUs, connected to the existing switch over SIP. | | Healthcare → | Personal health-data regulation and special-category data: health data inside the hospital perimeter, access by role and purpose, every access logged | On-prem in the hospital data centre; models inside, no external connection. For hospital groups, a dedicated tenant in the group data centre. | | Manufacturing & Automotive → | OT security (IEC 62443 approach): agents connect to the production network read-only and issue no commands; OT/IT separation kept | Hybrid: read-only agents and historian access near the production network on site, decision and document agents in our data centre in Türkiye or the organisation’s own. | | Logistics → | Customs law: declaration and document preparation by the agent, declaration and signature with the authorised broker | Sovereign shared deployment; TMS, telematics and carrier APIs connected through an allowlist, shipment data stays in Türkiye. | | Energy & Utilities → | Energy regulator consumer-service and connection regulations: dispute, correction and application deadlines coded into agent rules | Sovereign shared deployment for customer operations; grid-adjacent read-only agents on-prem in the company data centre with critical-infrastructure separation kept. | | Government & Public Sector → | Presidential information and communication security guide: deployment in the institution’s data centre, air-gap included; open-weight models inside | On-prem in the institution’s own data centre; air-gapped where required. For municipalities and affiliated bodies, a dedicated tenant in the public cloud. | | Defence & Aerospace → | Defence-industry agency and export-control law: control-list mapping by the agent, licence and sharing decisions with the authorised compliance officer | Air-gapped deployment: models, platform and data inside the closed network; updates as approved packages, remote maintenance only in windows the organisation opens. | | IoT & Connected Devices → | Personal-data law and device data: telemetry inside the organisation, user data purpose-limited | Hybrid: device cloud and time-series data on the existing infrastructure, agents and small models in our data centre in Türkiye or the manufacturer’s own. | | Software Delivery → | Source code and intellectual property: code never leaves the organisation; Code Aginies and completion models on-prem | Code Aginies and coding models in the organisation’s data centre or in our data centre in Türkiye through the closed-circuit deployment; SDLC agents connected to Jira, Git and CI/CD with service accounts. | ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Solutions URL: https://www.aginies.com/en/solutions Language: en Summary: Aginies solutions: customer interaction, operational automation, field service, IoT fleet operations, financial analysis, information management, data protection and export-control compliance, public procurement, HR and software delivery. All on one orchestration layer with one governance model. Solutions # Seven areas, 30 solution packages, one execution layer. From customer interaction to software delivery, every solution runs on the same orchestration layer, the same governance and the same observability. Each links to agents you can watch live in the agent library. Customer interaction & experience Agents that talk to customers and citizens and act in the system while talking: web, mobile, WhatsApp, phone. 10 solutions · 40 agents Retail & E-commerce4 agents ## E-commerce support & action Where is my order, returns, delivery changes: the agent completes the transaction in the order system. - High-volume requests resolved without growing the support team - Order updates, cancellations and delivery changes completed inside the conversation - Product answers grounded in your knowledge sources, no fabrication Read more → Insurance4 agents ## Voice AI for claims intake A voice agent takes the loss notice and opens the claim in the system. - Call-centre load drops and wait times fall - Hundreds of simultaneous conversations at consistent quality - Claims opened at first contact with complete information Read more → Enterprise Functions3 agents ## Call-centre quality monitoring 100% of conversations scored against objective criteria; sampling ends. - 100% evaluation instead of sampling - Consistent, objective quality measurement - Performance trends and improvement areas visible immediately Read more → Banking4 agents ## Conversational banking Complete mobile banking operations by typing or talking. - Banking completed by conversation, on any channel - Lower call-centre load, higher first-contact resolution - Lower operating cost per transaction Read more → Retail & E-commerce4 agents ## Sales & lead management The right offer at the right time: lead scoring, campaign and price recommendations. - Sales focuses on the leads most likely to convert - Personalised campaigns lift conversion and basket size - Contact before the lead cools: every scored lead is called the same day Read more → Healthcare4 agents ## Healthcare operations & clinical guidance Smart triage and patient referral; clinical decisions always stay with clinicians. - Patients routed to the right department and time at first contact - Fewer no-shows, higher capacity utilisation - Clinical teams relieved of administrative load Read more → Telecom4 agents ## Voice analysis & quality 100% of calls transcribed with Turkish STT, scored and coached; models run on-prem. - Measurement across all calls instead of sampling - Compliance breaches and customer risk visible the same day - Voice data never leaves the organisation Read more → Energy & Utilities4 agents ## Utility customer operations Outages, meters, connections and demand response: customer agents that speak with grid data. - Call volume drops in outages, satisfaction rises - Bill disputes and connection applications resolved within hours - The flexibility portfolio turns into revenue Read more → Government & Public Sector4 agents ## Citizen services & file pre-review The request to the right unit, the file complete, the decision with the officer: justified by clause. - Requests reach the right unit first time, with no lost deadline - File pre-review from hours to minutes - Every decision explainable and auditable Read more → Telecom5 agents ## Subscriber care and network support Bill, plan, contract and fault operations on voice and text channels, inside BSS and network systems. - Multi-topic calls closed in one session without a representative - A clear drop in unnecessary technician visits - Shorter handling time, higher satisfaction Read more → Operational automation From signal to work order, from request to delivery: agents that decide and dispatch on top of operational systems. 8 solutions · 32 agents Enterprise Functions4 agents ## Email classification & attachment verification The inbox reaches the right team; missing attachments are caught before work starts. - Requests reach the right team the first time - Instant replies to standard questions, higher satisfaction - Missing documents caught before entering operations Read more → Enterprise Functions4 agents ## Operational automation & demand management Agents that classify, route and answer demands; faster processes, lower cost. - Shorter resolution times, backlog cleared - Lower operating cost per demand - Teams freed from repetitive work to focus on exceptions Read more → IoT & Connected Devices4 agents ## Field service & dispatch automation Sensor alarm or customer report → the right technician, the right part, the right time. - Signal to service order in minutes, with no manual entry - First-visit fix with the right part and the right task card - Customer and quality records current at the same time Read more → Banking4 agents ## Fraud and dispute operations Alert triage, chargeback files, card blocks and early collections on one operations desk. - A large drop in review time per alert - No dispute file past its deadline - Fewer false-positive blocks, fewer customer complaints Read more → Insurance4 agents ## Claims operations end to end From notice to payment: voice intake, photo assessment, repair-shop assignment and subrogation in one flow. - Simple claims close in days - Adjuster capacity reserved for complex files - More subrogation, less leakage Read more → Fintech & Payments4 agents ## Merchant and payment operations Merchant onboarding, POS and settlement support, reconciliation exceptions and transaction monitoring on one desk. - Merchant issues resolved at first contact, on one channel - KYB within hours; the risk team only on exceptions - Reconciliation exceptions no longer wait for month-end Read more → Logistics4 agents ## Logistics control tower Operations agents that catch deviations early, reroute, inform the customer and compute the SLA. - Delays known and managed hours ahead - Higher on-time delivery and better customer information - Transparent, auditable SLA finance Read more → Manufacturing & Automotive4 agents ## Production and maintenance operations From line stoppage to work order, from maintenance request to warranty file: agents that decide on MES, CMMS and ERP. - A measurable drop in unplanned downtime - Minutes from alarm to work order - Days instead of weeks in warranty and invoice processes Read more → Financial analysis & operational intelligence From document to record, from record to report: agents that validate, match and report finance and operations data. 4 solutions · 15 agents Enterprise Functions4 agents ## Invoice & document extraction OCR → validation → ERP: clean records without manual entry. - Key fields extracted without manual entry - Finance cycles shrink from days to minutes - Validated data flows straight into ERP and business systems Read more → Fintech & Payments4 agents ## Financial analysis & operational intelligence Agents that read financial data and documents, interpret them and support decisions. - Reconciliation and invoice cycles drop from days to hours - Decisions taken on current, source-linked data - Exceptions and risk patterns visible before they materialise Read more → Banking4 agents ## Credit and onboarding operations Digital account opening, limit increases and loan documents under one decision logic, without a branch visit. - Accounts opened without a branch visit and credit decisions within minutes - A clear reduction in document review effort - Audit-ready, reasoned and consistent decisions Read more → Insurance3 agents ## Underwriting and renewal operations Quote documents read, risk variables extracted, renewal offers prepared within the rules. - Most renewals automatic, offers on time - Underwriter time spent on exceptional risks - A measurable rise in the renewal rate Read more → Information management & decision support Answers and decision support grounded in the organisation’s own knowledge, with citations; no fabrication. 1 solution · 4 agents Government & Public Sector4 agents ## Information management & decision support Source-linked access to regulations, policies and contracts in seconds. - Access to information drops from hours to seconds - Decisions grounded in current regulation and cited documents - Impact of regulatory change seen proactively Read more → Data protection & compliance Personal data, export control and public procurement: agents that apply the rule and leave the decision with the expert. 3 solutions · 11 agents Enterprise Functions4 agents ## KVKK / GDPR personal data protection Detect personal data in documents and calls, mask it, and keep an audit-ready data map. - Personal data masked before it reaches a model - An audit-ready, current data inventory - Compliance risk in document and call archives made visible Read more → Government & Public Sector3 agents ## Public procurement & tender compliance Bid files compared with the specification; the commission decides on a prepared findings list. - Commission preparation drops from days to hours - Consistent, justified, appeal-resistant evaluation - Every finding linked to its source document, audit-ready Read more → Defence & Aerospace4 agents ## Defence export control & supply-chain security Classification, party screening, licence file and supplier qualification: agents on a closed network that leave the decision with the expert. - The compliance expert spends minutes per file instead of hours - Controlled items are caught before shipment - Supplier risk visible before qualification Read more → HR & employee experience Employee experience from hiring to onboarding; the HR team spends its time on people. 2 solutions · 7 agents Enterprise Functions3 agents ## HR & employee experience From recruitment to onboarding: data-driven, fast and scalable HR processes. - Shorter HR cycle from evaluation to adaptation - New hires ready on day one: access, equipment, training - HR freed from repetitive work to focus on experience Read more → Enterprise Functions4 agents ## Agent outsourcing (agent workforce) Outsourced agents instead of outsourced people: iOS developer, business analyst, accountant, tester; with a work log. - Roles that start within days and work 24/7 - What was done, how long it took and what it cost in one record - Knowledge stays in the workspace; scaling takes minutes Read more → Software development & technology productivity The SDLC agent team, IoT fleet operations, Code Aginies: the whole software and device line. 2 solutions · 9 agents Software Delivery5 agents ## SDLC agent team PM, analyst, architect, developer and tester agents working together in the CI/CD pipeline. - Lead time from request to production drops from days to hours - Standard, measurable quality: coverage, security and regression on every PR - Engineers focus on direction and approval; agents take the repetitive work Read more → IoT & Connected Devices4 agents ## IoT device fleet operations Anomaly clustering, root cause and health-gated OTA rollouts: one incident, one decision per fleet. - Alarm noise collapses into incidents; the team looks at the real problem - Safe, evidenced and reversible firmware rollouts - Low unit cost with small on-prem models Read more → ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Agent outsourcing (agent workforce) URL: https://www.aginies.com/en/solutions/agent-workforce Language: en Summary: The outsourcing model sells person-months: starting takes weeks, knowledge leaves with the person, timesheets do not show what was done. Scaling means new hires. Enterprise Functions · HR & employee experience # Agent outsourcing (agent workforce) Outsourced agents instead of outsourced people: iOS developer, business analyst, accountant, tester; with a work log. In short Agent outsourcing (agent workforce) is an Aginies solution package for the Enterprise Functions sector in the “hr & employee experience” area. What the agents do: The job description becomes an agent spec: tasks, tools, outputs, KPIs, approval points; the agent gets accounts in your tools (Jira, Git, ERP, CRM); code and data stay in your tenant. Expected outcomes: roles that start within days and work 24/7; what was done, how long it took and what it cost in one record; knowledge stays in the workspace; scaling takes minutes. The package uses the Role specs, Work log, Shadow → sampled review, Human-in-the-loop capabilities of the Aginies platform and can be watched live through the iOS feature implementation (hired developer agent), Regression test cycle (hired QA agent), Month-end close and reconciliation (hired accountant agent), Backlog grooming and estimation scenarios in the agent library. The problem The outsourcing model sells person-months: starting takes weeks, knowledge leaves with the person, timesheets do not show what was done. Scaling means new hires. What the agents do - 01The job description becomes an agent spec: tasks, tools, outputs, KPIs, approval points. - 02The agent gets accounts in your tools (Jira, Git, ERP, CRM); code and data stay in your tenant. - 03Every output is reviewed in the shadow period; after the quality threshold, sampled review and threshold-based approval. - 04Every task goes to the work log: output, hour-equivalent, quality, approval, cost; weekly timesheet and monthly invoice from the same record. Package scope - Agents hired from the role catalogue: developer, analyst, accountant, tester, support - Works inside the organisation’s tools; work log with task, output and hour-equivalent - Shadow period, sampled review, autonomous; thresholds and kill-switch - A named team lead on the Aginies side; weekly report What you need to start - Tool access the role needs (Jira, Git, ERP, ticketing) - Task definition and acceptance criteria - A review owner on the organisation side Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors iOS Developer SupervisorQA SupervisorAccounting SupervisorPlanning Supervisor Specialist agents Analysis AgentDevelopment AgentReview AgentTest Plan AgentExecution AgentDefect AgentDocument AgentReconciliation AgentClose AgentProject Manager AgentAnalyst AgentArchitect Agent Models (Hub catalogue) aginies-sunkar-1aginies-talay-2aginies-tulpar-4aginies-konrul-1 Systems touched JiraGitHubXcode CloudFigmaWorkforce Work LogCI/CDDevice / Browser GridJira / XrayERP (SAP / Logo)Bank APIe-Invoice PortalConfluenceSlack / Teams Channels apiappbatchemailweb Simulation volumes 1,040runs / month 94.3%success rate 9.2 smedian duration $0.0191estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Roles that start within days and work 24/7 - What was done, how long it took and what it cost in one record - Knowledge stays in the workspace; scaling takes minutes Capabilities used Role specsWork logShadow → sampled reviewHuman-in-the-loopKill-switchAgent-month / output billing Explore the platform → Measured targets | Accepted output share | 85% and above | | Cost per role | 30–50% of an outsourced person-month | | Knowledge loss at turnover | Zero; knowledge stays in the workspace | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - iOS feature implementation (hired developer agent)Every PR is reviewed by a person; production deployment under the team lead’s approval - Regression test cycle (hired QA agent)The release “go / no-go” decision stays with the release manager - Month-end close and reconciliation (hired accountant agent)Every posted entry and every filing with the accountant’s approval - Backlog grooming and estimationSprint scope → product owner approval; agents only propose, never commit items to a sprint - packageProduction and financial record changes under the organisation’s approval - packageThe organisation owns the outputs; agent-month or output-based billing Rollout - 1 · Policy and role catalogue · 2 weeksHR policies, process owners and system access defined; personal-data boundaries written. - 2 · Pilot · 4 weeksOne department or one role; quality score and user feedback tracked weekly. - 3 · Scale-outAll departments; thresholds, review sampling and kill-switch owned by HR management. Related agents ## This package’s scenarios in the agent library Software DeliveryHITL ### iOS feature implementation (hired developer agent) The Jira story is taken, built in SwiftUI, tested and opened as a PR; review and release decisions stay with the team. 380 runs / month 94.4% success 4 agents Software DeliveryHITL ### Regression test cycle (hired QA agent) Regression runs on every release candidate, new bugs are filed with evidence, a coverage report is written; the “go” decision stays with the release manager. 260 runs / month 95.8% success 4 agents Enterprise FunctionsHITL ### Month-end close and reconciliation (hired accountant agent) Invoices are processed, bank and vendor reconciliations done, the close checklist run; entries are posted with the accountant’s approval. 60 runs / month 96.2% success 4 agents Software DeliveryHITL ### Backlog grooming and estimation Incoming requests are triaged, clarified and sized; a sprint proposal goes to the product owner for approval. 340 runs / month 92.8% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Call-centre quality monitoring URL: https://www.aginies.com/en/solutions/call-center-quality Language: en Summary: Quality teams can listen to 2% of calls. Scoring is subjective, coaching is late, compliance breaches surface weeks later. Enterprise Functions · Customer interaction & experience # Call-centre quality monitoring 100% of conversations scored against objective criteria; sampling ends. In short Call-centre quality monitoring is an Aginies solution package for the Enterprise Functions sector in the “customer interaction & experience” area. What the agents do: All calls are transcribed; sensitive data is masked; each call is scored against the company’s quality form, with rationale and timestamps. Expected outcomes: 100% evaluation instead of sampling; consistent, objective quality measurement; performance trends and improvement areas visible immediately. The package uses the Speech-to-text, Agent decision traceability, Dashboard analytics layer, Latency tracking capabilities of the Aginies platform and can be watched live through the Call centre quality monitoring, Review response and quality escalation, Bill dispute resolution scenarios in the agent library. The problem Quality teams can listen to 2% of calls. Scoring is subjective, coaching is late, compliance breaches surface weeks later. What the agents do - 01All calls are transcribed; sensitive data is masked. - 02Each call is scored against the company’s quality form, with rationale and timestamps. - 03Compliance breaches, repeat complaints and process bottlenecks are flagged automatically. - 04Agent and team dashboards with trends and coaching suggestions. Package scope - Transcription of every call and automatic scoring against the quality form - Compliance breach, impolite language and mandatory-phrase checks - Per-representative coaching note and trend dashboard - Full coverage instead of sampling; the quality team focuses on exceptions What you need to start - Access to call recordings and metadata (switch or recording system) - The existing quality form and scoring rules - CRM or switch data that maps representatives to calls Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Quality SupervisorVoice-of-Customer SupervisorBilling Supervisor Specialist agents Transcription AgentScoring AgentCoaching AgentClassification AgentResponse AgentQuality AgentIntent AgentBill Analysis AgentAdjustment Agent Models (Hub catalogue) aginies-voice-tr-v3aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Systems touched Genesys SIPQM Scorecard / Case ManagementQA Rubric Vector StoreWorkdayReview Platform APIShopify / OMSJira (QA)Billing (Amdocs)Usage Mediation / CDR StoreSMS Gateway Channels voicebatchwebapiappwhatsapp Simulation volumes 204,000runs / month 97.5%success rate 6 smedian duration $0.0012estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - 100% evaluation instead of sampling - Consistent, objective quality measurement - Performance trends and improvement areas visible immediately Capabilities used Speech-to-textAgent decision traceabilityDashboard analytics layerLatency trackingSensitive data maskingRouting accuracy measurement Explore the platform → Measured targets | Calls evaluated | From a 3–5% sample to 100% coverage | | Quality evaluation effort | 80% lower | | Time to detect a compliance breach | From weeks to same day | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Call centre quality monitoringScore < 60 or compliance breach → team leader review - Review response and quality escalation1-star review replies and defect escalations → community manager approval - Bill dispute resolutionGoodwill credit > 75 → team lead approval - packageA quality specialist validates a sample before scores reach representatives - packageRecordings processed in Türkiye; personal data masked in transcripts Rollout - 1 · Discovery and design · 2–3 weeksThe 20 most frequent intents and transactions are chosen, identity and approval thresholds written with information security, channel and system integrations listed. - 2 · Pilot · 4–6 weeksOne channel, a limited customer group; the test corpus runs, a human representative shadows, thresholds tuned on real traffic. - 3 · Scale-outRemaining channels and transactions added; dashboards, SIEM forwarding and kill-switch handed over; monthly model and rule review begins. Related agents ## This package’s scenarios in the agent library Enterprise FunctionsHITL ### Call centre quality monitoring Every call scored against the rubric, not a sample; compliance breaches and coaching flags ready by morning. 90,000 runs / month 98.8% success 4 agents Retail & E-commerceHITL ### Review response and quality escalation Reviews classified and answered in brand tone; recurring defects joined with return data and escalated to QA. 26,000 runs / month 97.2% success 4 agents TelecomHITL ### Bill dispute resolution Invoice lines and usage explained in seconds; goodwill credit applied by policy, larger amounts need a lead. 88,000 runs / month 96.3% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Citizen services & file pre-review URL: https://www.aginies.com/en/solutions/citizen-services Language: en Summary: Petitions reach the wrong unit, deadlines run, and missing-document checks in permit and assistance files take most of officers’ time. Black-box tools cannot be used in the public sector. Government & Public Sector · Customer interaction & experience # Citizen services & file pre-review The request to the right unit, the file complete, the decision with the officer: justified by clause. In short Citizen services & file pre-review is an Aginies solution package for the Government & Public Sector sector in the “customer interaction & experience” area. What the agents do: The request is classified, the statutory deadline and responsible unit found, the case opened; documents are read, verified against agency records, gaps reported to the citizen in plain language. Expected outcomes: requests reach the right unit first time, with no lost deadline; file pre-review from hours to minutes; every decision explainable and auditable. The package uses the Intent detection, Vision extraction, Multi-source RAG, Explainable scoring capabilities of the Aginies platform and can be watched live through the Citizen services assistant: application, documents, appointment and status, Citizen request routing, Permit application check, Social assistance eligibility pre-assessment scenarios in the agent library. The problem Petitions reach the wrong unit, deadlines run, and missing-document checks in permit and assistance files take most of officers’ time. Black-box tools cannot be used in the public sector. What the agents do - 01The request is classified, the statutory deadline and responsible unit found, the case opened. - 02Documents are read, verified against agency records, gaps reported to the citizen in plain language. - 03Eligibility criteria are scored; every finding is justified with a clause number and source page. - 04The administrative decision stays with the officer; sensitive fields are masked on screen, a trace kept. Package scope - Application intake, document reading and completion; e-Government integration - Eligibility assessment with reasoning; decision pack for the officer - Appointment, status tracking and notices; WhatsApp and call centre - Petition routing and deadline tracking What you need to start - API access to document management and the case system - Authorisation and integration for e-Government services - Criteria, document and deadline definitions per service Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Citizen Services SupervisorRequest SupervisorPermit SupervisorSocial Assistance Supervisor Specialist agents Dialogue AgentEligibility AgentDocument AgentCase AgentClassification AgentSLA AgentRegulation AgentFile AgentVerification AgentCitizen Comms Agent Models (Hub catalogue) aginies-talay-2aginies-tulpar-4aginies-konrul-1aginies-sunkar-1 Systems touched WhatsApp Businesse-Government GatewayCase ManagementDocument StoreAppointment Systeme-Devlet GatewayLegislation Vector StoreSMS GatewayGIS / Zoning PlanRegulation Vector StoreExchange / M365 MailSocial Assistance Information SystemCivil Registry / Income ServicesEligibility Criteria (vector) Channels whatsappvoicewebappemailbatch Simulation volumes 308,100runs / month 94.8%success rate 7 smedian duration $0.002estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Requests reach the right unit first time, with no lost deadline - File pre-review from hours to minutes - Every decision explainable and auditable Capabilities used Intent detectionVision extractionMulti-source RAGExplainable scoringPII maskingAir-gap deployment Explore the platform → Measured targets | Documents asked of the citizen again | 70% fewer | | Application pre-review time | From days to minutes | | Correct first routing | 95% and above | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Citizen services assistant: application, documents, appointment and statusBorderline eligibility, inconsistent document or another application in the household → officer review - Citizen request routingClassification confidence < 0.85 or staff complaint → registry officer approval - Permit application checkEvery permit decision is taken by the licensing officer; the system only pre-reviews - Social assistance eligibility pre-assessmentThe assistance decision is always with the case officer; the agent pre-assesses - packageThe decision always sits with the officer; the agent prepares the eligibility pack - packageData in the public cloud or the institution’s data centre; access by authorisation Rollout - 1 · Discovery and design · 2–3 weeksThe 20 most frequent intents and transactions are chosen, identity and approval thresholds written with information security, channel and system integrations listed. - 2 · Pilot · 4–6 weeksOne channel, a limited customer group; the test corpus runs, a human representative shadows, thresholds tuned on real traffic. - 3 · Scale-outRemaining channels and transactions added; dashboards, SIEM forwarding and kill-switch handed over; monthly model and rule review begins. Related agents ## This package’s scenarios in the agent library Government & Public SectorHITL ### Citizen services assistant: application, documents, appointment and status The citizen writes or calls; eligibility checked, documents read, application opened, appointment given, every step traceable. 230,000 runs / month 93.8% success 5 agents Government & Public SectorHITL ### Citizen request routing e-Government requests classified in seconds, given a legal deadline and routed to the right unit with an SLA. 54,000 runs / month 98.1% success 4 agents Government & Public SectorHITL ### Permit application check Missing documents and regulatory fit reported in minutes; the licensing officer always makes the decision. 3,100 runs / month 95.8% success 4 agents Government & Public SectorHITL ### Social assistance eligibility pre-assessment The application is verified against agency records, eligibility scored, gaps reported to the citizen; the decision belongs to the case officer. 21,000 runs / month 96.6% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Claims operations end to end URL: https://www.aginies.com/en/solutions/claims-operations-end-to-end Language: en Summary: A claim file travels across five teams and four systems; the customer calls at every step, adjusters visit simple files too, subrogation opportunities are missed. Insurance · Operational automation # Claims operations end to end From notice to payment: voice intake, photo assessment, repair-shop assignment and subrogation in one flow. In short Claims operations end to end is an Aginies solution package for the Insurance sector in the “operational automation” area. What the agents do: Notice taken by voice or app; the policy matched and the incident form completed in one conversation; parts and labour estimated from photos; fraud screening via prior-damage and metadata checks. Expected outcomes: simple claims close in days; adjuster capacity reserved for complex files; more subrogation, less leakage. The package uses the SIP voice, Vision damage assessment, Policy & claims tools (MCP), Human-in-the-loop capabilities of the Aginies platform and can be watched live through the End-to-end motor claim: from notice to payment, Voice claims intake (FNOL), Motor claim photo assessment, Subrogation recovery scenarios in the agent library. The problem A claim file travels across five teams and four systems; the customer calls at every step, adjusters visit simple files too, subrogation opportunities are missed. What the agents do - 01Notice taken by voice or app; the policy matched and the incident form completed in one conversation. - 02Parts and labour estimated from photos; fraud screening via prior-damage and metadata checks. - 03Cover, deductible and fault evaluated; a network repair shop assigned and the payment order prepared. - 04Subrogation file opened automatically; above-threshold files land on the adjuster’s approval card. Package scope - Notice by voice and app; policy matching and incident form - Photo assessment, cover and fault evaluation - Network repair-shop assignment, payment order and subrogation file - Customer updates and file tracking; adjuster approval card What you need to start - API access to policy, claims and payment systems - Repair network and parts price list - Thresholds and adjuster rules per claim type Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Claims SupervisorSubrogation Supervisor Specialist agents Intake AgentPhoto Assessment AgentAdjuster AgentCommunication AgentSpeech AgentPolicy AgentClaims Registration AgentVision AgentEstimate AgentRepair Routing AgentLiability Scan AgentEvidence AgentRecovery Agent Models (Hub catalogue) aginies-voice-tr-v3aginies-konrul-1aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Systems touched Call Centre SIPPolicy AdminClaims SystemRepair NetworkPayment HubGenesys SIPPolicy Admin SystemGuidewire ClaimCenterWhatsApp Business APIParts & Labour Catalogue (Audatex)Repair Network PortalDocument Store (S3)Insurance Registry (SBM)Email (Exchange) Channels voiceappwhatsappbatchemailapi Simulation volumes 73,900runs / month 94.0%success rate 8.1 smedian duration $0.0045estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Simple claims close in days - Adjuster capacity reserved for complex files - More subrogation, less leakage Capabilities used SIP voiceVision damage assessmentPolicy & claims tools (MCP)Human-in-the-loopFull audit trailPayment hub integration Explore the platform → Measured targets | File closure time (simple claims) | From weeks to days | | Files needing an adjuster | 40% fewer | | Files with subrogation started | +25% | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - End-to-end motor claim: from notice to paymentEstimate > 150,000, unclear fault share or overlap with prior damage → adjuster approval - Voice claims intake (FNOL)Injury or reserve > 250,000 → warm transfer to a claims specialist - Motor claim photo assessmentEstimate > 15,000 or model confidence < 80% → adjuster approval - Subrogation recoveryRecovery demand > 50,000 → subrogation counsel approval - packageAbove-threshold amounts, unclear fault and overlap with prior damage under adjuster approval - packagePhoto metadata and decision reasoning on file; fraud check on every file Rollout - 1 · Process map · 2 weeksSignal sources, decision points and work-order systems mapped; the automatic-decision boundary and approval chain defined. - 2 · Pilot · 4 weeksOne line, one region or one process; the agent prepares the recommendation, a person applies it; deviation and gain measured. - 3 · Autonomous operationDecisions below the threshold run automatically, above it go to approval; shift reports and SLA dashboards live. Related agents ## This package’s scenarios in the agent library InsuranceHITL ### End-to-end motor claim: from notice to payment Voice notice, photo assessment, repair-shop assignment and payment order in one flow; the adjuster steps in only above the threshold. 21,000 runs / month 91.2% success 5 agents InsuranceHITL ### Voice claims intake (FNOL) The customer calls and describes it; policy found by plate, cover checked, claim opened and adjuster assigned. 34,000 runs / month 95.8% success 5 agents InsuranceHITL ### Motor claim photo assessment Damage detected from photos, estimate built from parts and labour, adjuster only approves; repair slot booked. 15,500 runs / month 93.3% success 4 agents InsuranceHITL ### Subrogation recovery Closed claims scanned, the at-fault party's insurer identified, and the demand sent after counsel approval. 3,400 runs / month 95.5% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Conversational banking URL: https://www.aginies.com/en/solutions/conversational-banking Language: en Summary: Customers get lost in menus; the call centre fills with simple transactions. Existing assistants inform but cannot transact; those that can are not auditable. Banking · Customer interaction & experience # Conversational banking Complete mobile banking operations by typing or talking. In short Conversational banking is an Aginies solution package for the Banking sector in the “customer interaction & experience” area. What the agents do: Integrated with the existing identity flow; unauthorised transaction requests stop at the guardrail; limit increases, transfers, standing orders and statements completed in one dialogue. Expected outcomes: banking completed by conversation, on any channel; lower call-centre load, higher first-contact resolution; lower operating cost per transaction. The package uses the Identity & authorisation guardrails, Supervisor · sub-agents, Human-in-the-loop, Core banking tools (MCP) capabilities of the Aginies platform and can be watched live through the Dialogue banking: voice and text transaction assistant, Credit card limit increase, Voice banking money transfer, Card fraud alert triage scenarios in the agent library. The problem Customers get lost in menus; the call centre fills with simple transactions. Existing assistants inform but cannot transact; those that can are not auditable. What the agents do - 01Integrated with the existing identity flow; unauthorised transaction requests stop at the guardrail. - 02Limit increases, transfers, standing orders and statements completed in one dialogue. - 03Transactions above an amount or risk threshold route to human approval; every decision leaves a trace. - 04The voice channel connects to the call centre via SIP; the same workflow runs in text and voice. Package scope - The same transaction agent on mobile, web, WhatsApp and voice - Transfer, bill, limit, card and dispute operations with core-banking tools - Per-transaction identity and OTP thresholds; fraud score on every transaction - Warm SIP hand-off to the call centre with conversation summary and references What you need to start - Transaction APIs or MCP servers for core banking and card systems - Existing digital identity and OTP infrastructure - Transaction thresholds and approval rules written with information security Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Dialogue Banking SupervisorCard SupervisorVoice Transaction SupervisorFraud Supervisor Specialist agents Dialogue AgentIdentity AgentTransaction AgentRisk AgentIntent AgentOperations AgentSpeech AgentTransfer AgentTriage AgentOutreach AgentCard Agent Models (Hub catalogue) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-voice-tr-v3 Systems touched Mobile App ChatCore BankingCard ManagementFraud EngineSMS GatewayKYC / Customer 360Push & SMS GatewayGenesys SIPWhatsApp Business APISalesforce CRM Channels appvoicewhatsappsmsapi Simulation volumes 296,400runs / month 95.2%success rate 5.9 smedian duration $0.0031estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Banking completed by conversation, on any channel - Lower call-centre load, higher first-contact resolution - Lower operating cost per transaction Capabilities used Identity & authorisation guardrailsSupervisor · sub-agentsHuman-in-the-loopCore banking tools (MCP)Full audit trailSIP voice Explore the platform → Measured targets | Transactions completed by conversation | 55–70% | | Simple transactions reaching the call centre | 40% fewer | | Operating cost per transaction | 50% or more lower | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Dialogue banking: voice and text transaction assistantLimit increase > 60%, dispute amount > 5,000 or transfer to a new payee → operations approval - Credit card limit increaseLimit > 100,000 → branch manager approval - Voice banking money transferAmount > 50,000 or new payee → warm transfer to a call-centre agent - Card fraud alert triageConfirmed fraud loss > 2,500 → analyst approval before permanent cancellation - packageNew payee, high amount and dispute always require extra verification or human approval - packageDeployment aligned with banking IT regulation; all systems in Türkiye Rollout - 1 · Discovery and design · 2–3 weeksThe 20 most frequent intents and transactions are chosen, identity and approval thresholds written with information security, channel and system integrations listed. - 2 · Pilot · 4–6 weeksOne channel, a limited customer group; the test corpus runs, a human representative shadows, thresholds tuned on real traffic. - 3 · Scale-outRemaining channels and transactions added; dashboards, SIEM forwarding and kill-switch handed over; monthly model and rule review begins. Related agents ## This package’s scenarios in the agent library BankingHITL ### Dialogue banking: voice and text transaction assistant The customer talks or types; bill, limit, dispute and transfer are completed in core banking in one session. 186,000 runs / month 94.7% success 5 agents BankingHITL ### Credit card limit increase The customer types in the app or talks to the call centre; the decision takes seconds, and anything above threshold goes to a manager. 18,400 runs / month 96.4% success 4 agents BankingHITL ### Voice banking money transfer Customer calls and talks; voiceprint, security question and OTP verify them; transfer completes agent-free. 27,000 runs / month 93.1% success 5 agents BankingHITL ### Card fraud alert triage The card is blocked in seconds and the customer asked on WhatsApp; the reply releases or cancels the card. 65,000 runs / month 97.1% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Credit and onboarding operations URL: https://www.aginies.com/en/solutions/credit-and-onboarding-operations Language: en Summary: Account opening and loan applications are handled by different teams with different tools and rules. Document review takes days; the customer repeats the same information and the decision reasoning is scattered. Banking · Financial analysis & operational intelligence # Credit and onboarding operations Digital account opening, limit increases and loan documents under one decision logic, without a branch visit. In short Credit and onboarding operations is an Aginies solution package for the Banking sector in the “financial analysis & operational intelligence” area. What the agents do: Identity, liveness, sanctions and address verification in one flow; clean applications open within minutes; income, score and policy rules define the auto-approval band; out-of-band applications go to an analyst. Expected outcomes: accounts opened without a branch visit and credit decisions within minutes; A clear reduction in document review effort; audit-ready, reasoned and consistent decisions. The package uses the Identity & KYC tools, Policy engine, Document extraction (Vision), Human-in-the-loop capabilities of the Aginies platform and can be watched live through the Digital account opening (KYC), Credit card limit increase, Mortgage document review, Credit card debt payment scenarios in the agent library. The problem Account opening and loan applications are handled by different teams with different tools and rules. Document review takes days; the customer repeats the same information and the decision reasoning is scattered. What the agents do - 01Identity, liveness, sanctions and address verification in one flow; clean applications open within minutes. - 02Income, score and policy rules define the auto-approval band; out-of-band applications go to an analyst. - 03Payslips, title deeds and appraisals are read, fields cross-checked, missing items requested from the customer. - 04Branch, mobile and call centre use the same decision engine; every decision is stored with its reasoning. Package scope - Digital onboarding: identity, liveness, sanctions and address verification in one flow - Limit increase, loan application and mortgage document review - Income and score retrieval; policy engine; auto-approval band - The same decision logic in branch, mobile and call-centre channels What you need to start - API access to core banking, credit bureau and identity services - Credit policy and approval matrix - Sample set per document type (payslip, title deed, appraisal) Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Onboarding SupervisorCard SupervisorLoan File Supervisor Specialist agents Document AgentLiveness AgentScreening AgentAccount AgentIntent AgentRisk AgentOperations AgentExtraction AgentIncome AgentValuation AgentPayment AgentVoice Agent Models (Hub catalogue) aginies-konrul-1aginies-sunkar-1aginies-tulpar-4aginies-talay-2aginies-voice-tr-v3 Systems touched KYC / Customer 360Sanctions Screening (Dow Jones)Core BankingDocument Store (S3)SMS GatewayPush & SMS GatewayLoan Origination SystemCredit Bureau (KKB)Email (Exchange)Card ManagementKnowledge Base · kart_sss_v3OTP / SMS Gateway Channels appwebvoiceemailbranch Simulation volumes 118,600runs / month 96.0%success rate 7.8 smedian duration $0.0035estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Accounts opened without a branch visit and credit decisions within minutes - A clear reduction in document review effort - Audit-ready, reasoned and consistent decisions Capabilities used Identity & KYC toolsPolicy engineDocument extraction (Vision)Human-in-the-loopFull audit trailCore banking tools (MCP) Explore the platform → Measured targets | Accounts opened without a branch visit | 70–85% | | Credit decision time (auto band) | Minutes | | Document review effort | 60% lower | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Digital account opening (KYC)Sanctions/PEP fuzzy match score > 0.65 → compliance analyst approval - Credit card limit increaseLimit > 100,000 → branch manager approval - Mortgage document reviewLoan > 3,000,000 or DTI > 45% → underwriting officer approval - Credit card debt paymentAmount > 100,000 → branch manager approval - packageOut-of-band applications, high amounts and suspicious documents always with a credit analyst - packageDecision reasoning and data used stored per application; aligned with banking regulation Rollout - 1 · Documents and rule set · 2 weeksDocument types, field dictionary, matching and tolerance rules written with the finance team; a sample set labelled. - 2 · Parallel run · 4 weeksAgent and current process handle the same documents; the difference report reviewed weekly, cut-over when the accuracy threshold is met. - 3 · Close integrationTied to the month-end calendar; exception queue, audit trail and reconciliation reports opened to internal audit. Related agents ## This package’s scenarios in the agent library BankingHITL ### Digital account opening (KYC) ID, liveness and sanctions checks in one flow; clean applications open an account in minutes, no branch visit. 42,000 runs / month 93.8% success 5 agents BankingHITL ### Credit card limit increase The customer types in the app or talks to the call centre; the decision takes seconds, and anything above threshold goes to a manager. 18,400 runs / month 96.4% success 4 agents BankingHITL ### Mortgage document review Payslips, social security records and appraisal read automatically; income, LTV and missing documents checked. 6,200 runs / month 94.7% success 4 agents BankingHITL ### Credit card debt payment The customer types or talks; debt is fetched, an account chosen, OTP verified and the payment made in the core. 52,000 runs / month 97.8% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Defence export control & supply-chain security URL: https://www.aginies.com/en/solutions/defence-export-control-compliance Language: en Summary: Control-list mapping, party screening and licence files are manual per export item; supplier questionnaires and documents are read after qualification. No external model can be used in a classified environment. Defence & Aerospace · Data protection & compliance # Defence export control & supply-chain security Classification, party screening, licence file and supplier qualification: agents on a closed network that leave the decision with the expert. In short Defence export control & supply-chain security is an Aginies solution package for the Defence & Aerospace sector in the “data protection & compliance” area. What the agents do: Part parameters are mapped to control lists; the classification rationale is written with source clauses; buyer, intermediary and end user are screened against sanctions lists; ownership resolved, end-use statement verified. Expected outcomes: the compliance expert spends minutes per file instead of hours; controlled items are caught before shipment; supplier risk visible before qualification. The package uses the Air-gap deployment, Control-list RAG, Party screening tools, Vision extraction capabilities of the Aginies platform and can be watched live through the Technical publication assistant: maintenance manual, parts, procedure and revision control, Export control and end-user screening, Supplier security and compliance assessment, Platform maintenance (MRO) work order preparation scenarios in the agent library. The problem Control-list mapping, party screening and licence files are manual per export item; supplier questionnaires and documents are read after qualification. No external model can be used in a classified environment. What the agents do - 01Part parameters are mapped to control lists; the classification rationale is written with source clauses. - 02Buyer, intermediary and end user are screened against sanctions lists; ownership resolved, end-use statement verified. - 03Licence application draft and annex list are prepared; the order is blocked until the licence arrives. - 04Supplier questionnaires are scored against the requirement set; gaps and risk class presented to the security manager. Package scope - Control-list mapping and party screening; export licence file - Supplier security assessment and questionnaire - Technical data access: clearance and marking checks - Operation on a closed network in the organisation’s data centre What you need to start - Control lists and the organisation’s classification scheme - Supplier and part records - GPU infrastructure for the closed-network deployment Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Technical Publication SupervisorExport Compliance SupervisorSupplier Security SupervisorMaintenance Supervisor Specialist agents Clearance AgentPublication AgentProcedure AgentParts AgentClassification AgentParty Screening AgentLicence AgentDocument AgentAssessment AgentCase AgentFault AgentTech Pub AgentPlanning Agent Models (Hub catalogue) aginies-sunkar-1aginies-tulpar-4aginies-talay-2aginies-konrul-1 Systems touched Hangar Tablet AppAccess ControlTech Pub Repository (S1000D)MRO SystemAudit LogERP (SAP GTS)Control & Sanctions Lists (vector)Trade Compliance Case SystemE-mailSupplier Management (SRM)Security Requirements (vector)Document StoreMaintenance Management SystemTechnical Publications (S1000D vector)Spares InventoryTechnician App Channels appapiemailbatchweb Simulation volumes 17,400runs / month 95.7%success rate 7.6 smedian duration $0.0053estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - The compliance expert spends minutes per file instead of hours - Controlled items are caught before shipment - Supplier risk visible before qualification Capabilities used Air-gap deploymentControl-list RAGParty screening toolsVision extractionHuman-in-the-loopFull audit trail Explore the platform → Measured targets | Party screening time | From days to minutes | | Missing documents in licence files | 80% fewer | | Data leaving the network | Zero | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Technical publication assistant: maintenance manual, parts, procedure and revision controlProposed deviation from the procedure → authorised engineer approval; export-controlled content → clearance required - Export control and end-user screeningEvery controlled item → export compliance officer approval; the agent cannot make licence decisions - Supplier security and compliance assessmentThe qualification decision always sits with the facility security manager; the agent only prepares - Platform maintenance (MRO) work order preparationEvery work package affecting airworthiness → maintenance planner approval - packageExport and access decisions always with the authorised compliance officer - packageEvery access in the audit log with user, document and timestamp Rollout - 1 · Rule and control design · 2–3 weeksRegulatory articles translated into the control set, the decision matrix approved with compliance, trace requirements written. - 2 · Shadow period · 4 weeksThe agent produces findings and recommendations, the expert decides; disagreement rate and reasoning quality measured. - 3 · Audit-ready operationControl evidence collected automatically; audit reports and the exception register opened to internal audit. Related agents ## This package’s scenarios in the agent library Defence & AerospaceHITL ### Technical publication assistant: maintenance manual, parts, procedure and revision control The technician asks in the hangar; the right-revision procedure, part number and safety note arrive with clearance checks, every access traced. 9,800 runs / month 95.7% success 5 agents Defence & AerospaceHITL ### Export control and end-user screening Part, buyer and end use are classified in one flow; licence need is detected, the decision always stays with the compliance officer. 2,600 runs / month 96.1% success 4 agents Defence & AerospaceHITL ### Supplier security and compliance assessment Supplier questionnaires and documents are read, compared against facility security requirements, gaps listed; the qualification decision belongs to the security manager. 900 runs / month 95.2% success 4 agents Defence & AerospaceHITL ### Platform maintenance (MRO) work order preparation The fault report is matched to technical publications, task card and parts list are produced, the work order opens with planner approval. 4,100 runs / month 95.7% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # E-commerce support & action URL: https://www.aginies.com/en/solutions/e-commerce-support Language: en Summary: Support volume multiplies in season; teams answer the same ten questions while switching screens to act in the order system. Chatbots answer, but cannot cancel. Retail & E-commerce · Customer interaction & experience # E-commerce support & action Where is my order, returns, delivery changes: the agent completes the transaction in the order system. In short E-commerce support & action is an Aginies solution package for the Retail & E-commerce sector in the “customer interaction & experience” area. What the agents do: Intent and entity extraction: order number, product and request type understood in one turn; catalogue and policy knowledge retrieved with RAG; sizes, compatibility and return terms answered correctly. Expected outcomes: high-volume requests resolved without growing the support team; order updates, cancellations and delivery changes completed inside the conversation; product answers grounded in your knowledge sources, no fabrication. The package uses the Multi-source RAG, Supervisor · sub-agents, Session memory, REST / MCP tools capabilities of the Aginies platform and can be watched live through the Outfit stylist: a complete look by size, stock, budget and style, Where is my order? (WISMO), Returns and refunds, Product Q&A assistant scenarios in the agent library. The problem Support volume multiplies in season; teams answer the same ten questions while switching screens to act in the order system. Chatbots answer, but cannot cancel. What the agents do - 01Intent and entity extraction: order number, product and request type understood in one turn. - 02Catalogue and policy knowledge retrieved with RAG; sizes, compatibility and return terms answered correctly. - 03Tools wired to the order system change delivery details, start cancellations and issue return labels. - 04Out-of-policy requests (high-value refunds, campaign exceptions) route to human approval. Package scope - One dialogue agent on web, mobile and WhatsApp; order, return, delivery and product questions - Transaction tools wired to order management, carrier and payment systems - Catalogue and policy knowledge base; return and warranty terms with citations - Warm hand-off to a representative with a conversation summary What you need to start - Read/write API or MCP access to order and carrier systems - Current return, delivery and campaign policies - A session or OTP flow that verifies the customer Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Style SupervisorOrder SupervisorReturns SupervisorShopping Supervisor Specialist agents Vision AgentStylist AgentCatalogue AgentCart AgentIntent AgentTracking AgentReply AgentIntake AgentPolicy AgentFulfilment Agent Models (Hub catalogue) aginies-konrul-1aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Systems touched WhatsApp BusinessProduct Catalogue (PIM)InventoryCustomer ProfileCart & Checkout APIShopify / OMSCarrier API (Yurtiçi, DHL)WhatsApp Business APIZendeskPayment Gateway (iyzico)Email (Exchange)Vector StorePIM / CatalogueWeb Chat Widget Channels appwebwhatsappvoiceemail Simulation volumes 420,000runs / month 97.4%success rate 5.9 smedian duration $0.0009estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - High-volume requests resolved without growing the support team - Order updates, cancellations and delivery changes completed inside the conversation - Product answers grounded in your knowledge sources, no fabrication Capabilities used Multi-source RAGSupervisor · sub-agentsSession memoryREST / MCP toolsGuardrailsIntent detectionEntity extraction Explore the platform → Measured targets | First-contact resolution | +15–25 points | | Share of conversations handed to a representative | below 35% | | Extra representatives at seasonal peak | 40–60% lower | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Outfit stylist: a complete look by size, stock, budget and styleDiscount outside campaign rules or a bespoke price request → customer service - Returns and refundsRefund > 2,500 or > 5 returns in 12 months → team lead approval - packageNo refund or address change without identity verification - packageEvery transaction in the audit trail with customer, order and agent step Rollout - 1 · Discovery and design · 2–3 weeksThe 20 most frequent intents and transactions are chosen, identity and approval thresholds written with information security, channel and system integrations listed. - 2 · Pilot · 4–6 weeksOne channel, a limited customer group; the test corpus runs, a human representative shadows, thresholds tuned on real traffic. - 3 · Scale-outRemaining channels and transactions added; dashboards, SIEM forwarding and kill-switch handed over; monthly model and rule review begins. Related agents ## This package’s scenarios in the agent library Retail & E-commerceHITL ### Outfit stylist: a complete look by size, stock, budget and style The customer sends a photo and a sentence; the agent builds three looks that are in stock, fit and within budget, and adds the chosen one to the cart. 142,000 runs / month 96.1% success 5 agents Retail & E-commerce ### Where is my order? (WISMO) Customer asks, order and carrier events join instantly; on delay a new ETA is given and a trace opened. 110,000 runs / month 98.1% success 4 agents Retail & E-commerceHITL ### Returns and refunds Policy and history checked in seconds, label issued; the refund executes when the carrier scans the parcel. 48,000 runs / month 96.7% success 4 agents Retail & E-commerce ### Product Q&A assistant Size, compatibility and spec questions answered from catalogue and return data; stock and price verified. 120,000 runs / month 98.7% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Email classification & attachment verification URL: https://www.aginies.com/en/solutions/email-classification Language: en Summary: Shared inboxes receive thousands of emails a day. Misrouted requests are delayed; requests with missing attachments enter operations and bounce back. Enterprise Functions · Operational automation # Email classification & attachment verification The inbox reaches the right team; missing attachments are caught before work starts. In short Email classification & attachment verification is an Aginies solution package for the Enterprise Functions sector in the “operational automation” area. What the agents do: Email intent and urgency are classified and assigned to the right queue; attachments are checked by document type: missing, wrong or unreadable ones are flagged. Expected outcomes: requests reach the right team the first time; instant replies to standard questions, higher satisfaction; missing documents caught before entering operations. The package uses the Document embedding pipeline, Dynamic agent routing, Entity extraction, Email service integrations capabilities of the Aginies platform and can be watched live through the Email classification and attachment check, Accounts payable inbox, Citizen request routing, IT helpdesk L1 scenarios in the agent library. The problem Shared inboxes receive thousands of emails a day. Misrouted requests are delayed; requests with missing attachments enter operations and bounce back. What the agents do - 01Email intent and urgency are classified and assigned to the right queue. - 02Attachments are checked by document type: missing, wrong or unreadable ones are flagged. - 03Standard requests get instant answers from the corporate knowledge base. - 04Every routing decision and reply is logged with an audit trail. Package scope - Classification of shared mailboxes and routing to the right unit - Attachment checks: document type, legibility, missing attachment, malicious content - Draft reply and ticket creation; SLA timer - Personal-data masking and retention rules What you need to start - Mailbox access for Microsoft 365 or Google Workspace - API access to the ticketing system (ServiceNow, Jira SM) - Unit and topic taxonomy, SLA definitions Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Mailroom SupervisorAP SupervisorRequest SupervisorHelpdesk Supervisor Specialist agents Classification AgentAttachment AgentRouting AgentCapture AgentCoding AgentPosting AgentSLA AgentCase AgentIntent AgentAccess AgentTicket Agent Models (Hub catalogue) aginies-sunkar-1aginies-konrul-1aginies-talay-2aginies-tulpar-4 Systems touched Exchange / M365 MailMailbox Rules & Checklists Vector StoreServiceNowDocument Store (S3)Oracle EBSVendor Master / Procuremente-Devlet GatewayCase ManagementLegislation Vector StoreSMS GatewayActive Directory / Entra IDMicrosoft Teams Channels emailbatchwebappvoice Simulation volumes 215,000runs / month 97.8%success rate 5.7 smedian duration $0.001estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Requests reach the right team the first time - Instant replies to standard questions, higher satisfaction - Missing documents caught before entering operations Capabilities used Document embedding pipelineDynamic agent routingEntity extractionEmail service integrationsAgent templatingAudit trail logging Explore the platform → Measured targets | Correct first routing | 95% and above | | First response time | From hours to minutes | | Manual triage effort | 70% lower | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Accounts payable inboxNon-PO invoice > 25,000 TL → cost centre owner approval - Citizen request routingClassification confidence < 0.85 or staff complaint → registry officer approval - IT helpdesk L1Access to a finance or production group → group owner approval - packageLegal and complaint e-mails are never answered automatically - packageClassification decision and reasoning stored on the ticket Rollout - 1 · Process map · 2 weeksSignal sources, decision points and work-order systems mapped; the automatic-decision boundary and approval chain defined. - 2 · Pilot · 4 weeksOne line, one region or one process; the agent prepares the recommendation, a person applies it; deviation and gain measured. - 3 · Autonomous operationDecisions below the threshold run automatically, above it go to approval; shift reports and SLA dashboards live. Related agents ## This package’s scenarios in the agent library Enterprise Functions ### Email classification and attachment check Every e-mail in the shared mailbox is classified, attachments checked against a list, gaps asked for at once. 86,000 runs / month 97.6% success 4 agents Enterprise FunctionsHITL ### Accounts payable inbox Every invoice in the ap@ mailbox is captured, coded and posted to the ERP; only exceptions reach an analyst. 28,000 runs / month 96.8% success 4 agents Government & Public SectorHITL ### Citizen request routing e-Government requests classified in seconds, given a legal deadline and routed to the right unit with an SLA. 54,000 runs / month 98.1% success 4 agents Enterprise FunctionsHITL ### IT helpdesk L1 Password resets and access requests solved in chat; access to finance groups asks the owner via a Teams card. 47,000 runs / month 98.5% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Field service & dispatch automation URL: https://www.aginies.com/en/solutions/field-service-dispatch Language: en Summary: Fault and excursion signals exist, but service orders, parts reservation, technician assignment and customer notices are done by hand on separate screens. A critical event waits in the same queue as routine requests. IoT & Connected Devices · Operational automation # Field service & dispatch automation Sensor alarm or customer report → the right technician, the right part, the right time. In short Field service & dispatch automation is an Aginies solution package for the IoT & Connected Devices sector in the “operational automation” area. What the agents do: The signal is validated and prioritised: sensor series, device history and known issues read together; task card, parts list and tooling are derived from technical publications and stock. Expected outcomes: signal to service order in minutes, with no manual entry; first-visit fix with the right part and the right task card; customer and quality records current at the same time. The package uses the Event ingestion, Time-series tools, Multi-source RAG, REST / MCP tools capabilities of the Aginies platform and can be watched live through the Control tower: delay prediction, rerouting, customer notice, SLA, Cold-chain excursion response, Maintenance ticket triage, Platform maintenance (MRO) work order preparation scenarios in the agent library. The problem Fault and excursion signals exist, but service orders, parts reservation, technician assignment and customer notices are done by hand on separate screens. A critical event waits in the same queue as routine requests. What the agents do - 01The signal is validated and prioritised: sensor series, device history and known issues read together. - 02Task card, parts list and tooling are derived from technical publications and stock. - 03The nearest qualified technician is assigned, the service order opened, the customer told the ETA. - 04Decisions such as airworthiness, product disposal or high cost route to the responsible manager. Package scope - Diagnosis of fault and service requests; remote fix or field dispatch - Technician, part and route planning; customer appointment - Work order, procedure and part card to the field team - Customer updates and closure verification What you need to start - API access to the field service management (FSM) system - Device and asset records, part inventory - Technician skill and territory definitions Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Control Tower SupervisorCold-Chain SupervisorMaintenance Supervisor Specialist agents Prediction AgentRouting AgentCustomer AgentFinance AgentSensor AgentQuality AgentField AgentSignal AgentField Note AgentDiagnosis AgentPlanning AgentFault AgentTech Pub AgentParts Agent Models (Hub catalogue) aginies-tulpar-4aginies-talay-2aginies-sunkar-1aginies-voice-tr-v3 Systems touched TMSTelematicsCarrier APICustomer PortalERP BillingIoT PlatformWMS / Lot TrackingStability Rules (vector)Field ServiceWhatsApp Business APISCADA / HistorianCMMS (Maximo)SAP S/4HANA MMManuals & Work Order Vector StoreMaintenance Management SystemTechnical Publications (S1000D vector)Spares InventoryTechnician App Channels apiemailwhatsappappvoicebatch Simulation volumes 106,200runs / month 93.6%success rate 7.1 smedian duration $0.0031estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Signal to service order in minutes, with no manual entry - First-visit fix with the right part and the right task card - Customer and quality records current at the same time Capabilities used Event ingestionTime-series toolsMulti-source RAGREST / MCP toolsHuman-in-the-loopFull I/O logging Explore the platform → Measured targets | Requests resolved remotely | 30–45% | | First-visit fix rate | +15 points | | Jobs per technician per day | +20% | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Control tower: delay prediction, rerouting, customer notice, SLARoute cost increase > 15% or compensation > 10,000 → operations manager approval - Cold-chain excursion responseProduct transfer or disposal decision → quality manager approval - Maintenance ticket triagePlanned stop > 4 hours → shift supervisor approval - Platform maintenance (MRO) work order preparationEvery work package affecting airworthiness → maintenance planner approval - packageOut-of-warranty replacement and costly parts under human approval - packageRemote commands limited to the device owner’s consent scope Rollout - 1 · Process map · 2 weeksSignal sources, decision points and work-order systems mapped; the automatic-decision boundary and approval chain defined. - 2 · Pilot · 4 weeksOne line, one region or one process; the agent prepares the recommendation, a person applies it; deviation and gain measured. - 3 · Autonomous operationDecisions below the threshold run automatically, above it go to approval; shift reports and SLA dashboards live. Related agents ## This package’s scenarios in the agent library LogisticsHITL ### Control tower: delay prediction, rerouting, customer notice, SLA The telematics signal catches the delay early; alternative route and carrier chosen, customer informed, SLA compensation computed. 96,000 runs / month 93.4% success 5 agents IoT & Connected DevicesHITL ### Cold-chain excursion response A temperature excursion is assessed instantly, product risk computed, a field crew dispatched, the quality decision recorded. 2,700 runs / month 97.4% success 4 agents Manufacturing & AutomotiveHITL ### Maintenance ticket triage Sensor alarm and technician voice note become one work order, spare parts and a safe stop window included. 3,400 runs / month 95.2% success 5 agents Defence & AerospaceHITL ### Platform maintenance (MRO) work order preparation The fault report is matched to technical publications, task card and parts list are produced, the work order opens with planner approval. 4,100 runs / month 95.7% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Financial analysis & operational intelligence URL: https://www.aginies.com/en/solutions/financial-analysis-operational-intelligence Language: en Summary: Finance teams stitch data together by hand across invoices, statements and investment instructions. Reconciliation exceptions surface late; by the time the analysis is ready, the decision window has closed. Fintech & Payments · Financial analysis & operational intelligence # Financial analysis & operational intelligence Agents that read financial data and documents, interpret them and support decisions. In short Financial analysis & operational intelligence is an Aginies solution package for the Fintech & Payments sector in the “financial analysis & operational intelligence” area. What the agents do: The financial assistant answers questions straight from statements, ledgers and reports; every figure links to its source; the invoice-processing agent extracts the document, matches it to order and receipt, and flags exceptions. Expected outcomes: reconciliation and invoice cycles drop from days to hours; decisions taken on current, source-linked data; exceptions and risk patterns visible before they materialise. The package uses the Entity extraction, Multi-source RAG, ERP / core tools (MCP), Anomaly detection capabilities of the Aginies platform and can be watched live through the Merchant operations assistant: POS, settlement, refund and limit, Payment reconciliation exceptions, Supplier invoice three-way match, BNPL instant instalment decision scenarios in the agent library. The problem Finance teams stitch data together by hand across invoices, statements and investment instructions. Reconciliation exceptions surface late; by the time the analysis is ready, the decision window has closed. What the agents do - 01The financial assistant answers questions straight from statements, ledgers and reports; every figure links to its source. - 02The invoice-processing agent extracts the document, matches it to order and receipt, and flags exceptions. - 03Investment instructions are analysed: amount, instrument, limits and suitability rules are checked. - 04Above-threshold transactions and anomalous patterns reach the finance manager with a reasoned summary. Package scope - Revenue, cost and cash data collected and validated from several sources - Variance analysis, anomaly detection and root-cause explanation - Management report draft: tables, charts and commentary with sources - Q&A: data-backed answers to questions such as “why did marketing cost rise this month?” What you need to start - Read access to the data warehouse or ERP reporting layer - Chart of accounts, cost centres and KPI definitions - Report templates and distribution list Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Merchant SupervisorReconciliation SupervisorInvoice SupervisorCredit Decision Supervisor Specialist agents Dialogue AgentPOS AgentOperations AgentRisk AgentMatching AgentInvestigation AgentCorrection AgentCapture AgentPosting AgentIntake AgentBureau AgentDecision AgentBooking Agent Models (Hub catalogue) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-konrul-1 Systems touched WhatsApp BusinessTerminal ManagementSettlement LedgerRefund ServiceTicketingPSP / Acquirer APICard NetworkCore LedgerCase ManagementEmailSAP S/4HANAExchange / M365 MailTolerance Policy Vector StoreMerchant Checkout APICredit Bureau (KKB / Findeks)Open Banking APISMS Gateway Channels whatsappwebvoicebatchapiemailappsms Simulation volumes 139,620runs / month 95.7%success rate 7.2 smedian duration $0.0028estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Reconciliation and invoice cycles drop from days to hours - Decisions taken on current, source-linked data - Exceptions and risk patterns visible before they materialise Capabilities used Entity extractionMulti-source RAGERP / core tools (MCP)Anomaly detectionHuman-in-the-loopFull audit trail Explore the platform → Measured targets | Monthly report preparation time | 60–70% shorter | | Time to detect an anomaly | From month-end to the same week | | Report statements with a source | 100% | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Merchant operations assistant: POS, settlement, refund and limitRefund > 25,000 or limit increase > 50% → risk team approval - Payment reconciliation exceptionsSingle break > 10,000 or total > 50,000 → finance manager approval before adjustment - Supplier invoice three-way matchPrice deviation > 2% or > 1,000 TL → buyer approval - BNPL instant instalment decisionBasket > 15,000 or bureau score < 1,200 → underwriter approval - packageReports are distributed with the finance manager’s approval; the agent drafts - packageRole-based data access; payroll and salary data under a separate policy Rollout - 1 · Documents and rule set · 2 weeksDocument types, field dictionary, matching and tolerance rules written with the finance team; a sample set labelled. - 2 · Parallel run · 4 weeksAgent and current process handle the same documents; the difference report reviewed weekly, cut-over when the accuracy threshold is met. - 3 · Close integrationTied to the month-end calendar; exception queue, audit trail and reconciliation reports opened to internal audit. Related agents ## This package’s scenarios in the agent library Fintech & PaymentsHITL ### Merchant operations assistant: POS, settlement, refund and limit The merchant writes on WhatsApp; POS fault, settlement query, refund and limit increase are resolved in one dialogue. 64,000 runs / month 93.8% success 5 agents Fintech & PaymentsHITL ### Payment reconciliation exceptions PSP settlement files matched to the ledger; small breaks fix themselves, large ones go to finance approval. 620 runs / month 95.1% success 4 agents Manufacturing & AutomotiveHITL ### Supplier invoice three-way match Invoice, PO and goods receipt matched in SAP automatically; only out-of-tolerance lines reach a buyer. 21,000 runs / month 97.1% success 4 agents Fintech & PaymentsHITL ### BNPL instant instalment decision Checkout instalment decision in seconds from bureau and open-banking data; large tickets go to underwriting. 54,000 runs / month 97.3% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Fraud and dispute operations URL: https://www.aginies.com/en/solutions/fraud-and-dispute-operations Language: en Summary: Fraud alerts wait in the analyst queue, dispute files miss card-scheme deadlines, confirmation calls to customers are late. Every team keeps its own list. Banking · Operational automation # Fraud and dispute operations Alert triage, chargeback files, card blocks and early collections on one operations desk. In short Fraud and dispute operations is an Aginies solution package for the Banking sector in the “operational automation” area. What the agents do: Alerts are scored, merchant and behaviour analysis added, confirmation taken from the customer by SMS or call; dispute files open with evidence list, scheme code and deadline; missing evidence is requested automatically. Expected outcomes: A large drop in review time per alert; no dispute file past its deadline; fewer false-positive blocks, fewer customer complaints. The package uses the Fraud engine tools, Case management, Supervisor · sub-agents, Human-in-the-loop capabilities of the Aginies platform and can be watched live through the Card fraud alert triage, Chargeback dispute handling, Dialogue banking: voice and text transaction assistant, Early-stage collections outreach scenarios in the agent library. The problem Fraud alerts wait in the analyst queue, dispute files miss card-scheme deadlines, confirmation calls to customers are late. Every team keeps its own list. What the agents do - 01Alerts are scored, merchant and behaviour analysis added, confirmation taken from the customer by SMS or call. - 02Dispute files open with evidence list, scheme code and deadline; missing evidence is requested automatically. - 03Card block, replacement and provisional refund are executed in core systems; the customer is informed at every step. - 04Early collections contact comes with a payment plan offer; permanent decisions go to operations approval. Package scope - Fraud alert triage: score, merchant analysis, confirmation with the customer - Chargeback and dispute files: evidence collection, card-scheme rules - Card block, replacement card and provisional refund operations - Early-stage collections contact and payment plan What you need to start - API access to card management, fraud engine and dispute systems - Card-scheme dispute rules and deadlines - Customer contact channels and consent records Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Fraud SupervisorDispute SupervisorDialogue Banking SupervisorCollections Supervisor Specialist agents Triage AgentOutreach AgentCard AgentIntake AgentEvidence AgentFiling AgentDialogue AgentIdentity AgentTransaction AgentRisk AgentSegmentation AgentPayment Plan Agent Models (Hub catalogue) aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Systems touched Fraud EngineCard ManagementWhatsApp Business APISalesforce CRMSMS GatewayDocument Store (S3)Visa / Mastercard Dispute APIEmail (Exchange)Mobile App ChatCore BankingCollections Platform Channels whatsappsmsvoiceapiappemailbranchbatch Simulation volumes 356,800runs / month 94.2%success rate 7.1 smedian duration $0.0026estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - A large drop in review time per alert - No dispute file past its deadline - Fewer false-positive blocks, fewer customer complaints Capabilities used Fraud engine toolsCase managementSupervisor · sub-agentsHuman-in-the-loopFull audit trailSMS / voice channels Explore the platform → Measured targets | Review time per alert | 70% shorter | | Dispute files past the deadline | Near zero | | False-positive blocks | 30% fewer | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Card fraud alert triageConfirmed fraud loss > 2,500 → analyst approval before permanent cancellation - Chargeback dispute handlingAmount > 2,500 → dispute team lead approval before provisional credit - Dialogue banking: voice and text transaction assistantLimit increase > 60%, dispute amount > 5,000 or transfer to a new payee → operations approval - Early-stage collections outreachPlan balance > 25,000 or > 3 instalments → collections team lead approval - packagePermanent block and loss write-off always under operations approval - packageEvery decision on file with evidence and score; ready for internal audit and scheme review Rollout - 1 · Process map · 2 weeksSignal sources, decision points and work-order systems mapped; the automatic-decision boundary and approval chain defined. - 2 · Pilot · 4 weeksOne line, one region or one process; the agent prepares the recommendation, a person applies it; deviation and gain measured. - 3 · Autonomous operationDecisions below the threshold run automatically, above it go to approval; shift reports and SLA dashboards live. Related agents ## This package’s scenarios in the agent library BankingHITL ### Card fraud alert triage The card is blocked in seconds and the customer asked on WhatsApp; the reply releases or cancels the card. 65,000 runs / month 97.1% success 4 agents BankingHITL ### Chargeback dispute handling Dispute captured, evidence assembled and filed with the card scheme; large provisional credits go to approval. 9,800 runs / month 95.2% success 4 agents BankingHITL ### Dialogue banking: voice and text transaction assistant The customer talks or types; bill, limit, dispute and transfer are completed in core banking in one session. 186,000 runs / month 94.7% success 5 agents BankingHITL ### Early-stage collections outreach Past-due accounts are segmented and reminded on the right channel; customers set up their own payment plan. 96,000 runs / month 91.2% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Healthcare operations & clinical guidance URL: https://www.aginies.com/en/solutions/healthcare-operations-clinical-guidance Language: en Summary: Patient requests pile up across call centre, portal and messaging; wrong-department bookings and no-shows erode capacity. Clinical teams spend their time on administrative routing. Healthcare · Customer interaction & experience # Healthcare operations & clinical guidance Smart triage and patient referral; clinical decisions always stay with clinicians. In short Healthcare operations & clinical guidance is an Aginies solution package for the Healthcare sector in the “customer interaction & experience” area. What the agents do: Patient requests and complaints are analysed; urgency and the right department determined by the institution’s triage protocol; the scheduling agent proposes the best slot from clinician calendars and capacity; reminders and rebooking are automatic. Expected outcomes: patients routed to the right department and time at first contact; fewer no-shows, higher capacity utilisation; clinical teams relieved of administrative load. The package uses the Intent detection, Triage protocol guardrails, Scheduling tools (MCP), Human-in-the-loop capabilities of the Aginies platform and can be watched live through the Patient journey coordinator: appointment, preparation, approval and results, Patient intake and triage, Appointment scheduling and no-show reduction, Prior authorization scenarios in the agent library. The problem Patient requests pile up across call centre, portal and messaging; wrong-department bookings and no-shows erode capacity. Clinical teams spend their time on administrative routing. What the agents do - 01Patient requests and complaints are analysed; urgency and the right department determined by the institution’s triage protocol. - 02The scheduling agent proposes the best slot from clinician calendars and capacity; reminders and rebooking are automatic. - 03Prior-authorisation and referral documents are prepared; missing information is requested from the patient. - 04Red-flag symptoms escalate to a clinician immediately; diagnosis and treatment decisions never sit with the agent. Package scope - Appointment, pre-registration, preparation and result notice in one coordinator - Insurer pre-approval and public/private insurer queries - Preparation and contraindication checks tied to clinical protocols - Call centre and WhatsApp channel; approval card to the physician What you need to start - API access to the hospital information and scheduling systems - Clinical protocols and the preparation instruction set - Patient identity verification and consent records Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Patient Journey SupervisorTriage SupervisorAppointment SupervisorPrior Auth Supervisor Specialist agents Dialogue AgentScheduling AgentClinical Preparation AgentInsurer AgentVoice AgentSymptom AgentUrgency AgentRouting AgentNo-show Risk AgentBooking AgentOrder AgentPayer Rules AgentSubmission Agent Models (Hub catalogue) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-voice-tr-v3aginies-konrul-1 Systems touched WhatsApp BusinessHospital Information SystemSchedulingInsurer PortalLab / ImagingEHR (HL7/FHIR)HIS SchedulingTriage Protocol Vector StoreWhatsApp Business APIHospital Information System (HIS)SMS GatewayPayer Policy Vector StorePayer PortalExchange / M365 Mail Channels whatsappvoiceappsmsbatchapiemailweb Simulation volumes 150,700runs / month 96.4%success rate 7.3 smedian duration $0.0023estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Patients routed to the right department and time at first contact - Fewer no-shows, higher capacity utilisation - Clinical teams relieved of administrative load Capabilities used Intent detectionTriage protocol guardrailsScheduling tools (MCP)Human-in-the-loopSensitive data maskingFull audit trail Explore the platform → Measured targets | No-show rate | 30–40% lower | | Procedures postponed for missing preparation | 50% fewer | | Average handling time at the call centre | 35% shorter | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Patient journey coordinator: appointment, preparation, approval and resultsDrug interaction, suspected contraindication or extra test need → physician approval - Patient intake and triageA triage nurse confirms the urgency level on every case; "emergency" cases are directed to 112 immediately - Prior authorizationSurgical procedure or cost > 25,000 TL → utilisation management nurse approval - packageClinical decisions never sit with the agent; physician approval when in doubt - packageHealth data processed inside the hospital perimeter; access by role and purpose Rollout - 1 · Discovery and design · 2–3 weeksThe 20 most frequent intents and transactions are chosen, identity and approval thresholds written with information security, channel and system integrations listed. - 2 · Pilot · 4–6 weeksOne channel, a limited customer group; the test corpus runs, a human representative shadows, thresholds tuned on real traffic. - 3 · Scale-outRemaining channels and transactions added; dashboards, SIEM forwarding and kill-switch handed over; monthly model and rule review begins. Related agents ## This package’s scenarios in the agent library HealthcareHITL ### Patient journey coordinator: appointment, preparation, approval and results The patient writes or calls; appointment, preparation instructions, insurer pre-approval and result notice come together in one coordinator. 58,000 runs / month 95.1% success 5 agents HealthcareHITL ### Patient intake and triage The patient types or talks; intake and urgency scoring take seconds, and a nurse always makes the final call. 24,500 runs / month 95.8% success 5 agents Healthcare ### Appointment scheduling and no-show reduction High no-show-risk patients are reached on WhatsApp in advance; freed slots are refilled from the waiting list. 62,000 runs / month 98.2% success 4 agents HealthcareHITL ### Prior authorization Payer rules searched, clinical evidence gathered and the packet ready to submit within minutes of the order. 6,200 runs / month 94.1% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # HR & employee experience URL: https://www.aginies.com/en/solutions/hr-employee-experience Language: en Summary: Applications are screened by hand and new hires spend their first week chasing access and forms. HR answers the same questions repeatedly, leaving no time for strategic work. Enterprise Functions · HR & employee experience # HR & employee experience From recruitment to onboarding: data-driven, fast and scalable HR processes. In short HR & employee experience is an Aginies solution package for the Enterprise Functions sector in the “hr & employee experience” area. What the agents do: The recruitment agent pre-screens applications against the role criteria and schedules interviews; the final decision stays with HR and the manager; the onboarding agent coordinates accounts, equipment and training assignments across HR, IT and the manager. Expected outcomes: shorter HR cycle from evaluation to adaptation; new hires ready on day one: access, equipment, training; HR freed from repetitive work to focus on experience. The package uses the HRIS / ITSM tools (MCP), Multi-source RAG, Supervisor · sub-agents, Human-in-the-loop capabilities of the Aginies platform and can be watched live through the Employee onboarding, IT helpdesk L1, Email classification and attachment check scenarios in the agent library. The problem Applications are screened by hand and new hires spend their first week chasing access and forms. HR answers the same questions repeatedly, leaving no time for strategic work. What the agents do - 01The recruitment agent pre-screens applications against the role criteria and schedules interviews; the final decision stays with HR and the manager. - 02The onboarding agent coordinates accounts, equipment and training assignments across HR, IT and the manager. - 03The internal helpdesk answers leave, payroll and policy questions instantly from corporate sources. - 04Personal data is masked; every step is traced in HRIS and ITSM records. Package scope - Recruitment pre-screening, interview scheduling and candidate communication - Onboarding: accounts, equipment, training and first-week plan - Employee questions: leave, payroll, policy; cited answers - Off-boarding and role-change automation What you need to start - API access to the HR system (SAP SuccessFactors, Workday, local HRIS) - Integration with identity management and IT asset systems - Current HR policies and process owners Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Onboarding SupervisorHelpdesk SupervisorMailroom Supervisor Specialist agents HR AgentIdentity AgentIT AgentTraining AgentIntent AgentAccess AgentTicket AgentClassification AgentAttachment AgentRouting Agent Models (Hub catalogue) aginies-sunkar-1aginies-talay-2aginies-tulpar-4aginies-konrul-1 Systems touched SAP SuccessFactorsActive Directory / Entra IDServiceNowLMSExchange / M365 MailMicrosoft TeamsMailbox Rules & Checklists Vector StoreDocument Store (S3) Channels apiemailappvoiceweb Simulation volumes 134,900runs / month 97.9%success rate 5.5 smedian duration $0.0007estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Shorter HR cycle from evaluation to adaptation - New hires ready on day one: access, equipment, training - HR freed from repetitive work to focus on experience Capabilities used HRIS / ITSM tools (MCP)Multi-source RAGSupervisor · sub-agentsHuman-in-the-loopPII maskingAudit trail Explore the platform → Measured targets | Onboarding completion time | From weeks to days | | Repetitive questions reaching HR | 60% fewer | | Candidate response time | under 24 hours | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Employee onboardingAccess outside the role template or admin rights → line manager approval - IT helpdesk L1Access to a finance or production group → group owner approval - packageHiring decision and pay offer always with people; the agent matches and schedules - packageEmployee data purpose-limited; only authorised roles reach payroll information Rollout - 1 · Policy and role catalogue · 2 weeksHR policies, process owners and system access defined; personal-data boundaries written. - 2 · Pilot · 4 weeksOne department or one role; quality score and user feedback tracked weekly. - 3 · Scale-outAll departments; thresholds, review sampling and kill-switch owned by HR management. Related agents ## This package’s scenarios in the agent library Enterprise FunctionsHITL ### Employee onboarding Accounts, kit and training plan ready the moment HR creates the hire; off-template access goes to the manager. 1,900 runs / month 97.2% success 5 agents Enterprise FunctionsHITL ### IT helpdesk L1 Password resets and access requests solved in chat; access to finance groups asks the owner via a Teams card. 47,000 runs / month 98.5% success 4 agents Enterprise Functions ### Email classification and attachment check Every e-mail in the shared mailbox is classified, attachments checked against a list, gaps asked for at once. 86,000 runs / month 97.6% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Information management & decision support URL: https://www.aginies.com/en/solutions/information-management-decision-support Language: en Summary: Corporate knowledge is scattered across regulations, circulars, contracts and digital assets in different systems. Finding the right document takes hours; decisions rest on stale or partial information. Government & Public Sector · Information management & decision support # Information management & decision support Source-linked access to regulations, policies and contracts in seconds. In short Information management & decision support is an Aginies solution package for the Government & Public Sector sector in the “information management & decision support” area. What the agents do: Regulations and policies are indexed with versions; RAG answers questions clause by clause, with citations; the regulatory-analysis agent detects changes and lists the affected processes and documents. Expected outcomes: access to information drops from hours to seconds; decisions grounded in current regulation and cited documents; impact of regulatory change seen proactively. The package uses the Multi-source RAG, Document embedding pipeline, Version-aware retrieval, Citation grounding capabilities of the Aginies platform and can be watched live through the Contract clause extraction, Permit application check, Prior authorization, Customs document preparation scenarios in the agent library. The problem Corporate knowledge is scattered across regulations, circulars, contracts and digital assets in different systems. Finding the right document takes hours; decisions rest on stale or partial information. What the agents do - 01Regulations and policies are indexed with versions; RAG answers questions clause by clause, with citations. - 02The regulatory-analysis agent detects changes and lists the affected processes and documents. - 03Contract playbooks: clauses compared with standard positions, deviations flagged. - 04Digital assets (images, documents, records) are auto-tagged and searchable; access follows permissions. Package scope - Cited Q&A over regulation, procedures and the organisation’s documents - Document comparison, summarisation and change tracking - Access policy: who may see which document, enforced in the answer - Decision note and reasoning draft; published with expert approval What you need to start - Read access to the document management system or file shares - Document owners and update responsibles - Access policy mapped to SSO groups Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Contract SupervisorPermit SupervisorPrior Auth SupervisorCustoms Supervisor Specialist agents Clause AgentPlaybook AgentCLM AgentDocument AgentRegulation AgentFile AgentOrder AgentPayer Rules AgentSubmission AgentTariff AgentBroker Agent Models (Hub catalogue) aginies-tulpar-4aginies-talay-2aginies-konrul-1aginies-sunkar-1 Systems touched CLM (Contract Lifecycle Management)Legal Playbook Vector StoreDocument Store (S3)Exchange / M365 Maile-Devlet GatewayCase ManagementGIS / Zoning PlanRegulation Vector StoreEHR (HL7/FHIR)Payer Policy Vector StorePayer PortalSAP S/4HANAWMSTariff & Regulation Vector StoreCustoms Broker Portal Channels emailwebapibatch Simulation volumes 15,000runs / month 94.7%success rate 7.2 smedian duration $0.011estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Access to information drops from hours to seconds - Decisions grounded in current regulation and cited documents - Impact of regulatory change seen proactively Capabilities used Multi-source RAGDocument embedding pipelineVersion-aware retrievalCitation groundingRole-based accessEntity extraction Explore the platform → Measured targets | Answers with a citation | 100% | | Answers judged correct by experts | 90% and above | | Time spent searching for information | 60% lower | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Contract clause extractionHigh-risk deviation or contract value > 500,000 € → counsel approval - Permit application checkEvery permit decision is taken by the licensing officer; the system only pre-reviews - Prior authorizationSurgical procedure or cost > 25,000 TL → utilisation management nurse approval - Customs document preparationHS confidence < 0.90 or duty difference > 2,000 € → customs specialist approval - packageNo answer without a source; the agent says it does not know - packageLegal interpretation and external sharing always under expert approval Rollout - 1 · Source inventory · 2 weeksKnowledge sources, owners and access policies listed; chunking and tagging rules defined. - 2 · Pilot · 3–4 weeksWith one unit; cited answers scored by experts, coverage gaps closed. - 3 · Organisation-wideAccess policies bound to SSO groups; the update pipeline and quality dashboard live. Related agents ## This package’s scenarios in the agent library Enterprise FunctionsHITL ### Contract clause extraction Clauses are extracted and compared with the legal playbook; deviations reach counsel with redline suggestions. 1,400 runs / month 95.1% success 4 agents Government & Public SectorHITL ### Permit application check Missing documents and regulatory fit reported in minutes; the licensing officer always makes the decision. 3,100 runs / month 95.8% success 4 agents HealthcareHITL ### Prior authorization Payer rules searched, clinical evidence gathered and the packet ready to submit within minutes of the order. 6,200 runs / month 94.1% success 4 agents LogisticsHITL ### Customs document preparation HS codes, invoice and packing list prepared in minutes; only low-confidence lines reach a customs specialist. 4,300 runs / month 94.7% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Invoice & document extraction URL: https://www.aginies.com/en/solutions/invoice-document-extraction Language: en Summary: Invoices, contracts and forms arrive by email and portal in every format. Data entry is slow, error-prone, and a missing document stalls the process. Enterprise Functions · Financial analysis & operational intelligence # Invoice & document extraction OCR → validation → ERP: clean records without manual entry. In short Invoice & document extraction is an Aginies solution package for the Enterprise Functions sector in the “financial analysis & operational intelligence” area. What the agents do: Documents are classified; fields are extracted regardless of layout; extracted data is matched against purchase orders and goods receipts; tolerances are checked. Expected outcomes: key fields extracted without manual entry; finance cycles shrink from days to minutes; validated data flows straight into ERP and business systems. The package uses the Entity extraction, Document embedding pipeline, Stateful execution, Full I/O logging capabilities of the Aginies platform and can be watched live through the Procure-to-pay desk: request, quotes, contract and invoice in one place, Supplier invoice three-way match, Accounts payable inbox, Underwriting document extraction scenarios in the agent library. The problem Invoices, contracts and forms arrive by email and portal in every format. Data entry is slow, error-prone, and a missing document stalls the process. What the agents do - 01Documents are classified; fields are extracted regardless of layout. - 02Extracted data is matched against purchase orders and goods receipts; tolerances are checked. - 03Missing or inconsistent documents are flagged instantly with automatic feedback to the sender. - 04Validated records are posted to the ERP; every field keeps its source document and confidence score. Package scope - Invoice, delivery note, contract and form extraction; per-field confidence and source document - Three-way match with PO and goods receipt, tolerance rules - Exception queue: missing field, amount deviation, duplicate invoice - ERP posting and payment scheduling tools What you need to start - Posting and query API access to the ERP (SAP, Oracle, Logo, Netsis) - 200–500 sample documents covering the document types - Tolerance and approval rules from the finance team Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Procurement SupervisorInvoice SupervisorAP SupervisorUnderwriting Supervisor Specialist agents Request AgentSourcing AgentContract AgentInvoice AgentCapture AgentMatching AgentPosting AgentCoding AgentDocument AgentMedical Assessment AgentUnderwriting Agent Models (Hub catalogue) aginies-sunkar-1aginies-tulpar-4aginies-talay-2aginies-konrul-1 Systems touched Procurement InboxERPSupplier PortalContract RepositoryPayment HubSAP S/4HANAExchange / M365 MailTolerance Policy Vector StoreOracle EBSDocument Store (S3)Vendor Master / ProcurementUnderwriting WorkbenchVector StoreEmail (Exchange) Channels emailapibatchweb Simulation volumes 67,800runs / month 95.9%success rate 7.8 smedian duration $0.0044estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Key fields extracted without manual entry - Finance cycles shrink from days to minutes - Validated data flows straight into ERP and business systems Capabilities used Entity extractionDocument embedding pipelineStateful executionFull I/O loggingHuman-in-the-loopREST / MCP Explore the platform → Measured targets | Touchless document share | 70–85% | | Field accuracy (critical fields) | 98% and above | | Invoice processing time | From days to hours | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Procure-to-pay desk: request, quotes, contract and invoice in one placeAmount > 250,000, single quote or non-framework supplier → procurement committee - Supplier invoice three-way matchPrice deviation > 2% or > 1,000 TL → buyer approval - Accounts payable inboxNon-PO invoice > 25,000 TL → cost centre owner approval - Underwriting document extractionLoading ≥ 50% or sum assured > 5,000,000 → senior underwriter approval - packageOut-of-tolerance deviation and new supplier always to a person - packageSource document and extracted value stored together; audit sample ready Rollout - 1 · Documents and rule set · 2 weeksDocument types, field dictionary, matching and tolerance rules written with the finance team; a sample set labelled. - 2 · Parallel run · 4 weeksAgent and current process handle the same documents; the difference report reviewed weekly, cut-over when the accuracy threshold is met. - 3 · Close integrationTied to the month-end calendar; exception queue, audit trail and reconciliation reports opened to internal audit. Related agents ## This package’s scenarios in the agent library Enterprise FunctionsHITL ### Procure-to-pay desk: request, quotes, contract and invoice in one place The request arrives by email; quotes compared, contract clauses checked, PO opened, invoice three-way matched and payment scheduled. 14,000 runs / month 92.7% success 5 agents Manufacturing & AutomotiveHITL ### Supplier invoice three-way match Invoice, PO and goods receipt matched in SAP automatically; only out-of-tolerance lines reach a buyer. 21,000 runs / month 97.1% success 4 agents Enterprise FunctionsHITL ### Accounts payable inbox Every invoice in the ap@ mailbox is captured, coded and posted to the ERP; only exceptions reach an analyst. 28,000 runs / month 96.8% success 4 agents InsuranceHITL ### Underwriting document extraction Medical reports and labs read and assessed against the manual; underwriters only see files above threshold. 4,800 runs / month 94.9% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # IoT device fleet operations URL: https://www.aginies.com/en/solutions/iot-device-fleet-operations Language: en Summary: Thousands of devices raise hundreds of alarms from one root cause; firmware goes to the whole fleet at once with no rollback plan. Telemetry volume makes cloud model cost unsustainable. IoT & Connected Devices · Software development & technology productivity # IoT device fleet operations Anomaly clustering, root cause and health-gated OTA rollouts: one incident, one decision per fleet. In short IoT device fleet operations is an Aginies solution package for the IoT & Connected Devices sector in the “software development & technology productivity” area. What the agents do: Telemetry anomalies are clustered by model, firmware and gateway; one incident is opened; root-cause candidates are scored against firmware history and known issues; a confidence score is written. Expected outcomes: alarm noise collapses into incidents; the team looks at the real problem; safe, evidenced and reversible firmware rollouts; low unit cost with small on-prem models. The package uses the Event ingestion, Clustering, Time-series tools, Staged rollout capabilities of the Aginies platform and can be watched live through the Field device support assistant: telemetry, remote diagnosis, OTA and technician, Device fleet anomaly triage, Staged OTA update rollout, Cold-chain excursion response scenarios in the agent library. The problem Thousands of devices raise hundreds of alarms from one root cause; firmware goes to the whole fleet at once with no rollback plan. Telemetry volume makes cloud model cost unsustainable. What the agents do - 01Telemetry anomalies are clustered by model, firmware and gateway; one incident is opened. - 02Root-cause candidates are scored against firmware history and known issues; a confidence score is written. - 03OTA rollouts proceed in 1% → 10% → 100% stages; a health gate at each, automatic rollback on failure. - 04Bulk device commands and critical segments pass the fleet operations lead or product owner. Package scope - Clustering of telemetry anomalies into incidents - Staged OTA rollout; pilot wave, stop criteria, rollback - Remote diagnosis and commands; field-team integration - Fleet health dashboard and knowledge-base updates What you need to start - Access to the IoT platform or MQTT broker; device management API - Time-series database and fleet registry - OTA signing and rollout policy Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Device Support SupervisorFleet Health SupervisorRollout SupervisorCold-Chain Supervisor Specialist agents Dialogue AgentTelemetry AgentDiagnosis AgentDispatch AgentSignal AgentOperations AgentPlan AgentHealth AgentDeploy AgentSensor AgentQuality AgentField Agent Models (Hub catalogue) aginies-talay-2aginies-sunkar-1aginies-tulpar-4 Systems touched WhatsApp BusinessDevice Cloud (MQTT)Fleet RegistryOTA ServiceField ServiceIoT Platform (MQTT / Digital Twin)Time-series DBKnown Issues (vector)ITSM / IncidentSlackCI/CD (signed artefacts)Device Management / OTAObservability (metrics)IoT PlatformWMS / Lot TrackingStability Rules (vector)WhatsApp Business API Channels whatsappwebapibatchapp Simulation volumes 58,020runs / month 95.1%success rate 6.9 smedian duration $0.0025estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Alarm noise collapses into incidents; the team looks at the real problem - Safe, evidenced and reversible firmware rollouts - Low unit cost with small on-prem models Capabilities used Event ingestionClusteringTime-series toolsStaged rolloutKill-switchOn-prem models Explore the platform → Measured targets | Alarm-to-incident clustering | 80% less noise | | OTA rollback rate | under 1% | | Faults needing a field visit | 35% fewer | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Field device support assistant: telemetry, remote diagnosis, OTA and technicianBulk OTA (>100 devices) or out-of-warranty device replacement → product-owner approval - Device fleet anomaly triageBulk device command > 100 devices → fleet operations lead approval - Staged OTA update rolloutRollout to a critical fleet segment → product owner approval; automatic rollback if a health gate fails - Cold-chain excursion responseProduct transfer or disposal decision → quality manager approval - packageBulk OTA and critical segments always under product-owner approval - packageDevice commands signed and logged; read-only on safety-critical devices Rollout - 1 · Pipeline integration · 2 weeksJira, Git, CI/CD or device management connected; permission model and scope boundaries written. - 2 · Pilot · 4–6 weeksOne team or one device segment; every output passes human review, acceptance rate measured. - 3 · ScaleRemaining teams and segments; automated steps widen, production and architecture decisions stay with people. Related agents ## This package’s scenarios in the agent library IoT & Connected DevicesHITL ### Field device support assistant: telemetry, remote diagnosis, OTA and technician The customer says "the device is dead"; telemetry read, remote diagnosis made, update pushed, technician dispatched with the right part if needed. 37,000 runs / month 94.1% success 5 agents IoT & Connected DevicesHITL ### Device fleet anomaly triage Telemetry anomalies from thousands of devices are clustered, root-cause candidates listed, one incident reaches the right team. 18,000 runs / month 96.9% success 4 agents IoT & Connected DevicesHITL ### Staged OTA update rollout Firmware rolls out in 1% → 10% → 100% stages; a health gate at every stage, human approval for critical fleets. 320 runs / month 98.1% success 4 agents IoT & Connected DevicesHITL ### Cold-chain excursion response A temperature excursion is assessed instantly, product risk computed, a field crew dispatched, the quality decision recorded. 2,700 runs / month 97.4% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # KVKK / GDPR personal data protection URL: https://www.aginies.com/en/solutions/kvkk-personal-data-protection Language: en Summary: Personal data is scattered across contracts, emails and call recordings; nobody knows exactly what sits where. Every document and conversation sent to a model carries KVKK and GDPR risk unless it is masked. Enterprise Functions · Data protection & compliance # KVKK / GDPR personal data protection Detect personal data in documents and calls, mask it, and keep an audit-ready data map. In short KVKK / GDPR personal data protection is an Aginies solution package for the Enterprise Functions sector in the “data protection & compliance” area. What the agents do: The KVKK Personal Data Detection agent finds and classifies national IDs, IBANs, health and contact data in documents and conversations; the KVKK Document Masking agent masks or anonymises fields before data reaches any model. Expected outcomes: personal data masked before it reaches a model; an audit-ready, current data inventory; compliance risk in document and call archives made visible. The package uses the PII detection, PII masking, Entity extraction, On-prem deployment capabilities of the Aginies platform and can be watched live through the Underwriting document extraction, Mortgage document review, Call centre quality monitoring, Email classification and attachment check scenarios in the agent library. The problem Personal data is scattered across contracts, emails and call recordings; nobody knows exactly what sits where. Every document and conversation sent to a model carries KVKK and GDPR risk unless it is masked. What the agents do - 01The KVKK Personal Data Detection agent finds and classifies national IDs, IBANs, health and contact data in documents and conversations. - 02The KVKK Document Masking agent masks or anonymises fields before data reaches any model. - 03Findings feed a data map: which data, in which system, for what purpose, for how long. - 04Out-of-policy sharing and erasure requests route to the compliance team; every decision is kept with an audit trail. Package scope - Detection and classification of personal data in documents and conversations - Masking, anonymisation and retention rules - Recording and response process for data-subject requests (access, deletion) - Data inventory and processing-record updates What you need to start - Read access to document stores and communication channels - Data inventory and retention policy - Approval rules from the data controller and the KVKK team Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Underwriting SupervisorLoan File SupervisorQuality SupervisorMailroom Supervisor Specialist agents Document AgentMedical Assessment AgentUnderwriting AgentExtraction AgentIncome AgentValuation AgentTranscription AgentScoring AgentCoaching AgentClassification AgentAttachment AgentRouting Agent Models (Hub catalogue) aginies-konrul-1aginies-tulpar-4aginies-talay-2aginies-voice-tr-v3aginies-sunkar-1 Systems touched Underwriting WorkbenchDocument Store (S3)Vector StoreEmail (Exchange)Loan Origination SystemCredit Bureau (KKB)Genesys SIPQM Scorecard / Case ManagementQA Rubric Vector StoreWorkdayExchange / M365 MailMailbox Rules & Checklists Vector StoreServiceNow Channels emailwebbatchbranchvoice Simulation volumes 187,000runs / month 98.0%success rate 7.8 smedian duration $0.0017estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Personal data masked before it reaches a model - An audit-ready, current data inventory - Compliance risk in document and call archives made visible Capabilities used PII detectionPII maskingEntity extractionOn-prem deploymentData map & lineageHuman-in-the-loopAudit trail Explore the platform → Measured targets | Personal-data detection accuracy | 97% and above | | Response time to data-subject requests | Half the statutory deadline | | Manual masking effort | 80% lower | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Underwriting document extractionLoading ≥ 50% or sum assured > 5,000,000 → senior underwriter approval - Mortgage document reviewLoan > 3,000,000 or DTI > 45% → underwriting officer approval - Call centre quality monitoringScore < 60 or compliance breach → team leader review - packageDeletion and external sharing always under the data controller’s approval - packageThe detection model runs on-prem; data never leaves Türkiye Rollout - 1 · Rule and control design · 2–3 weeksRegulatory articles translated into the control set, the decision matrix approved with compliance, trace requirements written. - 2 · Shadow period · 4 weeksThe agent produces findings and recommendations, the expert decides; disagreement rate and reasoning quality measured. - 3 · Audit-ready operationControl evidence collected automatically; audit reports and the exception register opened to internal audit. Related agents ## This package’s scenarios in the agent library InsuranceHITL ### Underwriting document extraction Medical reports and labs read and assessed against the manual; underwriters only see files above threshold. 4,800 runs / month 94.9% success 4 agents BankingHITL ### Mortgage document review Payslips, social security records and appraisal read automatically; income, LTV and missing documents checked. 6,200 runs / month 94.7% success 4 agents Enterprise FunctionsHITL ### Call centre quality monitoring Every call scored against the rubric, not a sample; compliance breaches and coaching flags ready by morning. 90,000 runs / month 98.8% success 4 agents Enterprise Functions ### Email classification and attachment check Every e-mail in the shared mailbox is classified, attachments checked against a list, gaps asked for at once. 86,000 runs / month 97.6% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Logistics control tower URL: https://www.aginies.com/en/solutions/logistics-control-tower Language: en Summary: Delays are learned when the customer calls; alternative routes and carriers are chosen by hand, SLA compensation argued at month-end, customs documents prepared at the last minute. Logistics · Operational automation # Logistics control tower Operations agents that catch deviations early, reroute, inform the customer and compute the SLA. In short Logistics control tower is an Aginies solution package for the Logistics sector in the “operational automation” area. What the agents do: Deviations predicted from telematics and carrier data; SLA impact computed per shipment; route, transfer and carrier options compared on cost; rebooking executed when inside the threshold. Expected outcomes: delays known and managed hours ahead; higher on-time delivery and better customer information; transparent, auditable SLA finance. The package uses the TMS & telematics tools, Carrier APIs, Supervisor · sub-agents, Human-in-the-loop capabilities of the Aginies platform and can be watched live through the Control tower: delay prediction, rerouting, customer notice, SLA, Shipment exception handling, Customs document preparation, Cold-chain excursion response scenarios in the agent library. The problem Delays are learned when the customer calls; alternative routes and carriers are chosen by hand, SLA compensation argued at month-end, customs documents prepared at the last minute. What the agents do - 01Deviations predicted from telematics and carrier data; SLA impact computed per shipment. - 02Route, transfer and carrier options compared on cost; rebooking executed when inside the threshold. - 03Consignee and shipper informed with the new ETA and options; cold-chain excursions handled by a separate protocol. - 04Customs documents prepared from shipment data; SLA record, compensation and carrier recourse opened automatically. Package scope - Early prediction of shipment deviations from telematics and carrier data - Route and carrier alternatives, rebooking - Customs document preparation and cold-chain excursion response - Customer notices, SLA record and compensation calculation What you need to start - API access to TMS, telematics and carrier systems - SLA contracts and compensation rules - Customs and product requirements (cold-chain tolerances) Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Control Tower SupervisorShipment SupervisorCustoms SupervisorCold-Chain Supervisor Specialist agents Prediction AgentRouting AgentCustomer AgentFinance AgentException AgentImpact AgentAction AgentTariff AgentDocument AgentBroker AgentSensor AgentQuality AgentField Agent Models (Hub catalogue) aginies-tulpar-4aginies-talay-2aginies-sunkar-1 Systems touched TMSTelematicsCarrier APICustomer PortalERP BillingTMS (Transport Management)Salesforce CRMWhatsApp Business APISAP S/4HANAWMSTariff & Regulation Vector StoreCustoms Broker PortalDocument Store (S3)IoT PlatformWMS / Lot TrackingStability Rules (vector)Field Service Channels apiemailwhatsappbatchapp Simulation volumes 115,800runs / month 93.8%success rate 6.9 smedian duration $0.0032estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Delays known and managed hours ahead - Higher on-time delivery and better customer information - Transparent, auditable SLA finance Capabilities used TMS & telematics toolsCarrier APIsSupervisor · sub-agentsHuman-in-the-loopCustomer portal integrationFull audit trail Explore the platform → Measured targets | Delay detection lead | Hours ahead | | On-time delivery | +5–10 points | | Customer calls (where is my shipment) | 50% fewer | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Control tower: delay prediction, rerouting, customer notice, SLARoute cost increase > 15% or compensation > 10,000 → operations manager approval - Shipment exception handlingExtra cost > 1,000 € → dispatch lead approval - Customs document preparationHS confidence < 0.90 or duty difference > 2,000 € → customs specialist approval - Cold-chain excursion responseProduct transfer or disposal decision → quality manager approval - packageCost increases and compensation above the threshold under operations-manager approval - packageEvery routing decision recorded with cost and SLA impact Rollout - 1 · Process map · 2 weeksSignal sources, decision points and work-order systems mapped; the automatic-decision boundary and approval chain defined. - 2 · Pilot · 4 weeksOne line, one region or one process; the agent prepares the recommendation, a person applies it; deviation and gain measured. - 3 · Autonomous operationDecisions below the threshold run automatically, above it go to approval; shift reports and SLA dashboards live. Related agents ## This package’s scenarios in the agent library LogisticsHITL ### Control tower: delay prediction, rerouting, customer notice, SLA The telematics signal catches the delay early; alternative route and carrier chosen, customer informed, SLA compensation computed. 96,000 runs / month 93.4% success 5 agents LogisticsHITL ### Shipment exception handling When a delay or damage lands, impact is costed, the carrier rebooked and the customer hears before they ask. 12,800 runs / month 95.5% success 4 agents LogisticsHITL ### Customs document preparation HS codes, invoice and packing list prepared in minutes; only low-confidence lines reach a customs specialist. 4,300 runs / month 94.7% success 4 agents IoT & Connected DevicesHITL ### Cold-chain excursion response A temperature excursion is assessed instantly, product risk computed, a field crew dispatched, the quality decision recorded. 2,700 runs / month 97.4% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Production and maintenance operations URL: https://www.aginies.com/en/solutions/manufacturing-operations Language: en Summary: When a stoppage alarm fires, root cause is guesswork, part stock sits in another system, schedule changes happen by phone; warranty claims and supplier invoices wait for weeks. Manufacturing & Automotive · Operational automation # Production and maintenance operations From line stoppage to work order, from maintenance request to warranty file: agents that decide on MES, CMMS and ERP. In short Production and maintenance operations is an Aginies solution package for the Manufacturing & Automotive sector in the “operational automation” area. What the agents do: Root causes ranked from sensor history and maintenance records; part stock checked and reserved; CMMS work order opened with procedure and estimate; the production schedule simulated and presented to the shift lead. Expected outcomes: A measurable drop in unplanned downtime; minutes from alarm to work order; days instead of weeks in warranty and invoice processes. The package uses the MES / SCADA tools, CMMS & ERP integration, Supervisor · sub-agents, Human-in-the-loop capabilities of the Aginies platform and can be watched live through the Line stoppage response: from alarm to work order and shift report, Maintenance ticket triage, Warranty claim processing, Supplier invoice three-way match scenarios in the agent library. The problem When a stoppage alarm fires, root cause is guesswork, part stock sits in another system, schedule changes happen by phone; warranty claims and supplier invoices wait for weeks. What the agents do - 01Root causes ranked from sensor history and maintenance records; part stock checked and reserved. - 02CMMS work order opened with procedure and estimate; the production schedule simulated and presented to the shift lead. - 03Maintenance requests triaged; predictive alerts turned into planned maintenance. - 04Warranty claims and supplier invoices read from documents, matched, exceptions routed to people. Package scope - Line stoppage response: root cause, part, work order, schedule update - Maintenance request triage and predictive-maintenance alerts - Warranty claim processing and supplier invoice matching - Shift reports and the OEE dashboard What you need to start - API access to MES, SCADA historian and CMMS - Spare-part inventory and supplier records (ERP) - Maintenance procedures and safety rules Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Production SupervisorMaintenance SupervisorWarranty SupervisorInvoice Supervisor Specialist agents Diagnosis AgentParts AgentPlanning AgentReport AgentSignal AgentField Note AgentClaim AgentValidation AgentSettlement AgentCapture AgentMatching AgentPosting Agent Models (Hub catalogue) aginies-tulpar-4aginies-sunkar-1aginies-talay-2aginies-voice-tr-v3aginies-konrul-1 Systems touched MESSCADA HistorianCMMSERP Spare PartsShift Channel (Teams)SCADA / HistorianCMMS (Maximo)SAP S/4HANA MMManuals & Work Order Vector StoreDealer PortalSAP S/4HANAWarranty Policy Vector StoreDocument Store (S3)Exchange / M365 MailTolerance Policy Vector Store Channels apiappvoicewebemailbatch Simulation volumes 36,100runs / month 96.3%success rate 7.1 smedian duration $0.005estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - A measurable drop in unplanned downtime - Minutes from alarm to work order - Days instead of weeks in warranty and invoice processes Capabilities used MES / SCADA toolsCMMS & ERP integrationSupervisor · sub-agentsHuman-in-the-loopDocument extraction (Vision)Full audit trail Explore the platform → Measured targets | Unplanned downtime | 20–30% lower | | Alarm-to-work-order time | Minutes | | Warranty claim processing time | 60% shorter | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Line stoppage response: from alarm to work order and shift reportSchedule change > 2 hours or urgent purchase above 50,000 → shift lead approval - Maintenance ticket triagePlanned stop > 4 hours → shift supervisor approval - Warranty claim processingWarranty claim > 5,000 € → regional warranty manager approval - Supplier invoice three-way matchPrice deviation > 2% or > 1,000 TL → buyer approval - packageProduction schedule changes and urgent purchases under shift-lead approval - packageThe agent issues no equipment commands; it recommends, the maintenance team acts Rollout - 1 · Process map · 2 weeksSignal sources, decision points and work-order systems mapped; the automatic-decision boundary and approval chain defined. - 2 · Pilot · 4 weeksOne line, one region or one process; the agent prepares the recommendation, a person applies it; deviation and gain measured. - 3 · Autonomous operationDecisions below the threshold run automatically, above it go to approval; shift reports and SLA dashboards live. Related agents ## This package’s scenarios in the agent library Manufacturing & AutomotiveHITL ### Line stoppage response: from alarm to work order and shift report An MES alarm lands; root cause found from history, spare part checked, work order and plan update ready within minutes. 3,100 runs / month 92.2% success 5 agents Manufacturing & AutomotiveHITL ### Maintenance ticket triage Sensor alarm and technician voice note become one work order, spare parts and a safe stop window included. 3,400 runs / month 95.2% success 5 agents Manufacturing & AutomotiveHITL ### Warranty claim processing Dealer claims validated for coverage, parts and labour in minutes; big amounts go to the regional manager. 8,600 runs / month 96.3% success 4 agents Manufacturing & AutomotiveHITL ### Supplier invoice three-way match Invoice, PO and goods receipt matched in SAP automatically; only out-of-tolerance lines reach a buyer. 21,000 runs / month 97.1% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Merchant and payment operations URL: https://www.aginies.com/en/solutions/merchant-and-payment-operations Language: en Summary: At a payment institution the merchant desk, risk team and reconciliation team work separately on the same merchant; KYB takes days, settlement questions fill the call centre, exceptions wait for month-end. Fintech & Payments · Operational automation # Merchant and payment operations Merchant onboarding, POS and settlement support, reconciliation exceptions and transaction monitoring on one desk. In short Merchant and payment operations is an Aginies solution package for the Fintech & Payments sector in the “operational automation” area. What the agents do: KYB: trade registry, signatory and sanctions checks, risk class and contract in one flow; the operations assistant resolves POS diagnosis, settlement breakdown, refund and limit requests on WhatsApp and the portal. Expected outcomes: merchant issues resolved at first contact, on one channel; KYB within hours; the risk team only on exceptions; reconciliation exceptions no longer wait for month-end. The package uses the KYB & sanctions tools, Terminal & settlement tools, Policy engine, Human-in-the-loop capabilities of the Aginies platform and can be watched live through the Merchant operations assistant: POS, settlement, refund and limit, Merchant KYB onboarding, Payment reconciliation exceptions, AML transaction monitoring alert triage scenarios in the agent library. The problem At a payment institution the merchant desk, risk team and reconciliation team work separately on the same merchant; KYB takes days, settlement questions fill the call centre, exceptions wait for month-end. What the agents do - 01KYB: trade registry, signatory and sanctions checks, risk class and contract in one flow. - 02The operations assistant resolves POS diagnosis, settlement breakdown, refund and limit requests on WhatsApp and the portal. - 03Reconciliation exceptions matched and explained; AML alerts presented to the risk team with evidence. - 04BNPL and instalment decisions through the policy engine; out-of-band requests always to a person. Package scope - Merchant onboarding (KYB), risk class and contract - Operations assistant: POS, settlement, refund and limit in one dialogue - Reconciliation exceptions and transaction-monitoring (AML) alerts - BNPL and instalment decisions with the policy engine What you need to start - API access to terminal management, settlement ledger and refund service - Risk policy and rules per central-bank and AML obligations - WhatsApp Business and merchant-portal channels Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Merchant SupervisorMerchant Onboarding SupervisorReconciliation SupervisorAML Supervisor Specialist agents Dialogue AgentPOS AgentOperations AgentRisk AgentBusiness Identity AgentUBO AgentContract AgentMatching AgentInvestigation AgentCorrection AgentEnrichment AgentScreening AgentNarrative AgentFiling Agent Models (Hub catalogue) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-konrul-1 Systems touched WhatsApp BusinessTerminal ManagementSettlement LedgerRefund ServiceTicketingMerchant PortalTrade Registry (MERSİS)Sanctions Screeninge-SignEmailPSP / Acquirer APICard NetworkCore LedgerCase ManagementMASAK Reporting Portal Channels whatsappwebvoiceemailbatchapi Simulation volumes 74,520runs / month 94.0%success rate 8.1 smedian duration $0.0043estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Merchant issues resolved at first contact, on one channel - KYB within hours; the risk team only on exceptions - Reconciliation exceptions no longer wait for month-end Capabilities used KYB & sanctions toolsTerminal & settlement toolsPolicy engineHuman-in-the-loopWhatsApp BusinessFull audit trail Explore the platform → Measured targets | Merchant support first-contact resolution | 70% and above | | KYB completion time | From days to hours | | Reconciliation exception closure time | 60% shorter | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Merchant operations assistant: POS, settlement, refund and limitRefund > 25,000 or limit increase > 50% → risk team approval - Merchant KYB onboardingRisk tier high or UBO fuzzy match > 0.65 → compliance manager approval - Payment reconciliation exceptionsSingle break > 10,000 or total > 50,000 → finance manager approval before adjustment - AML transaction monitoring alert triageEvery alert → compliance analyst decision mandatory; the agent never files an STR or closes a case alone - packageHigh refunds, limit increases and AML alerts always with the risk team - packageCommunity cloud or the institution’s data centre; deployment aligned with payment regulation Rollout - 1 · Process map · 2 weeksSignal sources, decision points and work-order systems mapped; the automatic-decision boundary and approval chain defined. - 2 · Pilot · 4 weeksOne line, one region or one process; the agent prepares the recommendation, a person applies it; deviation and gain measured. - 3 · Autonomous operationDecisions below the threshold run automatically, above it go to approval; shift reports and SLA dashboards live. Related agents ## This package’s scenarios in the agent library Fintech & PaymentsHITL ### Merchant operations assistant: POS, settlement, refund and limit The merchant writes on WhatsApp; POS fault, settlement query, refund and limit increase are resolved in one dialogue. 64,000 runs / month 93.8% success 5 agents Fintech & PaymentsHITL ### Merchant KYB onboarding Registry, UBO and sanctions checks in one flow; low-risk merchants start accepting payments the same day. 1,400 runs / month 93.4% success 5 agents Fintech & PaymentsHITL ### Payment reconciliation exceptions PSP settlement files matched to the ledger; small breaks fix themselves, large ones go to finance approval. 620 runs / month 95.1% success 4 agents Fintech & PaymentsHITL ### AML transaction monitoring alert triage Alerts arrive enriched with a draft narrative; the analyst always decides and the filing goes out in minutes. 8,500 runs / month 95.8% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Operational automation & demand management URL: https://www.aginies.com/en/solutions/operational-automation-demand Language: en Summary: Demands arrive scattered across email, portal and phone; each is read and assigned by hand. Misrouting and rework inflate cycle time and cost. Enterprise Functions · Operational automation # Operational automation & demand management Agents that classify, route and answer demands; faster processes, lower cost. In short Operational automation & demand management is an Aginies solution package for the Enterprise Functions sector in the “operational automation” area. What the agents do: Each incoming demand is classified regardless of channel: topic, urgency and owning team determined; dynamic routing sends the demand to the right queue or agent; missing information is requested from the requester. Expected outcomes: shorter resolution times, backlog cleared; lower operating cost per demand; teams freed from repetitive work to focus on exceptions. The package uses the Dynamic agent routing, Intent detection, Email service integrations, Ticketing tools (MCP) capabilities of the Aginies platform and can be watched live through the Line stoppage response: from alarm to work order and shift report, Control tower: delay prediction, rerouting, customer notice, SLA, IT helpdesk L1, Maintenance ticket triage scenarios in the agent library. The problem Demands arrive scattered across email, portal and phone; each is read and assigned by hand. Misrouting and rework inflate cycle time and cost. What the agents do - 01Each incoming demand is classified regardless of channel: topic, urgency and owning team determined. - 02Dynamic routing sends the demand to the right queue or agent; missing information is requested from the requester. - 03Standard demands (passwords, status queries, forms) resolved instantly with corporate knowledge and tools. - 04SLA risk and recurring demand patterns surface on dashboards; critical decisions stay with people. Package scope - Classification, prioritisation and assignment of requests and tickets - End-to-end handling of standard requests (access, password, form, information) - Approval chains and SLA tracking; delay alerts - First-level support with knowledge base and automation tools What you need to start - API access to the ITSM / ticketing system - Service catalogue, SLA and approval rules - Authorised access to identity management and automation tools Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Production SupervisorControl Tower SupervisorHelpdesk SupervisorMaintenance Supervisor Specialist agents Diagnosis AgentParts AgentPlanning AgentReport AgentPrediction AgentRouting AgentCustomer AgentFinance AgentIntent AgentAccess AgentTicket AgentSignal AgentField Note Agent Models (Hub catalogue) aginies-tulpar-4aginies-sunkar-1aginies-talay-2aginies-voice-tr-v3 Systems touched MESSCADA HistorianCMMSERP Spare PartsShift Channel (Teams)TMSTelematicsCarrier APICustomer PortalERP BillingActive Directory / Entra IDServiceNowMicrosoft TeamsSCADA / HistorianCMMS (Maximo)SAP S/4HANA MMManuals & Work Order Vector Store Channels apiappemailwhatsappvoiceweb Simulation volumes 149,500runs / month 95.0%success rate 7.9 smedian duration $0.0023estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Shorter resolution times, backlog cleared - Lower operating cost per demand - Teams freed from repetitive work to focus on exceptions Capabilities used Dynamic agent routingIntent detectionEmail service integrationsTicketing tools (MCP)Agent templatingAudit trail logging Explore the platform → Measured targets | Requests closed automatically | 45–60% | | Mean time to resolve | 50% shorter | | SLA breaches | 60% fewer | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Line stoppage response: from alarm to work order and shift reportSchedule change > 2 hours or urgent purchase above 50,000 → shift lead approval - Control tower: delay prediction, rerouting, customer notice, SLARoute cost increase > 15% or compensation > 10,000 → operations manager approval - IT helpdesk L1Access to a finance or production group → group owner approval - Maintenance ticket triagePlanned stop > 4 hours → shift supervisor approval - packagePrivilege elevation and production access always approved - packageEvery automatic action recorded on the ticket with who/what/when Rollout - 1 · Process map · 2 weeksSignal sources, decision points and work-order systems mapped; the automatic-decision boundary and approval chain defined. - 2 · Pilot · 4 weeksOne line, one region or one process; the agent prepares the recommendation, a person applies it; deviation and gain measured. - 3 · Autonomous operationDecisions below the threshold run automatically, above it go to approval; shift reports and SLA dashboards live. Related agents ## This package’s scenarios in the agent library Manufacturing & AutomotiveHITL ### Line stoppage response: from alarm to work order and shift report An MES alarm lands; root cause found from history, spare part checked, work order and plan update ready within minutes. 3,100 runs / month 92.2% success 5 agents LogisticsHITL ### Control tower: delay prediction, rerouting, customer notice, SLA The telematics signal catches the delay early; alternative route and carrier chosen, customer informed, SLA compensation computed. 96,000 runs / month 93.4% success 5 agents Enterprise FunctionsHITL ### IT helpdesk L1 Password resets and access requests solved in chat; access to finance groups asks the owner via a Teams card. 47,000 runs / month 98.5% success 4 agents Manufacturing & AutomotiveHITL ### Maintenance ticket triage Sensor alarm and technician voice note become one work order, spare parts and a safe stop window included. 3,400 runs / month 95.2% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Public procurement & tender compliance URL: https://www.aginies.com/en/solutions/public-procurement-compliance Language: en Summary: Hundreds of documents and dozens of specification clauses are compared by hand; gaps surface in the commission meeting and appeal risk grows. Government & Public Sector · Data protection & compliance # Public procurement & tender compliance Bid files compared with the specification; the commission decides on a prepared findings list. In short Public procurement & tender compliance is an Aginies solution package for the Government & Public Sector sector in the “data protection & compliance” area. What the agents do: Bid documents are classified and read: guarantee, qualification, experience, technical annexes, e-signature; every bid is matched with administrative and technical clauses; non-conformity candidates are flagged. Expected outcomes: commission preparation drops from days to hours; consistent, justified, appeal-resistant evaluation; every finding linked to its source document, audit-ready. The package uses the Vision extraction, Document classification, Clause matching, Explainable findings capabilities of the Aginies platform and can be watched live through the Public procurement bid file pre-check, Contract clause extraction, Supplier security and compliance assessment scenarios in the agent library. The problem Hundreds of documents and dozens of specification clauses are compared by hand; gaps surface in the commission meeting and appeal risk grows. What the agents do - 01Bid documents are classified and read: guarantee, qualification, experience, technical annexes, e-signature. - 02Every bid is matched with administrative and technical clauses; non-conformity candidates are flagged. - 03Checklist and minutes draft are produced with clause numbers. - 04Conformity and elimination decisions stay with the tender commission; ambiguous clauses are flagged to it. Package scope - Comparison of bid files with the specification and regulation - Findings on missing documents, non-compliant bids and conflicts of interest - Findings list and reasoning draft for the commission - Tender calendar and deadline tracking What you need to start - Read access to the e-procurement platform and the institution’s tender files - Specification templates and evaluation criteria - Commission members and approval flow Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Tender SupervisorContract SupervisorSupplier Security Supervisor Specialist agents Document AgentSpecification AgentMinutes AgentClause AgentPlaybook AgentCLM AgentAssessment AgentCase Agent Models (Hub catalogue) aginies-konrul-1aginies-tulpar-4aginies-talay-2 Systems touched e-Procurement PlatformSpecification Vector StoreDocument StoreCLM (Contract Lifecycle Management)Legal Playbook Vector StoreDocument Store (S3)Exchange / M365 MailSupplier Management (SRM)Security Requirements (vector)E-mail Channels batchwebemailapi Simulation volumes 2,680runs / month 95.1%success rate 8 smedian duration $0.0177estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Commission preparation drops from days to hours - Consistent, justified, appeal-resistant evaluation - Every finding linked to its source document, audit-ready Capabilities used Vision extractionDocument classificationClause matchingExplainable findingsHuman-in-the-loopOn-prem models Explore the platform → Measured targets | File pre-review time | 70% shorter | | Missed missing documents | Near zero | | Appeals against commission decisions | Fewer | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Public procurement bid file pre-checkConformity and elimination decisions always with the tender commission; the agent only prepares - Contract clause extractionHigh-risk deviation or contract value > 500,000 € → counsel approval - Supplier security and compliance assessmentThe qualification decision always sits with the facility security manager; the agent only prepares - packageElimination and award decisions only with the commission; the agent produces findings - packageEvery finding with a document and article reference; audit-ready Rollout - 1 · Rule and control design · 2–3 weeksRegulatory articles translated into the control set, the decision matrix approved with compliance, trace requirements written. - 2 · Shadow period · 4 weeksThe agent produces findings and recommendations, the expert decides; disagreement rate and reasoning quality measured. - 3 · Audit-ready operationControl evidence collected automatically; audit reports and the exception register opened to internal audit. Related agents ## This package’s scenarios in the agent library Government & Public SectorHITL ### Public procurement bid file pre-check Bid files are compared with the specification, missing and non-conforming documents listed; the evaluation decision belongs to the commission. 380 runs / month 94.8% success 4 agents Enterprise FunctionsHITL ### Contract clause extraction Clauses are extracted and compared with the legal playbook; deviations reach counsel with redline suggestions. 1,400 runs / month 95.1% success 4 agents Defence & AerospaceHITL ### Supplier security and compliance assessment Supplier questionnaires and documents are read, compared against facility security requirements, gaps listed; the qualification decision belongs to the security manager. 900 runs / month 95.2% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Sales & lead management URL: https://www.aginies.com/en/solutions/sales-lead-management Language: en Summary: Leads pile up in the CRM, yet sales cannot tell which are warm; campaigns go to everyone alike. Price and gift suggestions rely on gut feel, and opportunities go cold. Retail & E-commerce · Customer interaction & experience # Sales & lead management The right offer at the right time: lead scoring, campaign and price recommendations. In short Sales & lead management is an Aginies solution package for the Retail & E-commerce sector in the “customer interaction & experience” area. What the agents do: Lead AI combines web, form and CRM signals; each lead is prioritised with an intent and fit score; the campaign agent produces personalised offers and gift suggestions from segment and past behaviour. Expected outcomes: sales focuses on the leads most likely to convert; personalised campaigns lift conversion and basket size; contact before the lead cools: every scored lead is called the same day. The package uses the Lead scoring, Recommendation engine, CRM tools (MCP), Multi-source RAG capabilities of the Aginies platform and can be watched live through the Sales lead qualification, Plan upgrade advisor, Review response and quality escalation, Product Q&A assistant scenarios in the agent library. The problem Leads pile up in the CRM, yet sales cannot tell which are warm; campaigns go to everyone alike. Price and gift suggestions rely on gut feel, and opportunities go cold. What the agents do - 01Lead AI combines web, form and CRM signals; each lead is prioritised with an intent and fit score. - 02The campaign agent produces personalised offers and gift suggestions from segment and past behaviour. - 03The price-analysis agent reads competitor and stock data and proposes prices within margin rules. - 04Warm leads are handed to a rep with a conversation summary; offers above the discount threshold go to approval. Package scope - Scoring and enrichment of leads from web, e-mail and WhatsApp - CRM record, representative assignment and first-contact draft - Offer and plan suggestion from the product knowledge base - Follow-up reminders and lost-opportunity analysis What you need to start - API access to the CRM (Salesforce, HubSpot, Dynamics) - Scoring criteria and territory/representative assignment rules - Current product, price and campaign information Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Sales SupervisorPlan SupervisorVoice-of-Customer SupervisorShopping Supervisor Specialist agents Enrichment AgentScoring AgentCRM AgentUsage AgentOffer AgentProvisioning AgentClassification AgentResponse AgentQuality AgentIntent AgentCatalogue AgentReply Agent Models (Hub catalogue) aginies-tulpar-4aginies-talay-2aginies-sunkar-1 Systems touched HubSpotEnrichment APIMicrosoft TeamsBilling (Amdocs)Product Catalogue / Offer EngineHLR / ProvisioningWhatsApp Business APISMS GatewayReview Platform APIShopify / OMSJira (QA)Vector StorePIM / CatalogueWeb Chat Widget Channels webapiemailappwhatsappvoice Simulation volumes 213,600runs / month 97.8%success rate 5.4 smedian duration $0.0007estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Sales focuses on the leads most likely to convert - Personalised campaigns lift conversion and basket size - Contact before the lead cools: every scored lead is called the same day Capabilities used Lead scoringRecommendation engineCRM tools (MCP)Multi-source RAGHuman-in-the-loopSession memory Explore the platform → Measured targets | Time to first contact | From hours to minutes | | Qualified-lead conversion | +10–20 points | | CRM data completeness | 95% and above | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Sales lead qualificationEnterprise (> 1,000 employees) or suspected competitor → SDR lead acceptance - Plan upgrade advisorCommitment-based or discounted bespoke offer → campaign manager approval - Review response and quality escalation1-star review replies and defect escalations → community manager approval - packagePrice and discount offers follow the authority matrix; out-of-band offers to a person - packageNo marketing contact without consent; the consent record checked on every contact Rollout - 1 · Discovery and design · 2–3 weeksThe 20 most frequent intents and transactions are chosen, identity and approval thresholds written with information security, channel and system integrations listed. - 2 · Pilot · 4–6 weeksOne channel, a limited customer group; the test corpus runs, a human representative shadows, thresholds tuned on real traffic. - 3 · Scale-outRemaining channels and transactions added; dashboards, SIEM forwarding and kill-switch handed over; monthly model and rule review begins. Related agents ## This package’s scenarios in the agent library Enterprise FunctionsHITL ### Sales lead qualification Inbound leads enriched, scored and in the CRM within seconds; the SDR starts the call with a context brief. 15,600 runs / month 97.9% success 4 agents TelecomHITL ### Plan upgrade advisor Usage analysed, the right plan offered with a comparison; on acceptance the change provisions instantly. 52,000 runs / month 95.8% success 4 agents Retail & E-commerceHITL ### Review response and quality escalation Reviews classified and answered in brand tone; recurring defects joined with return data and escalated to QA. 26,000 runs / month 97.2% success 4 agents Retail & E-commerce ### Product Q&A assistant Size, compatibility and spec questions answered from catalogue and return data; stock and price verified. 120,000 runs / month 98.7% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # SDLC agent team URL: https://www.aginies.com/en/solutions/sdlc-agent-team Language: en Summary: Software teams lose time in hand-offs between request, analysis, design, coding, testing and deployment. Code-generating assistants exist; a team that owns the process from ticket to production with defined approval points does not. Software Delivery · Software development & technology productivity # SDLC agent team PM, analyst, architect, developer and tester agents working together in the CI/CD pipeline. In short SDLC agent team is an Aginies solution package for the Software Delivery sector in the “software development & technology productivity” area. What the agents do: The Project Manager Agent triages requests, the Analyst Agent writes acceptance criteria, the Architect Agent produces the design and impact; the Developer Agent implements on a branch; the Tester Agent generates and runs unit and integration tests and reports coverage. Expected outcomes: lead time from request to production drops from days to hours; standard, measurable quality: coverage, security and regression on every PR; engineers focus on direction and approval; agents take the repetitive work. The package uses the Supervisor · sub-agents, Tools / MCP (Git, CI, Jira), Human-in-the-loop, Stateful execution capabilities of the Aginies platform and can be watched live through the Legacy module modernisation: analysis, tests, rewrite, PR, Feature delivery pipeline, CI/CD release pipeline, Incident response and hotfix, Code review and security gate scenarios in the agent library. The problem Software teams lose time in hand-offs between request, analysis, design, coding, testing and deployment. Code-generating assistants exist; a team that owns the process from ticket to production with defined approval points does not. What the agents do - 01The Project Manager Agent triages requests, the Analyst Agent writes acceptance criteria, the Architect Agent produces the design and impact. - 02The Developer Agent implements on a branch; the Tester Agent generates and runs unit and integration tests and reports coverage. - 03SAST, dependency scanning and staging smoke tests run automatically in CI/CD; production deployment goes to release-manager approval. - 04Every step is traced in Jira, Git and CI; code review and the production decision always stay with people. Package scope - PM, analyst, architect, developer and test agents on Jira, Git and CI/CD - From requirement to PR: analysis note, design decision, code, tests, review - In-IDE completion with Code Aginies; code never leaves the organisation - Incident response and hotfix lane; release notes What you need to start - Service accounts and permission model for Jira, Git and CI/CD - Coding standards, review rules and definition of done - Scope boundaries: which repositories, which environments Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Modernisation SupervisorDelivery SupervisorRelease SupervisorIncident SupervisorReview Supervisor Specialist agents Analysis AgentTest AgentDeveloper AgentReview AgentAnalyst AgentArchitect AgentTester AgentBuild AgentSecurity AgentRelease AgentPost-mortem AgentReviewer Agent Models (Hub catalogue) aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Systems touched JiraGit (self-hosted)CI PipelineStatic AnalysisGitHubCI Runner (GitHub Actions)ConfluenceSonarQube + SnykKubernetes / ArgoCDSlackObservability (Grafana / Datadog)SonarQubeDependency Scanner (Snyk) Channels apibatchweb Simulation volumes 8,570runs / month 92.6%success rate 16 smedian duration $0.0144estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Lead time from request to production drops from days to hours - Standard, measurable quality: coverage, security and regression on every PR - Engineers focus on direction and approval; agents take the repetitive work Capabilities used Supervisor · sub-agentsTools / MCP (Git, CI, Jira)Human-in-the-loopStateful executionFull audit trailPrompt versioning Explore the platform → Measured targets | PRs accepted on first review | 60–75% | | Requirement-to-PR lead time | 40–50% shorter | | Test coverage | +15–25 points | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Legacy module modernisation: analysis, tests, rewrite, PRProduction schema change or out-of-scope file → architect approval; PR merge always human - Feature delivery pipelineEvery PR → at least 1 human code review approval; agents never push to main directly - CI/CD release pipelineEvery production deploy → release-manager approval mandatory; a critical CVE closes the gate automatically - Incident response and hotfixEvery emergency production deploy → on-call engineer approval; the hotfix PR still gets a human review - Code review and security gateEvery PR → human CODEOWNERS approval; a high SAST finding or coverage < 80% blocks the gate - packagePR merge and production deployment always under human approval - packageThe agent writes only in assigned repositories and branches; secret scanner on every PR Rollout - 1 · Pipeline integration · 2 weeksJira, Git, CI/CD or device management connected; permission model and scope boundaries written. - 2 · Pilot · 4–6 weeksOne team or one device segment; every output passes human review, acceptance rate measured. - 3 · ScaleRemaining teams and segments; automated steps widen, production and architecture decisions stay with people. Related agents ## This package’s scenarios in the agent library Software DeliveryHITL ### Legacy module modernisation: analysis, tests, rewrite, PR Starts from a Jira epic; the old module is read, its behaviour pinned with tests, rewritten and opened as a PR with evidence. 1,900 runs / month 90.4% success 5 agents Software DeliveryHITL ### Feature delivery pipeline A Jira epic turns into criteria, design, code and tests via agents; humans only review and approve the PR. 1,200 runs / month 91.2% success 5 agents Software DeliveryHITL ### CI/CD release pipeline Build, tests, scans and staging run unattended; production ships only on release-manager approval, via ArgoCD. 950 runs / month 94.7% success 5 agents Software DeliveryHITL ### Incident response and hotfix Alert fires, bug reproduced, patch and regression run ready; on-call approves, post-mortem already drafted. 320 runs / month 90.5% success 5 agents Software DeliveryHITL ### Code review and security gate Agents review, scan and coverage-check every PR; fixes land first, a human approves only clean PRs. 4,200 runs / month 93.6% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Subscriber care and network support URL: https://www.aginies.com/en/solutions/telecom-subscriber-care Language: en Summary: A subscriber talks about bill, plan and fault in the same call; the IVR tree and representative screens cannot resolve it in one session, and technicians are often dispatched for nothing. Telecom · Customer interaction & experience # Subscriber care and network support Bill, plan, contract and fault operations on voice and text channels, inside BSS and network systems. In short Subscriber care and network support is an Aginies solution package for the Telecom sector in the “customer interaction & experience” area. What the agents do: Intents separated; bill lines explained, correction and plan change applied in the BSS; line and modem diagnosed remotely; a technician is dispatched only if firmware and restart do not fix it. Expected outcomes: multi-topic calls closed in one session without a representative; A clear drop in unnecessary technician visits; shorter handling time, higher satisfaction. The package uses the BSS / OSS tools (MCP), SIP voice, Supervisor · sub-agents, Policy engine capabilities of the Aginies platform and can be watched live through the Subscriber care assistant: bill, plan, fault and contract, Bill dispute resolution, Plan upgrade advisor, SIM swap fraud check, Network outage communications scenarios in the agent library. The problem A subscriber talks about bill, plan and fault in the same call; the IVR tree and representative screens cannot resolve it in one session, and technicians are often dispatched for nothing. What the agents do - 01Intents separated; bill lines explained, correction and plan change applied in the BSS. - 02Line and modem diagnosed remotely; a technician is dispatched only if firmware and restart do not fix it. - 03Contract and exit terms calculated, the eligible offer presented; churn intent passed to the retention team. - 04SIM swaps and suspicious requests pass fraud checks; outage information sent proactively. Package scope - Bill disputes, plan changes and contract operations on the BSS - Line and modem diagnosis, remote repair, technician dispatch - SIM-swap fraud checks and outage communication - The same assistant on IVR, app and WhatsApp; warm hand-off to a representative What you need to start - API access to BSS/billing and OSS/network systems - Tariff, contract and exit-fee rules - IVR/SIP and digital channel integration Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Subscriber Care SupervisorBilling SupervisorPlan SupervisorSIM Security SupervisorOutage Communications Supervisor Specialist agents Dialogue AgentAccount AgentNetwork AgentRetention AgentIntent AgentBill Analysis AgentAdjustment AgentUsage AgentOffer AgentProvisioning AgentRisk AgentVerification AgentDetection AgentImpact AgentCommunications Agent Models (Hub catalogue) aginies-talay-2aginies-sunkar-1aginies-tulpar-4 Systems touched IVR / SIPBSS / BillingOSS NetworkCRMSMS GatewayBilling (Amdocs)Usage Mediation / CDR StoreProduct Catalogue / Offer EngineHLR / ProvisioningWhatsApp Business APISalesforce CRMBank SIM-Swap Signal APIServiceNowNetwork OSS Channels voiceappwhatsappbranchapismsbatch Simulation volumes 592,400runs / month 95.0%success rate 7.3 smedian duration $0.0016estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Multi-topic calls closed in one session without a representative - A clear drop in unnecessary technician visits - Shorter handling time, higher satisfaction Capabilities used BSS / OSS tools (MCP)SIP voiceSupervisor · sub-agentsPolicy engineHuman-in-the-loopFull audit trail Explore the platform → Measured targets | Requests resolved without a representative | 60–70% | | Faults closed without a technician | 35–45% | | Average handling time | 40% shorter | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Subscriber care assistant: bill, plan, fault and contractContract cancellation, bill correction above 500 or technician priority → representative approval - Bill dispute resolutionGoodwill credit > 75 → team lead approval - Plan upgrade advisorCommitment-based or discounted bespoke offer → campaign manager approval - SIM swap fraud checkRisk score ≥ 70 or credential change within 72 hours → fraud desk approval - Network outage communicationsAffected subscribers > 10,000 or P1 outage → network operations manager approval - packageContract cancellation, large bill corrections and SIM swaps under human approval - packageSubscriber data processed per telecom-regulator obligations; every operation leaves a trace Rollout - 1 · Discovery and design · 2–3 weeksThe 20 most frequent intents and transactions are chosen, identity and approval thresholds written with information security, channel and system integrations listed. - 2 · Pilot · 4–6 weeksOne channel, a limited customer group; the test corpus runs, a human representative shadows, thresholds tuned on real traffic. - 3 · Scale-outRemaining channels and transactions added; dashboards, SIEM forwarding and kill-switch handed over; monthly model and rule review begins. Related agents ## This package’s scenarios in the agent library TelecomHITL ### Subscriber care assistant: bill, plan, fault and contract The subscriber calls or writes; bill dispute, plan change and line fault are resolved in BSS and network systems in one session. 410,000 runs / month 94.3% success 5 agents TelecomHITL ### Bill dispute resolution Invoice lines and usage explained in seconds; goodwill credit applied by policy, larger amounts need a lead. 88,000 runs / month 96.3% success 4 agents TelecomHITL ### Plan upgrade advisor Usage analysed, the right plan offered with a comparison; on acceptance the change provisions instantly. 52,000 runs / month 95.8% success 4 agents TelecomHITL ### SIM swap fraud check Every SIM swap is risk-scored before provisioning; suspicious requests are held and banks are signalled. 41,000 runs / month 97.7% success 4 agents TelecomHITL ### Network outage communications When an outage ticket opens, affected subscribers are found and the approved message sent before they call. 1,400 runs / month 98.5% success 4 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Underwriting and renewal operations URL: https://www.aginies.com/en/solutions/underwriting-and-renewal Language: en Summary: Renewal offers are prepared by hand, documents arrive incomplete, underwriters lose time on simple files; when the customer is called late, the policy goes to a competitor. Insurance · Financial analysis & operational intelligence # Underwriting and renewal operations Quote documents read, risk variables extracted, renewal offers prepared within the rules. In short Underwriting and renewal operations is an Aginies solution package for the Insurance sector in the “financial analysis & operational intelligence” area. What the agents do: Risk variables extracted from quote and renewal documents; missing items requested from the customer; offer prepared with the pricing engine and underwriting rules; sent if within the authority limit. Expected outcomes: most renewals automatic, offers on time; underwriter time spent on exceptional risks; A measurable rise in the renewal rate. The package uses the Document extraction (Vision), Pricing engine tools, Policy engine, Human-in-the-loop capabilities of the Aginies platform and can be watched live through the Policy renewal offer, Underwriting document extraction, Voice claims intake (FNOL) scenarios in the agent library. The problem Renewal offers are prepared by hand, documents arrive incomplete, underwriters lose time on simple files; when the customer is called late, the policy goes to a competitor. What the agents do - 01Risk variables extracted from quote and renewal documents; missing items requested from the customer. - 02Offer prepared with the pricing engine and underwriting rules; sent if within the authority limit. - 03Renewal reminder, offer and missing-document communication on the customer’s preferred channel. - 04Above-limit and exceptional risks land in the underwriter queue with a decision pack. Package scope - Extraction of quote and renewal documents; risk variables - Renewal offer and price proposal within underwriting rules - Customer communication: renewal reminder, offer, missing document - Decision pack for the underwriter and an exception queue What you need to start - API access to policy administration and the pricing engine - Underwriting rules and authority limits - Sample quote and renewal documents Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Renewal SupervisorUnderwriting SupervisorClaims Supervisor Specialist agents Pricing AgentOffer AgentIssuance AgentDocument AgentMedical Assessment AgentUnderwriting AgentSpeech AgentIntake AgentPolicy AgentClaims Registration Agent Models (Hub catalogue) aginies-talay-2aginies-tulpar-4aginies-sunkar-1aginies-konrul-1aginies-voice-tr-v3 Systems touched Policy Admin SystemRating EngineWhatsApp Business APIEmail (Exchange)Underwriting WorkbenchDocument Store (S3)Vector StoreGenesys SIPGuidewire ClaimCenter Channels batchwhatsappemailappwebvoice Simulation volumes 60,800runs / month 95.2%success rate 6.2 smedian duration $0.003estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Most renewals automatic, offers on time - Underwriter time spent on exceptional risks - A measurable rise in the renewal rate Capabilities used Document extraction (Vision)Pricing engine toolsPolicy engineHuman-in-the-loopMulti-channel messagingFull audit trail Explore the platform → Measured targets | Policies renewed automatically | 60–75% | | Quote preparation time | 60% shorter | | Renewal rate | +5–8 points | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Policy renewal offerPremium increase > 20% or discount > 10% → renewal team lead approval - Underwriting document extractionLoading ≥ 50% or sum assured > 5,000,000 → senior underwriter approval - Voice claims intake (FNOL)Injury or reserve > 250,000 → warm transfer to a claims specialist - packagePrices above the authority limit and exceptional risks always under underwriter approval - packagePrice reasoning and variables used stored in the policy file Rollout - 1 · Documents and rule set · 2 weeksDocument types, field dictionary, matching and tolerance rules written with the finance team; a sample set labelled. - 2 · Parallel run · 4 weeksAgent and current process handle the same documents; the difference report reviewed weekly, cut-over when the accuracy threshold is met. - 3 · Close integrationTied to the month-end calendar; exception queue, audit trail and reconciliation reports opened to internal audit. Related agents ## This package’s scenarios in the agent library InsuranceHITL ### Policy renewal offer Policies re-priced, sharp increases go to approval; customer accepts with one tap and the policy is issued. 22,000 runs / month 94.4% success 4 agents InsuranceHITL ### Underwriting document extraction Medical reports and labs read and assessed against the manual; underwriters only see files above threshold. 4,800 runs / month 94.9% success 4 agents InsuranceHITL ### Voice claims intake (FNOL) The customer calls and describes it; policy found by plate, cover checked, claim opened and adjuster assigned. 34,000 runs / month 95.8% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Utility customer operations URL: https://www.aginies.com/en/solutions/utility-customer-operations Language: en Summary: Customer channels and grid systems are disconnected: hundreds of calls in an outage, a backlog of estimated-bill disputes, engineer time spent on document checks for rooftop solar. Energy & Utilities · Customer interaction & experience # Utility customer operations Outages, meters, connections and demand response: customer agents that speak with grid data. In short Utility customer operations is an Aginies solution package for the Energy & Utilities sector in the “customer interaction & experience” area. What the agents do: Outage reports are clustered with SCADA and feeder topology; everyone affected hears before the crew leaves; the meter photo is read by a vision model, regulation checked, the bill corrected. Expected outcomes: call volume drops in outages, satisfaction rises; bill disputes and connection applications resolved within hours; the flexibility portfolio turns into revenue. The package uses the Voice (STT/TTS), Vision extraction, Multi-source RAG, REST / MCP tools capabilities of the Aginies platform and can be watched live through the Utility customer assistant: bill, outage, consumption and instalments in one dialogue, Outage report handling, Meter reading dispute, Rooftop solar connection application scenarios in the agent library. The problem Customer channels and grid systems are disconnected: hundreds of calls in an outage, a backlog of estimated-bill disputes, engineer time spent on document checks for rooftop solar. What the agents do - 01Outage reports are clustered with SCADA and feeder topology; everyone affected hears before the crew leaves. - 02The meter photo is read by a vision model, regulation checked, the bill corrected. - 03Connection applications are pre-assessed for documents, capacity and regulation; the opinion leaves with the engineer’s signature. - 04In demand-response events participants are selected, offers sent, reduction measured, settlement computed. Package scope - Bill explanation, dispute and instalments; with smart-meter data - Outage information and ETA; OMS integration - Connection, subscription and rooftop-solar applications - IVR, WhatsApp and app channel; warm hand-off to a representative What you need to start - API access to billing (CIS) and outage management (OMS) - Smart-meter data access - Tariff, instalment and dispute policies Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Customer Service SupervisorOutage SupervisorBilling Dispute SupervisorConnection Supervisor Specialist agents Dialogue AgentBilling AgentGrid AgentCollections AgentVoice AgentReport AgentClustering AgentComms AgentVision AgentOperations AgentDocument AgentRegulation AgentCase Agent Models (Hub catalogue) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-voice-tr-v3aginies-konrul-1 Systems touched IVR / SIPBilling (CIS)Outage ManagementSmart Meter DataSMS GatewayCustomer CRM / BillingSCADA / HistorianNetwork OSSWhatsApp Business APIBilling / CIS (SAP IS-U)Meter Data ManagementRegulation Vector Store (EPDK)Connection PortalGIS / Network ModelE-mail Channels voicewhatsappappwebemailapi Simulation volumes 572,600runs / month 94.8%success rate 6.5 smedian duration $0.0016estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Call volume drops in outages, satisfaction rises - Bill disputes and connection applications resolved within hours - The flexibility portfolio turns into revenue Capabilities used Voice (STT/TTS)Vision extractionMulti-source RAGREST / MCP toolsHuman-in-the-loopBatch events Explore the platform → Measured targets | Calls resolved without a representative | 55–65% | | Bill dispute resolution time | From days to minutes | | Call load during outages | 50% lower | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Utility customer assistant: bill, outage, consumption and instalments in one dialogueInstalments > 6 months, bill cancellation or meter suspicion → customer service approval - Outage report handlingMass notification > 1,000 customers → control room dispatcher approval - Meter reading disputeCredit / correction > 1,000 TL → billing team lead approval - Rooftop solar connection applicationThe connection opinion is always signed by a network planning engineer - packageBill cancellation and long instalment plans always under customer-service approval - packageMeter data used only for the customer’s own account Rollout - 1 · Discovery and design · 2–3 weeksThe 20 most frequent intents and transactions are chosen, identity and approval thresholds written with information security, channel and system integrations listed. - 2 · Pilot · 4–6 weeksOne channel, a limited customer group; the test corpus runs, a human representative shadows, thresholds tuned on real traffic. - 3 · Scale-outRemaining channels and transactions added; dashboards, SIEM forwarding and kill-switch handed over; monthly model and rule review begins. Related agents ## This package’s scenarios in the agent library Energy & UtilitiesHITL ### Utility customer assistant: bill, outage, consumption and instalments in one dialogue The customer calls or writes; the high bill is explained, outage ETA given, meter data shown, an instalment plan set up. 520,000 runs / month 94.6% success 5 agents Energy & UtilitiesHITL ### Outage report handling Reports are clustered by feeder, one OSS ticket is raised and everyone affected hears before the crew leaves. 38,000 runs / month 97.7% success 5 agents Energy & UtilitiesHITL ### Meter reading dispute Customer sends a meter photo; the index is read by vision, the estimated bill fixed, big credits approved. 9,200 runs / month 94.9% success 4 agents Energy & UtilitiesHITL ### Rooftop solar connection application Application documents are read, transformer capacity and regulation checked, gaps reported the same day; the connection opinion belongs to the engineer. 5,400 runs / month 95.3% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Voice analysis & quality URL: https://www.aginies.com/en/solutions/voice-analysis-quality Language: en Summary: Millions of call minutes go to the archive unheard; customer sentiment, intent and compliance breaches stay invisible. Cloud STT services fall short in Turkish and move voice data outside the organisation. Telecom · Customer interaction & experience # Voice analysis & quality 100% of calls transcribed with Turkish STT, scored and coached; models run on-prem. In short Voice analysis & quality is an Aginies solution package for the Telecom sector in the “customer interaction & experience” area. What the agents do: Voice Aginies models transcribe 100% of calls in Turkish on-prem; personal data is masked; sentiment, intent and topic extracted from every call; customer vulnerability and churn signals flagged. Expected outcomes: measurement across all calls instead of sampling; compliance breaches and customer risk visible the same day; voice data never leaves the organisation. The package uses the Speech-to-text (Turkish), On-prem models, Sentiment & intent, Compliance phrase checks capabilities of the Aginies platform and can be watched live through the Call centre quality monitoring, Subscriber care assistant: bill, plan, fault and contract, Dialogue banking: voice and text transaction assistant, Voice claims intake (FNOL) scenarios in the agent library. The problem Millions of call minutes go to the archive unheard; customer sentiment, intent and compliance breaches stay invisible. Cloud STT services fall short in Turkish and move voice data outside the organisation. What the agents do - 01Voice Aginies models transcribe 100% of calls in Turkish on-prem; personal data is masked. - 02Sentiment, intent and topic extracted from every call; customer vulnerability and churn signals flagged. - 03Quality forms and mandatory compliance phrases (disclosure, recording notice) scored automatically, with rationale. - 04Coaching suggestions produced per agent; training and disciplinary decisions stay with the team leader. Package scope - Call transcript, sentiment and topic analysis; history per subscriber - Compliance checks in campaign and contract communication - Early detection of complaint and churn signals, retention action - Representative coaching and quality dashboard What you need to start - Access to the switch and recording system; SIP or recording API - Campaign, contract and cancellation policies - Subscriber mapping with CRM and billing Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Quality SupervisorSubscriber Care SupervisorDialogue Banking SupervisorClaims Supervisor Specialist agents Transcription AgentScoring AgentCoaching AgentDialogue AgentAccount AgentNetwork AgentRetention AgentIdentity AgentTransaction AgentRisk AgentSpeech AgentIntake AgentPolicy AgentClaims Registration Agent Models (Hub catalogue) aginies-voice-tr-v3aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Systems touched Genesys SIPQM Scorecard / Case ManagementQA Rubric Vector StoreWorkdayIVR / SIPBSS / BillingOSS NetworkCRMSMS GatewayMobile App ChatCore BankingCard ManagementFraud EnginePolicy Admin SystemGuidewire ClaimCenterWhatsApp Business API Channels voicebatchappwhatsapp Simulation volumes 720,000runs / month 95.0%success rate 7.4 smedian duration $0.0024estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Measurement across all calls instead of sampling - Compliance breaches and customer risk visible the same day - Voice data never leaves the organisation Capabilities used Speech-to-text (Turkish)On-prem modelsSentiment & intentCompliance phrase checksSensitive data maskingDashboard analytics layerCoaching insights Explore the platform → Measured targets | Calls analysed | 100% | | Time to detect a churn signal | Same day | | Quality evaluation effort | 75% lower | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - Call centre quality monitoringScore < 60 or compliance breach → team leader review - Subscriber care assistant: bill, plan, fault and contractContract cancellation, bill correction above 500 or technician priority → representative approval - Dialogue banking: voice and text transaction assistantLimit increase > 60%, dispute amount > 5,000 or transfer to a new payee → operations approval - Voice claims intake (FNOL)Injury or reserve > 250,000 → warm transfer to a claims specialist - packageSubscriber-level actions (offer, call) with the retention team’s approval - packageRecordings and transcripts retained under personal-data rules Rollout - 1 · Discovery and design · 2–3 weeksThe 20 most frequent intents and transactions are chosen, identity and approval thresholds written with information security, channel and system integrations listed. - 2 · Pilot · 4–6 weeksOne channel, a limited customer group; the test corpus runs, a human representative shadows, thresholds tuned on real traffic. - 3 · Scale-outRemaining channels and transactions added; dashboards, SIEM forwarding and kill-switch handed over; monthly model and rule review begins. Related agents ## This package’s scenarios in the agent library Enterprise FunctionsHITL ### Call centre quality monitoring Every call scored against the rubric, not a sample; compliance breaches and coaching flags ready by morning. 90,000 runs / month 98.8% success 4 agents TelecomHITL ### Subscriber care assistant: bill, plan, fault and contract The subscriber calls or writes; bill dispute, plan change and line fault are resolved in BSS and network systems in one session. 410,000 runs / month 94.3% success 5 agents BankingHITL ### Dialogue banking: voice and text transaction assistant The customer talks or types; bill, limit, dispute and transfer are completed in core banking in one session. 186,000 runs / month 94.7% success 5 agents InsuranceHITL ### Voice claims intake (FNOL) The customer calls and describes it; policy found by plate, cover checked, claim opened and adjuster assigned. 34,000 runs / month 95.8% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Voice AI for claims intake URL: https://www.aginies.com/en/solutions/voice-claims-intake Language: en Summary: Loss notices pile up at the call centre; wait times grow while agents fill forms with the customer on the line. On storm days capacity runs out. Insurance · Customer interaction & experience # Voice AI for claims intake A voice agent takes the loss notice and opens the claim in the system. In short Voice AI for claims intake is an Aginies solution package for the Insurance sector in the “customer interaction & experience” area. What the agents do: Calls land on the voice agent through SIP; streaming speech recognition works without lag; policy, location, date, parties and loss type are extracted and confirmed within the conversation. Expected outcomes: call-centre load drops and wait times fall; hundreds of simultaneous conversations at consistent quality; claims opened at first contact with complete information. The package uses the Speech-to-text (streaming), Call-centre SIP integration, Intent detection, Stateful execution capabilities of the Aginies platform and can be watched live through the End-to-end motor claim: from notice to payment, Voice claims intake (FNOL), Dialogue banking: voice and text transaction assistant, Utility customer assistant: bill, outage, consumption and instalments in one dialogue scenarios in the agent library. The problem Loss notices pile up at the call centre; wait times grow while agents fill forms with the customer on the line. On storm days capacity runs out. What the agents do - 01Calls land on the voice agent through SIP; streaming speech recognition works without lag. - 02Policy, location, date, parties and loss type are extracted and confirmed within the conversation. - 03The claim is opened in the policy system and adjuster assignment triggers by rule. - 04Sensitive data is masked; suspicious patterns are flagged to the fraud team. Package scope - Claims and request calls taken by a voice agent at the call centre - Policy matching, incident form and missing information completed in one call - File creation in the claims system and first action (courtesy car, repair shop, adjuster) - Recording, transcript and summary; warm hand-off to a representative What you need to start - SIP integration with the call-centre switch - API access to policy and claims systems - Mandatory field list per claim type Architecture Derived from the linked scenarios: the supervisors, specialist agents, models and systems that run this package. Supervisors Claims SupervisorDialogue Banking SupervisorCustomer Service Supervisor Specialist agents Intake AgentPhoto Assessment AgentAdjuster AgentCommunication AgentSpeech AgentPolicy AgentClaims Registration AgentDialogue AgentIdentity AgentTransaction AgentRisk AgentBilling AgentGrid AgentCollections Agent Models (Hub catalogue) aginies-voice-tr-v3aginies-konrul-1aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Systems touched Call Centre SIPPolicy AdminClaims SystemRepair NetworkPayment HubGenesys SIPPolicy Admin SystemGuidewire ClaimCenterWhatsApp Business APIMobile App ChatCore BankingCard ManagementFraud EngineSMS GatewayIVR / SIPBilling (CIS)Outage ManagementSmart Meter Data Channels voiceappwhatsapp Simulation volumes 761,000runs / month 94.6%success rate 7.8 smedian duration $0.0022estimated cost / run Summed from the simulation definitions of the linked scenarios; recalculated for your scale in the discovery session. Outcomes - Call-centre load drops and wait times fall - Hundreds of simultaneous conversations at consistent quality - Claims opened at first contact with complete information Capabilities used Speech-to-text (streaming)Call-centre SIP integrationIntent detectionStateful executionSensitive data masking Explore the platform → Measured targets | Notices completed without a representative | 60–75% | | Average time per notice | 40% shorter | | Files opened with missing information | 80% fewer | Target ranges from comparable deployments; baseline and target are fixed together in the first week of the pilot. Governance and approval rules Scenario-level approval rules plus the package’s general controls. - End-to-end motor claim: from notice to paymentEstimate > 150,000, unclear fault share or overlap with prior damage → adjuster approval - Voice claims intake (FNOL)Injury or reserve > 250,000 → warm transfer to a claims specialist - Dialogue banking: voice and text transaction assistantLimit increase > 60%, dispute amount > 5,000 or transfer to a new payee → operations approval - Utility customer assistant: bill, outage, consumption and instalments in one dialogueInstalments > 6 months, bill cancellation or meter suspicion → customer service approval - packageInjury, major loss and fraud suspicion transferred to a person immediately - packageRecording retention and access according to insurance regulation Rollout - 1 · Discovery and design · 2–3 weeksThe 20 most frequent intents and transactions are chosen, identity and approval thresholds written with information security, channel and system integrations listed. - 2 · Pilot · 4–6 weeksOne channel, a limited customer group; the test corpus runs, a human representative shadows, thresholds tuned on real traffic. - 3 · Scale-outRemaining channels and transactions added; dashboards, SIEM forwarding and kill-switch handed over; monthly model and rule review begins. Related agents ## This package’s scenarios in the agent library InsuranceHITL ### End-to-end motor claim: from notice to payment Voice notice, photo assessment, repair-shop assignment and payment order in one flow; the adjuster steps in only above the threshold. 21,000 runs / month 91.2% success 5 agents InsuranceHITL ### Voice claims intake (FNOL) The customer calls and describes it; policy found by plate, cover checked, claim opened and adjuster assigned. 34,000 runs / month 95.8% success 5 agents BankingHITL ### Dialogue banking: voice and text transaction assistant The customer talks or types; bill, limit, dispute and transfer are completed in core banking in one session. 186,000 runs / month 94.7% success 5 agents Energy & UtilitiesHITL ### Utility customer assistant: bill, outage, consumption and instalments in one dialogue The customer calls or writes; the high bill is explained, outage ETA given, meter data shown, an instalment plan set up. 520,000 runs / month 94.6% success 5 agents ## Time to move from experimenting with AI to transforming with it. In a 30-minute discovery session we take your 2–3 priority business problems, show a live demo of a similar scenario, and draft a roadmap that starts with the Value layer. Book a 30-minute discovery session → dahi@aginies.com - Your 2–3 priority problems - Live demo of a similar scenario - Roadmap starting with value analysis --- # Aginies — Yapay zekâyı pilottan üretime taşıyan orkestrasyon katmanı URL: https://www.aginies.com/tr Language: tr Summary: Kurumsal agentic AI orkestrasyon platformu. Supervisor ajanlar planlar, uzman ajanlar çekirdek sistemlerde aksiyon alır; her adım izlenir, onaylanır ve maliyetlendirilir. On-prem, hibrit veya bulut. Agentic orkestrasyon · Egemen tasarım # Pilotta kalan yapay zekâyı üretime taşıyan katman. Supervisor ajanlar planlar, uzman ajanlar çekirdek sistemlerinizde aksiyon alır; her adım izlenir, gerekirse insan onayından geçer ve maliyetlendirilir. Modelden bağımsız; on-prem, hibrit ya da bulutta. 30 dakikalık keşif oturumu → Canlı ajanı izle Aşağıdaki ajanlar gerçek ajan tanımlarından türetilmiş simülasyonlardır. Sektör seçin, farklı bir senaryo izleyin. Bankacılık ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı Ajanı aç → [Diyalog bankacılığı: sesli ve yazılı işlem asistanı] Denetim izi00:00.00 - ◦Uygulama sohbeti: "Elektrik faturamı öde, limitimi 40 bine çıkar, dünkü 1.250 TL'yi tanımıyorum" - ∿Üç niyet ayrıştırıldı: fatura ödeme · limit artışı · işlem itirazı; sıra ve bağımlılık belirlendi çalıştırma #41.287.618süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Limit artışı > %60, itiraz tutarı > 5.000 TL veya yeni alıcıya transfer → operasyon onayı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Üretimden sayılar · perakende, sigorta, otomotiv ve bankacılık kurumlarında canlı 13.120.000+çalıştırma · son 30 gün %97,4başarı oranı · tüm ajanlar 1.920+üretimde aktif ajan 13.100+toplam ajan · tüm kurulumlar Gözlemlenebilirlik katmanından · Ekim 2026 Sorun ## Yapay zekâda asıl zorluk üretime geçiştir. Kurumların büyük çoğunluğu AI ile bir şeyler deniyor; çok azı bunu üretimde, ölçekte ve kontrollü biçimde işletebiliyor. ### PoC çukuru Pilotlar etkileyici, ama üretime geçiş yolu tanımsız. ### Dağınık girişimler Her birim kendi aracını deniyor; kurumsal bütünlük yok. ### Ölçülemeyen değer “Çalışıyor” ile “işe yarıyor” arasındaki fark raporlanamıyor. ### Yönetişim boşluğu Güvenlik, uyum ve denetlenebilirlik sonradan akla geliyor. ### Dış kaynak bağımlılığı Kişi-ay satın alınıyor; bilgi kişiyle gidiyor, zaman çizelgesi ne yapıldığını göstermiyor. Sonuç: bütçe harcanıyor, dönüşüm gerçekleşmiyor. Tek muhatap ## Danışmanlıktan platforma, altyapıdan operasyona. Kurumsal AI dönüşümünü sahiplenen ve uçtan uca işleten bir çözüm ortağı. Dört bileşen tek çatı altında; hesap verebilirlik tek yerde. 01 ### Yetkin ekip Strateji, veri, mühendislik ve değişim yönetimini bir arada yürüten deneyimli kadro. Danışman gelip gitmez; ürünü birlikte işletiriz. 02 ### Aginies platformu Tüm AI iş yüklerini merkezden yöneten kurumsal sınıf orkestrasyon katmanı: kurumun bütün AI senaryoları aynı yürütme katmanında koşar, çekirdek sistemlerde işlem yapar ve denetim izi bırakır. 03 ### Egemen altyapı GPU-as-a-Service ile ülke sınırları içinde bağımsız işletim. Veri ve model kontrolü sizde; sermaye yatırımı yapmadan, kullanım bazlı ölçeklenen kapasite. 04 ### Ajan iş gücü Dış kaynak insan yerine dış kaynak ajan: iOS geliştirici, iş analisti, muhasebeci, testçi gibi roller sizin araçlarınızda çalışır; her görev iş kaydında, kararlar sizde. Otonom Ajan ## Her ajan bir diyalogla kurulur. Ne istediğinizi yazın; Otonom Ajan süpervizörü ve uzman ajanları kurar, araçları ve bilgi tabanını bağlar, test korpusunu koşturur ve onayınızla üretime alır. Kütüphanedeki her ajan böyle üretilebilir. Otonom Ajan ve ürün ailesi → Otonom AjanAginies Workspace · Diyalog bankacılığı: sesli ve yazılı işlem asistanı Otonom Ajan’a yazın… AjanTestlerCanlı Ajan burada oluşacak Ürün ailesiPlatform Aginies çatıdır; diğerleri onun modülleridir. - Platform AginiesHepsi burada toplanır Neler dahil - Görsel ajan tasarımcısı, workspace ve tenant yönetimi, SSO ile roller - Çalıştırma motoru, onay kartları, kill-switch, panolar ve raporlar - Bilgi tabanı yönetimi ve Otonom Ajan Entegrasyon Kurumun kimlik sağlayıcısı (SAML/OIDC), SIEM, veri ambarı ve bilet sistemleriyle. 77 / 77 senaryoda kullanılıyor Ekranı ve ayrıntıları gör → - Voice AginiesSTT · TTS · ASR Neler dahil - Akışlı Türkçe ASR ve STT (aginies-voice-tr-v3), doğal TTS (tts-tr-v2) - Ses izi doğrulama, OTP okutma, sıcak aktarım - Çağrı kalite analizi: transkript, duygu, uyum kontrolü Entegrasyon Genesys, Avaya ve açık kaynak santrallerle SIP; kayıt sistemleriyle API. 23 / 77 senaryoda kullanılıyor Ekranı ve ayrıntıları gör → - Code AginiesIDE Registry · IDE eklentileri · kod tamamlama Neler dahil - IDE eklentileri (VS Code, JetBrains) ve kurum içi kod tamamlama - IDE Registry: onaylı eklenti ve model dağıtımı, sürüm sabitleme - SDLC ajan ekibi: analiz, tasarım, geliştirme, test, inceleme Entegrasyon Jira, Git (GitHub, GitLab, Bitbucket, self-hosted), CI/CD ve statik analiz araçlarıyla. 8 / 77 senaryoda kullanılıyor Ekranı ve ayrıntıları gör → - Hub AginiesModel Hub · MCP Hub · Model proxy Neler dahil - Model kataloğu: tulpar, talay, sunkar, konrul aileleri ve güncel açık sürümler - Model proxy ve anahtar yönetimi; kullanım ve maliyet izleme - MCP Hub: onaylı araç sunucuları, çağrı politikaları, bağlantı testleri Entegrasyon Ülke içi GPU havuzlarıyla; kurumun seçtiği bulut model sağlayıcılarıyla proxy üzerinden. 77 / 77 senaryoda kullanılıyor Ekranı ve ayrıntıları gör → - Vision AginiesDoküman · görüntü · maskeleme Neler dahil - Fatura, sözleşme, form ve kimlik belgesi çıkarımı; alan bazında güven skoru - Hasar fotoğrafı değerlendirme ve görsel kalite kontrol - Kişisel veri tespiti ve doküman maskeleme Entegrasyon Doküman depoları (S3, SharePoint), tarayıcı ve ECM sistemleriyle; sonuçlar ERP ve hasar sistemlerine araçla yazılır. 21 / 77 senaryoda kullanılıyor Ekranı ve ayrıntıları gör → - Workforce AginiesKiralık ajan iş gücü Neler dahil - Rol kataloğu: mühendislik, iş analizi, finans, destek, İK, hukuk - İş kaydı: görev, çıktı, saat karşılığı, kalite puanı, inceleme durumu - Gölge dönem → örneklemeli inceleme → otonom; adı belli ekip lideri Entegrasyon Kurumun kendi araçlarıyla: Jira, Git, ERP, bilet sistemi, e-posta. 17 / 77 senaryoda kullanılıyor Ekranı ve ayrıntıları gör → - Dialog AginiesDiyalog arayüzü Neler dahil - Web ve mobil diyalog arayüzü; gömülebilir widget - WhatsApp, SMS ve e-posta kanalları aynı ajanla - Onay kartları, dosya ve görsel paylaşımı, sıcak aktarım Entegrasyon Kurumun kimlik ve OTP altyapısıyla; WhatsApp Business API ve SMS geçitleriyle. 60 / 77 senaryoda kullanılıyor Ekranı ve ayrıntıları gör → Ajan Kiralama · Workforce Aginies ## Kaynak değil, ajan kiralayın. Dış kaynak firmasından iOS geliştirici, iş analisti, muhasebeci ya da testçi kiralamak yerine aynı rolü üstlenen ajanları kiralarsınız. Ajan sizin araçlarınızda çalışır, ekibinize rapor verir; ne yaptığı, ne kadar sürdüğü ve kaça mal olduğu iş kaydındadır. Rol kataloğu ve çalışma modeli → - iOS Geliştirici Ajanıyerine: Dış kaynak iOS geliştiriciPR / hafta · İlk incelemede geçme oranı - İş Analisti Ajanıyerine: Dış kaynak iş analistiHikâye / hafta · Geliştirmede geri dönüş oranı - Muhasebe Ajanıyerine: Dış kaynak muhasebeciKapanış günü · Mutabakat farkı - Test Mühendisi Ajanıyerine: Dış kaynak testçiKaçan hata / sürüm · Hata doğruluk oranı Workforce Aginies· kadro ve iş kaydı · fintech-opsPzt 09:00 kiralık ajanlar · 4 ios-dev-01iOS Geliştirici hazır mobil0 h bu hafta qa-01Test Mühendisi hazır mobil0 h bu hafta ba-01İş Analisti hazır ödeme0 h bu hafta acct-01Muhasebe hazır finans0 h bu hafta Tek panoda izlenebilirlik ve raporlama ## Ajan etkinliğini, platform sağlığını ve model maliyetlerini tek panodan izleyin. Her çalıştırma adım adım kayıtlı; tüm istekler maskelenebilir ve anonimleştirilebilir. Workspace ve tenant bazında ayrılır; ekipler ajanlarını kurum içinde paylaşır, birlikte geliştirir. - Ajan bazında başarı ısı haritası ve hata zaman çizelgesi - Platform sağlığı, model kullanımı ve maliyet - Kendi anahtarlarınızla bulut modellerine proxy; tek noktadan izlenir - Maskeleme, anonimleştirme, tam denetim izi Platformu keşfet → Platform Aginies · Pano · son 30 güntüm workspace’ler · tüm tenant’lar çalıştırma13.120.000 başarı%97,4 p50 gecikme1,4 s platform sağlığısağlıklı en yoğun ajanlar · her hücre ≈ 1 gün dialog_banking_assistant%97,4 card_debt_payment%99,1 kvkk_data_detection%96,2 kvkk_document_masking%98,8 motor_claim_end_to_end%98,1 merchant_operations%100 model kullanımı aginies-tulpar-438% aginies-talay-229% aginies-sunkar-119% aginies-voice-tr-v39% cloud · proxy$2.960 maliyet · senaryo bazında Diyalog bankacılığı26% Kredi ve edinim24% Hasar operasyonları18% KVKK maskeleme14% Üye işyeri10% Diğer8% PII maskeleme: açıkanonimleştirme: açıkdenetim izi: tam1.920+ aktif ajan Nasıl çalışır ## Agentic AI gerçek iş süreçlerini dört adımda yürütür. - 01 ### Sinyalleri anla Uygulamalar, API’ler, dokümanlar, ses kanalları ve olaylardan gelen girdiler yürütmeye hazır görevlere dönüşür. ApplicationsAPIsDocumentsVoiceEvents - 02 ### Süpervizör planlar Karmaşık hedefler yapılandırılmış adımlara bölünür; sorumluluklar uzman ajanlara dağıtılır. SupervisorPlanningDecompositionDelegation - 03 ### Ajanlar aksiyon alır Bilgi getirilir, belgeler işlenir, araçlar çağrılır, kurumsal sistemlerde gerçek işlemler tetiklenir. RetrievalTools / MCPActionsIntegrations - 04 ### Her şey izlenir Ajan kararları, gecikme, maliyet ve araç kullanımı ortamlar arasında tam görünürlükle izlenir ve iyileştirilir. TracesLatencyGuardrailsAudit Platform ## Sekiz entegre yetenek katmanı. Kurumsal ölçek, yönetişim ve dayanıklılık için tasarlandı. Katmanlar tek platformda birlikte çalışır. Platformu keşfet → A ### Orkestrasyon & Yürütme Zekâsı Karmaşık işi uçtan uca yürütür; talebi akıllıca yönlendirir, çok adımlı süreçleri bağımlılıklar başarısız olsa bile kararlı tutar. B ### AI / Model Orkestrasyonu Her iş için doğru modeli kullanır; vendor bağımlılığını azaltır, bir sağlayıcı düştüğünde operasyonu sürdürür. C ### Bilgi & Erişim Katmanı Dağınık kurumsal bilgiyi yönetilen, aranabilir ve güvenilir bilgiye dönüştürür; cevaplar kaynağa bağlanır. D ### Yönetişim, Risk & Uyum AI’ı denetlenebilir, güvenli ve regüle ortamlara uygun kılar; her karar izlenebilir. E ### Gözlemlenebilirlik & Optimizasyon Doğruluk, hız, maliyet ve sağlık görünürlüğü sağlar; sürekli iyileştirmenin verisini üretir. F ### Entegrasyon & Bağlantı Mevcut teknoloji ortamını yeniden kurmadan kurumsal sistemlere ve müşteri kanallarına bağlanır. G ### Çok Kanallı Deneyim & Ses Dijital, mesajlaşma ve ses kanallarında tutarlı bir orkestrasyon üzerinde AI sunar. H ### Dağıtım & Kurumsal Mimari Büyük kurumların on-prem, hibrit ve egemenlik gereksinimlerini destekler. Üretimde çalışan gerçek çözümler ## Aginies üzerinde bugün canlıda çalışan iş yükleri. Perakende & E-ticaret ### E-ticaret müşteri desteği ve aksiyon Sipariş nerede, iade, teslimat değişikliği: ajan işlemi sipariş sisteminde tamamlar. Devamı → Kurumsal Fonksiyonlar ### Fatura ve doküman veri çıkarımı OCR → doğrulama → ERP: elle veri girişi olmadan temiz kayıt. Devamı → Sigorta ### Sesli AI ile hasar ve talep kaydı Çağrı merkezine gelen hasar bildirimini sesli ajan alır, kaydı sistemde açar. Devamı → Yazılım Geliştirme ### SDLC ajan ekibi PM, analist, mimar, geliştirici ve test ajanları CI/CD hattında birlikte çalışır. Devamı → Kurumsal Fonksiyonlar ### Ajan kiralama (Agent Outsourcing) Dış kaynak insan yerine dış kaynak ajan: iOS geliştirici, iş analisti, muhasebeci, testçi; iş kaydıyla. Devamı → Bankacılık ### Diyalog bankacılığı Mobil bankacılık işlemlerini yazarak ya da konuşarak gerçekleştir. Devamı → Ajan kütüphanesinde 77 canlı ajan → 77canlı senaryo 30çözüm paketi 14sektör 200bağlı sistem 73onay kuralı 657araç tanımı Ajan kütüphanesinin senaryo tanımlarından derleme sırasında türetilir. SektörlerOn dört sektör, tek platform. - Bankacılık9 - Finansal Teknolojiler5 - Sigorta6 - Perakende & E-ticaret6 - Telekom5 - Sağlık4 - Üretim & Otomotiv4 - Lojistik3 - Enerji5 - Kamu ve Hükümet5 - Kurumsal Fonksiyonlar9 - Yazılım Geliştirme8 - Savunma Sanayii4 - IoT ve Bağlı Cihazlar4 Farkımız ## Disiplinli bir dönüşüm metodolojisi. Hangi AI fikrinin yapılmaya değdiğini ve ne zaman başlanacağını sayılarla belirleriz. Her kullanım senaryosunu beş katmanlı değerlendirme çerçevesinden geçiririz; her aşamanın sonunda net bir karar noktası vardır: devam, düzelt veya durdur. - 01 Değer Bu iş neden yapılmalı? - 02 Hazırlık Gerekli girdiler elimizde var mı? - 03 Uygulanabilirlik Bunu gerçekten yapabilir miyiz? - 04 Operasyon Canlıya çıktıktan sonra nasıl yaşayacak? - 05 Yönetişim Güvenli, kontrollü ve denetlenebilir mi? Senaryonuzu 3 dakikada değerlendirin → Metodoloji Güvenlik, uyum ve egemenlik ## Regüle ortamlar için tasarlandı. AI, kurumun diğer kritik sistemleriyle aynı standartlara tabi olmalı. Finans, sigorta, sağlık, kamu ve savunma gibi regüle sektörlerin gereksinimlerine göre kurgulandı; kapalı ağ ve air-gap kurulum dahil. SOC 2 Type IIuyumlu platform ISO 27001uyumlu platform KVKK & GDPRveri yerleşimi garantisi On-prem / air-gapkurum dışına veri çıkmaz Mimariyi ve dağıtım seçeneklerini gör → ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Ajan Kiralama URL: https://www.aginies.com/tr/ajan-kiralama Language: tr Summary: Dış kaynak insan yerine dış kaynak ajan: iOS geliştirici, iş analisti, muhasebeci, test mühendisi gibi roller kurumunuzun araçlarında çalışan ajanlar olarak kiralanır; her görev iş kaydında, kararlar sizde. Ajan Kiralama · Workforce Aginies # Kaynak değil, ajan kiralayın. Dış kaynak firmasından iOS geliştirici, iş analisti, muhasebeci ya da testçi kiralamak yerine aynı rolü üstlenen ajanları kiralarsınız. Ajan sizin araçlarınızda çalışır, sizin ekibinize rapor verir; ne yaptığı, ne kadar sürdüğü ve kaça mal olduğu iş kaydındadır. Kararlar sizin onaylayanınızda, Aginies tarafında adı belli bir ekip lideri. 12+hazır rol · sınırsız özel rol 2 haftagölge dönem, %100 inceleme 7/24erişilebilir; boşta ödeme yok 1adı belli Aginies ekip lideri Workforce Aginies· kadro ve iş kaydı · fintech-opsPzt 09:00 kiralık ajanlar · 4 ios-dev-01iOS Geliştirici hazır mobil0 h bu hafta qa-01Test Mühendisi hazır mobil0 h bu hafta ba-01İş Analisti hazır ödeme0 h bu hafta acct-01Muhasebe hazır finans0 h bu hafta Simülasyon: dört kiralık ajanın bir haftası. Sekmelerden günlük özet, haftalık rapor, sürüm bülteni ve iş kaydını izleyin; alttaki kutudan bir ajana talimat verin. Neden ## Dış kaynak modelinin üç kronik sorunu 01 ### Devir ve bilgi kaybı Kişi gider, bildikleri de gider; yeni gelen sıfırdan başlar. Ajanın bilgisi workspace’te kalır, bir sonraki aynı korpusla başlar. 02 ### Görünmezlik Zaman çizelgesi yalnızca kaç saat yazıldığını gösterir. İş kaydı her görevi, çıktıyı, kaliteyi ve onayı gösterir. 03 ### Ölçekleme gecikmesi Yeni kişi aylar sürer; aynı rolden ikinci ajan dakikalar. Talep düşünce kapasite de düşer, fatura da. Rol kataloğu ## Bugün kiralanabilen roller Her rol bir ajan spesifikasyonudur: görevler, araçlar, çıktılar, KPI’lar ve onay noktaları. Kataloğa sığmayan roller keşif oturumunda Otonom Ajan ile günler içinde tanımlanır. Mühendislik ### iOS Geliştirici Ajanı yerine: Dış kaynak iOS geliştirici aginies-sunkar-1 · Code Aginies Ne yapar - Jira’daki hikâyeyi alır, SwiftUI ile geliştirir, birim ve UI testlerini yazar - PR açar, inceleme yorumlarını uygular, CI’ı yeşile getirir - Sürüm notu ve App Store metadata taslağını hazırlar Nerede çalışır JiraGitHub / GitLabXcode CloudTestFlightFigma Çıktılar Birleştirilen PRKapanan hikâye puanıTest kapsamı artışı Haftalık KPI PR / haftaİlk incelemede geçme oranıBuild başarı oranıSaat karşılığı İnsan kontrolü: Her PR bir insan tarafından incelenir; üretim dağıtımı ekip liderinin onayında Kütüphanede örnek → ### Backend Geliştirici Ajanı yerine: Dış kaynak backend geliştirici aginies-sunkar-1 · Code Aginies Ne yapar - API uç noktaları, veri modeli ve migrasyonları yazar - Mevcut kod tabanına uyar: lint, mimari kurallar, gözlemlenebilirlik - Entegrasyon testlerini ve API dokümantasyonunu üretir Nerede çalışır JiraGitHubCI/CDPostgreSQLOpenAPI Çıktılar Birleştirilen PRUç nokta / servisKapatılan hata Haftalık KPI PR / haftaHata yeniden açılma oranıp95 gecikme etkisiSaat karşılığı İnsan kontrolü: Şema değişikliği ve üretim dağıtımı insan onaylı Kütüphanede örnek → ### Test Mühendisi Ajanı yerine: Dış kaynak testçi aginies-talay-2 · on-prem Ne yapar - Kabul kriterlerinden test senaryoları üretir, regresyon setini korur - Her sürümde regresyonu koşar, hataları kanıt ve adımlarla açar - Kapsama ve kaçan hata raporu yazar Nerede çalışır Jira / XrayPlaywrightXCUITestCI/CDBrowserStack Çıktılar Koşulan test / sürümAçılan hataRegresyon kapsamı Haftalık KPI Kaçan hata / sürümHata doğruluk oranıRegresyon süresiSaat karşılığı İnsan kontrolü: Sürüm “go” kararı sürüm yöneticisinde Kütüphanede örnek → ### DevOps / SRE Ajanı yerine: Dış kaynak DevOps mühendisi aginies-tulpar-4 · on-prem Ne yapar - Pipeline, IaC ve izleme kurallarını yazar ve bakımını yapar - Olaylarda ilk müdahaleyi yapar, runbook uygular, geri alır - Bulut maliyet ve kapasite raporu üretir Nerede çalışır GitHub ActionsTerraformKubernetesGrafanaPagerDuty Çıktılar Kapanan olayPipeline değişikliğiMaliyet tasarrufu önerisi Haftalık KPI MTTRDağıtım sıklığıDeğişiklik hata oranıSaat karşılığı İnsan kontrolü: Üretim altyapı değişikliği ve geri alma onaylı Kütüphanede örnek → İş analizi ve veri ### İş Analisti Ajanı yerine: Dış kaynak iş analisti aginies-tulpar-4 · on-prem Ne yapar - Talep ve toplantı notlarından kullanıcı hikâyesi ve kabul kriteri yazar (Gherkin) - Süreç haritası, veri sözlüğü ve etki analizi çıkarır - Backlog’u düzenler, tahmin için hazır hale getirir Nerede çalışır JiraConfluenceMiroTeams / Slack Çıktılar Hazır hikâyeSüreç dokümanıEtki analizi Haftalık KPI Hikâye / haftaGeliştirmede geri dönüş oranıKriter netlik puanıSaat karşılığı İnsan kontrolü: Ürün sahibi hikâyeleri onaylar Kütüphanede örnek → ### Veri Analisti Ajanı yerine: Dış kaynak veri analisti aginies-tulpar-4 · on-prem Ne yapar - SQL sorguları, panolar ve haftalık raporlar üretir - Anomali ve trendleri açıklar, kaynak veriye bağlar - Veri kalitesi kontrolleri koşar Nerede çalışır BigQuery / SnowflakedbtPower BI / LookerSlack Çıktılar RaporPanoCevaplanan analiz talebi Haftalık KPI Talep → rapor süresiYeniden çalışma oranıVeri kalitesi bulgusuSaat karşılığı İnsan kontrolü: Dışa paylaşılan rapor analitik liderinin onayında Finans ve operasyon ### Muhasebe Ajanı yerine: Dış kaynak muhasebeci aginies-tulpar-4 · Vision Aginies Ne yapar - Fatura ve ekstreleri işler, yevmiye kayıtlarını hazırlar - Banka ve tedarikçi mutabakatını yapar, farkları açıklar - Ay sonu kapanış kontrol listesini yürütür, KDV hazırlığını yapar Nerede çalışır ERP (SAP / Logo / Netsis)e-FaturaBanka APIExcel Çıktılar KayıtMutabakatKapanış görevi Haftalık KPI Kapanış günüMutabakat farkıDüzeltme oranıSaat karşılığı İnsan kontrolü: Kayıtlar mali müşavir onayıyla defterlere işlenir; beyanname insanda Kütüphanede örnek → ### Satın Alma Uzmanı Ajanı yerine: Dış kaynak satın alma uzmanı aginies-talay-2 · on-prem Ne yapar - Talepleri toplar, RFQ hazırlar, tedarikçi cevaplarını karşılaştırır - Sipariş, mal kabul ve fatura üçlü eşleştirmesini yapar - Tedarikçi performans raporu üretir Nerede çalışır ERPE-postaTedarikçi portalıExcel Çıktılar RFQEşleşen faturaTasarruf raporu Haftalık KPI Talep → sipariş süresiEşleşme istisnasıTasarrufSaat karşılığı İnsan kontrolü: Sipariş onayı ve tedarikçi seçimi satın alma yöneticisinde Kütüphanede örnek → Müşteri, satış ve İK ### Müşteri Destek Temsilcisi Ajanı yerine: Dış kaynak çağrı merkezi temsilcisi aginies-talay-2 · on-prem Ne yapar - Bilet ve sohbetleri cevaplar, sistemde işlem yapar - Politika dışı talepleri gerekçeyle insana devreder - Haftalık konu ve kök neden raporu yazar Nerede çalışır Zendesk / SalesforceDialog AginiesWhatsAppCRM Çıktılar Çözülen biletİlk temasta çözümDevir Haftalık KPI CSATOrtalama çözüm süresiDevir oranıSaat karşılığı İnsan kontrolü: İade, iptal ve tutar üstü işlemler onaylı; %10 örneklemeli kalite incelemesi Kütüphanede örnek → ### Satış Geliştirme Ajanı yerine: Dış kaynak SDR aginies-tulpar-4 · on-prem Ne yapar - Fırsatları araştırır, puanlar, CRM’i günceller - Kişiselleştirilmiş ilk temas taslağı yazar; gönderim insan onaylı - Toplantı planlar ve brifing hazırlar Nerede çalışır HubSpot / SalesforceLinkedInE-postaTakvim Çıktılar Nitelendirilen fırsatPlanlanan toplantıBrifing Haftalık KPI Toplantı / haftaNitelendirme doğruluğuCevap oranıSaat karşılığı İnsan kontrolü: Dış iletişim gönderimi her zaman insan onayıyla Kütüphanede örnek → ### İşe Alım Uzmanı Ajanı yerine: Dış kaynak işe alım uzmanı aginies-tulpar-4 · on-prem Ne yapar - Başvuruları kriterlere göre ön eler, gerekçe yazar - Mülakat planlar, adayla iletişimi yürütür - İşe alım hunisi raporu üretir Nerede çalışır ATS (Greenhouse / Kariyer.net)TakvimE-postaHRIS Çıktılar Ön elenen adayPlanlanan mülakatHuni raporu Haftalık KPI Pozisyon kapama süresiMülakat / işe alım oranıAday memnuniyetiSaat karşılığı İnsan kontrolü: Eleme ve teklif kararı İK’da; ajan öneri ve gerekçe sunar Kütüphanede örnek → Hukuk ve uyum ### Sözleşme Analisti Ajanı yerine: Dış kaynak hukuk asistanı aginies-tulpar-4 · on-prem Ne yapar - Sözleşme maddelerini çıkarır, playbook’la karşılaştırır, sapmaları işaretler - Redline önerisi ve özet yazar - Sözleşme takvimi ve yenileme uyarıları üretir Nerede çalışır CLMWordE-postaDMS Çıktılar İncelenen sözleşmeRedlineÖzet Haftalık KPI İnceleme süresiKaçan sapmaAvukat kabul oranıSaat karşılığı İnsan kontrolü: Her redline avukat onayıyla karşı tarafa gider Kütüphanede örnek → + ### Kataloğa sığmayan bir rol mü? Rol sayısında sınır yok. İş tanımını yazın; Otonom Ajan süpervizörü ve uzman ajanları kurar, araçlarınıza bağlar, test korpusunu koşturur. İki haftalık gölge dönemden sonra rol kataloğa girer. Tipik örnekler: içerik editörü, tedarik zinciri planlamacısı, hukuk sekreteri, pazarlama analisti, teknik yazar. Rolü birlikte tanımlayalım → Bir talimatın yolu ## Talimattan bültene yedi adım Ajan bir Jira hikâyesi, e-posta ya da doğrudan mesajla görev alır; planı görünür; işi sizin araçlarınızda yapar; kontroller koşar; eşik aşan çıktı insana düşer; sonuç iş kaydına yazılır; özet, rapor ve bülten bu kayıttan üretilir. - 01TalimatJira hikâyesi, e-posta ya da doğrudan mesaj: “PAY-119 karanlık mod hatasını düzelt” - 02PlanKapsam, dokunulacak araçlar, onay noktası; 3 adımlık plan ekip liderine görünür - 03İşSizin araçlarınızda: kod, test, kayıt, rapor; her adım izli - 04KontrollerLint, test, politika, PII; kontrol aileleri puanlanır - 05OnayEşik aşan çıktı insana düşer: PR incelemesi, tutar üstü kayıt - 06İş kaydıGörev · çıktı · saat karşılığı · kalite · onay · maliyet - 07BültenGünlük özet 09:00 · haftalık rapor Cuma · sürüm bülteni yayında Rutinler ## Bir çalışan gibi: sabah özeti, Cuma raporu, sürümde bülten Kiralık ajanın günü bir çalışanın günü gibi akar. Aradaki her an yeni talimat verilebilir; ajan planını ve ilerlemesini gösterir, yetkisi dışındaki adımlarda durup sorar. | ne zaman | ne | içerik | kime | | --- | --- | --- | --- | | Her sabah 09:00 | Günlük özet | Dün biten, bugün planlanan, engeller; ajan bazında | ekip lideri · Slack / e-posta | | Sürekli | İş kaydı | Her görev bir satır: çıktı, saat karşılığı, kalite, onay, maliyet | onaylayan · onay kuyruğu | | Cuma 17:00 | Haftalık rapor | Çıktılar, KPI’lar, kalite, maliyet; zaman çizelgesi dışa aktarımı | ekip lideri · yönetim | | Her sürüm | Sürüm bülteni | Ne çıktı, kim ne yaptı, testler, onaylar, yayın kademesi | ekip · paydaşlar | | Ay sonu | Kapanış ve fatura | Muhasebe kapanış görevleri; iş kaydından fatura kalemi | finans · satın alma | | Talep üzerine | Talimat | Doğrudan mesaj, Jira ya da e-posta ile yeni görev; plan ve ilerleme görünür | herkes · yetki dahilinde | Örnekler ## Haftalık rapor ve sürüm bülteni, ajanların kaleminden Haftalık rapor · 38. hafta · fintech-opsCuma 17:00 · zaman çizelgesi dışa aktarıldı · fatura kalemi hazır | ajan | rol | çıktı | ne | kalite | saat karş. | maliyet | | --- | --- | --- | --- | --- | --- | --- | | ios-dev-01 | iOS Geliştirici | 5 | 4 PR · 21 puan · sürüm notları | 4,7 | 20 h | $6,7 | | qa-01 | Test Mühendisi | 6 | 1.245 test · 1 hata · go önerisi | 4,8 | 24 h | $4,8 | | ba-01 | İş Analisti | 5 | 14 hikâye · 2 süreç · retro | 4,6 | 20 h | $4,5 | | acct-01 | Muhasebe | 5 | 236 kayıt · 3 mutabakat · kapanış | 4,9 | 23.5 h | $4,6 | 21çıktı 87,5 hsaat karşılığı 9insan onayı $20,6toplam maliyet - Sürüm 2.14.0 yayında; kaçan hata 0; çökme %0,08 (eşik %0,3). - Eylül kapanışı 5. iş gününde bitti (önceki ay 9); 3 kalem mali müşavir onayıyla. - Aynı işin insan karşılığı ~88 saat; dört rolün ajan-ay maliyeti bunun altında. Rakamlar temsilidir. sürüm bülteni2.14.0 · fintech-ops · mobilPer 13:00 · ekibe ve paydaşlara e-posta Ne çıktı Ödeme özeti ekranı (PAY-117), karanlık mod düzeltmeleri (PAY-119), 6 hata düzeltmesi ios-dev-01 PR #4818, #4823 birleştirildi · 10 puan · sürüm notları ve App Store metadata qa-01 rc2 → rc4: 421/421 · BUG-2291 bulundu, düzeltme doğrulandı · kaçan hata 0 ba-01 PAY-118 kabul kriterleri sürümdeki 9 testin kaynağı; 3 belirsiz hikâye ürün sahibinde Onaylar d.kaya (PR incelemesi) · e.demir (sürüm go) · kayıt: kim, ne zaman, gerekçe Yayın %10 kademeli → 24 saat izleme (çökme eşiği %0,3) → %50 → %100 Sonraki PAY-121 ve PAY-122 geliştirmeye hazır; PAY-124 incelemede Bülten ajanlar tarafından iş kaydından yazılır; aynı biçimde günlük özet (her sabah 09:00) ve haftalık rapor (Cuma 17:00) üretilir. Takip ## Neyi ne kadar yaptığı: iş kaydı Her görev bir satır: bilet numarası, çıktı, saat karşılığı, kalite puanı, onay ve maliyet. Haftalık zaman çizelgesi ve aylık fatura aynı kayıttan üretilir; sürpriz yoktur. GörevBilet / hikâye numarası, başlangıç ve bitiş, durum ÇıktıPR, rapor, kayıt, bilet: somut ve sayılabilir Saat karşılığıAynı işin insan süresi tahmini; fatura ve karşılaştırma için Kaliteİnceleme sonucu, yeniden çalışma, kontrol ailesi puanları OnayKim, ne zaman, gerekçe; eşik aşan her adım MaliyetToken, araç çağrısı, model; görev ve rol bazında günlükiş kaydı · onay kuyruğu haftalıkzaman çizelgesi · KPI · kalite aylıkfatura · gözden geçirme · ölçek kararı Her satır bir görevdir; haftalık zaman çizelgesi ve aylık fatura aynı kayıttan üretilir. Ne yapıldığı, ne kadar sürdüğü, kimin onayladığı ve kaça mal olduğu sorularının cevabı tek yerdedir. Çalışma modeli ## Beş adımda işe başlar, ölçeklenir Rol tanımından aylık gözden geçirmeye kadar her adımda sorumluluk nettir. Gölge dönemde her çıktı incelenir; kalite eşiği geçilince örneklemeli incelemeye geçilir. Kill-switch her zaman açıktır. - 01Rol tanımıİş tanımı ajan spesifikasyonuna çevrilir: görevler, araçlar, çıktılar, KPI’lar, onay noktaları.siz + Aginies ekip lideri - 02Erişim ve onboardingAjan sizin araçlarınızda hesap alır: SSO, roller, okuma / yazma izinleri, veri sınırları. Kod ve veri sizde kalır.BT + güvenlik - 03Gölge dönemİlk iki hafta her çıktı bir insan tarafından incelenir; hatalar test korpusuna eklenir, ajan düzeltilir.ekip lideriniz + Aginies - 04Örneklemeli incelemeKalite puanı eşiği geçince inceleme %10 örnekleme ve eşik bazlı onaya iner; kill-switch her zaman açık.ekip lideriniz - 05Aylık gözden geçirme ve ölçekİş kaydı, KPI’lar ve maliyet birlikte okunur; rol genişletilir, yeni rol eklenir ya da sözleşme sonlandırılır.siz + Aginies Karşılaştırma ## İnsan dış kaynak ile ajan dış kaynak yan yana Aynı işi yapan iki model; farkı işe başlama süresi, izlenebilirlik, bilgi birikimi ve fiyatlamada ortaya çıkar. Her ikisinde de son karar sizin ekibinizdedir. | | dış kaynak insan | dış kaynak ajan · Workforce Aginies | | --- | --- | --- | | İşe başlama | 4–8 hafta arama, sözleşme, onboarding | Günler: rol tanımı, erişim, gölge dönem | | Erişilebilirlik | Mesai saatleri, izin, devir | 7/24; talep düşünce durur, artınca ölçeklenir | | Bilgi birikimi | Kişiyle gelir, kişiyle gider | Workspace’te kalır; bir sonraki ajan aynı korpusla başlar | | İzlenebilirlik | Zaman çizelgesi ve haftalık rapor | Adım adım iş kaydı: görev, çıktı, kalite, onay, maliyet | | Kalite tutarlılığı | Kişiye ve güne göre değişir | Test korpusu ve kontrol aileleriyle sürümlü | | Veri güvenliği | Dizüstü, ev ağı, üçüncü taraf ofis | Sizin tenant’ınızda; kod ve veri dışarı çıkmaz | | Ölçekleme | Yeni işe alım, aylar | Aynı rolden ikinci ajan dakikalar | | Fiyatlama | Kişi-ay, kullanım fark etmez | Ajan-ay ya da çıktı bazlı; kullanılmayan kapasiteye ödeme yok | | İnsan sorumluluğu | Tedarikçi proje yöneticisi | Aginies tarafında adı belli ekip lideri; kararlar sizin onaylayanınızda | Ticari model ## Üç fiyatlama modeli, liste fiyatı yok Rol, hacim ve onay yoğunluğuna göre keşif oturumunda birlikte belirlenir. Çıktıların mülkiyeti sizdedir; sözleşme aylık sonlandırılabilir. ### Ajan-ay Rol başına aylık sabit ücret; sınırsız görev, tanımlı KPI ve SLA. Uygun olduğu durumSürekli iş yükü olan roller ### Çıktı bazlı Birleştirilen PR, çözülen bilet, işlenen fatura başına ücret; iş kaydı fatura kalemidir. Uygun olduğu durumDalgalı ya da proje bazlı işler ### Ajan pod’u Birden çok rol + Aginies ekip lideri; ürün ekibi gibi çalışır, haftalık planlama ve retrospektif. Uygun olduğu durumUçtan uca teslimat isteyen ekipler Ölçülen hedefler ve yol haritası ## İlk ajanınız 4 haftada gölge dönemden çıkar. Hedefler benzer kurulumlardan aralık olarak; başlangıç değeri ve hedef ilk haftada birlikte sabitlenir. Ölçülen hedefler | Kabul edilen çıktı payı | %85 ve üzeri | | Rol başına maliyet | Dış kaynak kişi-ayın %30–50’si | | Bilgi kaybı (devirde) | Sıfır; bilgi workspace’te kalır | Başlamak için gerekenler - Rol için gerekli araç erişimleri (Jira, Git, ERP, bilet sistemi) - Görev tanımı ve kabul kriterleri - Kurum tarafında inceleme sorumlusu Yol haritası - 1 · Politika ve rol kataloğu · 2 haftaİK politikaları, süreç sahipleri ve sistem erişimleri tanımlanır; kişisel veri sınırları yazılır. - 2 · Pilot · 4 haftaBir departman ya da bir rolle; kalite puanı ve kullanıcı geri bildirimi haftalık izlenir. - 3 · YaygınlaştırmaTüm departmanlar; eşikler, inceleme örneklemi ve kill-switch İK yönetiminde. Yönetişim ## Ajan çalışan gibi yönetilir, sistem gibi denetlenir - Sizin araçlarınızda, sizin tenant’ınızda hesap; kod ve veri dışarı çıkmaz - Araç bazında okuma / yazma izni; eşik aşan adım onaya düşer - Her çıktı için kimin onayladığı ve gerekçesi kayıtlı - Aginies tarafında adı belli ekip lideri; haftalık planlama, aylık gözden geçirme - Kill-switch: rol ya da görev bazında anında durdurma - Çıktıların ve üretilen bilginin mülkiyeti sizde; ayrılırken korpus size kalır Kütüphanede ## Kiralanan rollerin canlı örnekleri Yazılım GeliştirmeHITL ### iOS özellik geliştirme (kiralık geliştirici ajanı) Jira hikâyesi alınır, SwiftUI ile geliştirilir, testler yazılır, PR açılır; inceleme ve dağıtım kararı ekipte kalır. 380 çalıştırma / ay %94,4 başarı 4 ajan Yazılım GeliştirmeHITL ### Özellik teslimat hattı Bir Jira epic'i ajanlarla kriter, tasarım, kod ve teste dönüşür; insan yalnızca PR'ı inceler ve onaylar. 1.200 çalıştırma / ay %91,2 başarı 5 ajan Yazılım GeliştirmeHITL ### Regresyon test döngüsü (kiralık test ajanı) Her sürüm adayında regresyon koşulur, yeni hatalar kanıtla açılır, kapsama raporu yazılır; “go” kararı sürüm yöneticisinde. 260 çalıştırma / ay %95,8 başarı 4 ajan Yazılım GeliştirmeHITL ### Olay müdahalesi ve acil düzeltme Alarm gelir, hata yeniden üretilir, yama ve regresyon hazır; nöbetçi onaylar, post-mortem taslağı bekler. 320 çalıştırma / ay %90,5 başarı 5 ajan Yazılım GeliştirmeHITL ### Backlog düzenleme ve tahminleme Talepler triyajlanır, netleştirilir, karmaşıklığı tahminlenir; sprint önerisi ürün sahibinin onayına gider. 340 çalıştırma / ay %92,8 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Ay sonu kapanış ve mutabakat (kiralık muhasebe ajanı) Faturalar işlenir, banka ve tedarikçi mutabakatı yapılır, kapanış listesi yürütülür; kayıtlar mali müşavir onayıyla deftere işlenir. 60 çalıştırma / ay %96,2 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Ajan Kütüphanesi URL: https://www.aginies.com/tr/ajanlar Language: tr Summary: 77 canlı ajan senaryosu: bankacılık, fintech, sigorta, perakende, telekom, sağlık, üretim, lojistik, enerji, kamu, savunma sanayii, IoT, yazılım geliştirme ve kurumsal fonksiyonlar. Her ajan supervisor, uzman alt ajanlar, sistemler ve onay noktalarıyla simüle edilir. Ajan kütüphanesi # Üretimde çalışan kurumsal ajan senaryoları, canlı simülasyonla. 77 ajan; her biri supervisor, uzman alt ajanlar, dokunduğu sistemler, onay kuralı ve maliyetiyle. Küresel örneklerden derlendi, Aginies ilkelleriyle yeniden kuruldu. Sektör seçin, ajanı izleyin, kendi senaryonuzla karşılaştırın. 77canlı senaryo 14sektör 261uzman ajan 200bağlı sistem 73onay kuralı 3.320.970simüle çalıştırma / ay Bankacılık ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı Ajanı aç → [Diyalog bankacılığı: sesli ve yazılı işlem asistanı] Denetim izi00:00.00 - ◦Uygulama sohbeti: "Elektrik faturamı öde, limitimi 40 bine çıkar, dünkü 1.250 TL'yi tanımıyorum" - ∿Üç niyet ayrıştırıldı: fatura ödeme · limit artışı · işlem itirazı; sıra ve bağımlılık belirlendi çalıştırma #41.287.381süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Limit artışı > %60, itiraz tutarı > 5.000 TL veya yeni alıcıya transfer → operasyon onayı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır BankacılıkHITL ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı Müşteri konuşur ya da yazar; fatura, limit, itiraz ve transfer aynı oturumda çekirdek sistemde tamamlanır. 186.000 çalıştırma / ay %94,7 başarı 5 ajan Finansal TeknolojilerHITL ### Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit İşyeri WhatsApp'tan yazar; POS arızası, hakediş sorgusu, iade ve limit artışı tek diyalogda çözülür. 64.000 çalıştırma / ay %93,8 başarı 5 ajan SigortaHITL ### Uçtan uca kasko hasarı: ihbardan ödemeye Sesli ihbar, fotoğraf ekspertizi, anlaşmalı servis ataması ve ödeme talimatı tek akışta; eksper yalnızca eşik üstünde devreye girer. 21.000 çalıştırma / ay %91,2 başarı 5 ajan Perakende & E-ticaretHITL ### Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin Müşteri bir fotoğraf ve bir cümle gönderir; ajan stokta olan, bedenine uyan ve bütçeye sığan üç kombin kurar, seçileni sepete ekler. 142.000 çalıştırma / ay %96,1 başarı 5 ajan TelekomHITL ### Abone hizmet asistanı: fatura, paket, arıza ve taahhüt Abone arar ya da yazar; fatura itirazı, paket değişikliği ve hat arızası aynı oturumda BSS ve şebeke sistemlerinde çözülür. 410.000 çalıştırma / ay %94,3 başarı 5 ajan Yazılım GeliştirmeHITL ### Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR Jira'daki bir epic'ten başlar; eski modül okunur, davranışı testlerle sabitlenir, yeniden yazılır ve kanıtlı bir PR açılır. 1.900 çalıştırma / ay %90,4 başarı 5 ajan SağlıkHITL ### Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuç Hasta yazar ya da arar; randevu, ön hazırlık talimatı, sigorta ön onayı ve sonuç bildirimi tek koordinatörde birleşir. 58.000 çalıştırma / ay %95,1 başarı 5 ajan Üretim & OtomotivHITL ### Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna MES alarmı düşer; kök neden tarihsel veriyle bulunur, yedek parça kontrol edilir, iş emri ve plan güncellemesi dakikalar içinde hazırdır. 3.100 çalıştırma / ay %92,2 başarı 5 ajan LojistikHITL ### Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA Telematik sinyali gecikmeyi erken yakalar; alternatif rota ve taşıyıcı seçilir, müşteri bilgilendirilir, SLA tazmini hesaplanır. 96.000 çalıştırma / ay %93,4 başarı 5 ajan EnerjiHITL ### Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda Abone arar ya da yazar; yüksek fatura açıklanır, kesinti süresi verilir, sayaç verisi gösterilir, taksit planı kurulur. 520.000 çalıştırma / ay %94,6 başarı 5 ajan Kamu ve HükümetHITL ### Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi Vatandaş yazar ya da arar; uygunluk kontrol edilir, belgeler okunur, başvuru açılır, randevu verilir, her adım izlenebilir. 230.000 çalıştırma / ay %93,8 başarı 5 ajan Kurumsal FonksiyonlarHITL ### Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada Talep e-postayla gelir; teklifler karşılaştırılır, sözleşme maddeleri kontrol edilir, sipariş açılır, fatura üç yönlü eşleşir ve ödeme planlanır. 14.000 çalıştırma / ay %92,7 başarı 5 ajan Savunma SanayiiHITL ### Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolü Teknisyen hangarda sorar; doğru revizyondaki prosedür, parça numarası ve emniyet notu yetki kontrolüyle gelir, her erişim iz bırakır. 9.800 çalıştırma / ay %95,7 başarı 5 ajan IoT ve Bağlı CihazlarHITL ### Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyen Müşteri "cihaz çalışmıyor" der; telemetri okunur, uzaktan tanı konur, güncelleme gönderilir, gerekirse teknisyen doğru parçayla atanır. 37.000 çalıştırma / ay %94,1 başarı 5 ajan BankacılıkHITL ### Kredi kartı limit artışı Müşteri mobilde yazar ya da çağrı merkezinde konuşur; karar saniyeler içinde verilir, eşik üstü yönetici onayına gider. 18.400 çalıştırma / ay %96,4 başarı 4 ajan BankacılıkHITL ### Dijital müşteri edinimi (KYC) Kimlik, canlılık ve yaptırım kontrolü tek akışta; temiz başvurular şubeye gitmeden dakikalarda hesap açar. 42.000 çalıştırma / ay %93,8 başarı 5 ajan BankacılıkHITL ### Kart dolandırıcılık alarmı triyajı Kart saniyeler içinde bloklanır, müşteriye WhatsApp'tan sorulur; yanıta göre kart açılır ya da iptal edilir. 65.000 çalıştırma / ay %97,1 başarı 4 ajan BankacılıkHITL ### Ters ibraz (chargeback) itirazı İtiraz alınır, kanıt paketi toplanır ve kart şemasına dosyalanır; eşik üstü geçici alacak onaya gider. 9.800 çalıştırma / ay %95,2 başarı 4 ajan BankacılıkHITL ### Erken dönem tahsilat iletişimi Gecikmiş hesaplar segmentlenir, doğru kanaldan nazikçe hatırlatılır; müşteri ödeme planını kendisi kurar. 96.000 çalıştırma / ay %91,2 başarı 4 ajan BankacılıkHITL ### Konut kredisi belge incelemesi Bordro, SGK dökümü ve ekspertiz otomatik okunur; gelir, LTV ve eksik belge kontrolü danışman beklemeden biter. 6.200 çalıştırma / ay %94,7 başarı 4 ajan BankacılıkHITL ### Sesli bankacılıkla para transferi Müşteri arar ve konuşur; ses izi, güvenlik sorusu ve OTP ile doğrulanır, transfer temsilcisiz tamamlanır. 27.000 çalıştırma / ay %93,1 başarı 5 ajan BankacılıkHITL ### Kredi kartı borcu ödeme Müşteri yazar ya da konuşur; borç sorgulanır, hesap seçilir, OTP ile doğrulanır, ödeme çekirdek sistemde yapılır. 52.000 çalıştırma / ay %97,8 başarı 4 ajan Finansal TeknolojilerHITL ### Üye işyeri KYB onboarding Sicil, gerçek faydalanıcı ve yaptırım kontrolü tek akışta; düşük riskli işyeri aynı gün ödeme almaya başlar. 1.400 çalıştırma / ay %93,4 başarı 5 ajan Finansal TeknolojilerHITL ### AML işlem izleme alarm triyajı Alarm zenginleştirilir, anlatı taslağı hazır gelir; kararı analist verir, bildirim dakikalarda dosyalanır. 8.500 çalıştırma / ay %95,8 başarı 5 ajan Finansal TeknolojilerHITL ### BNPL anlık taksit kararı Ödemede saniyeler içinde taksit kararı; büro ve açık bankacılık verisi birleşir, limit üstü uzmana gider. 54.000 çalıştırma / ay %97,3 başarı 5 ajan Finansal TeknolojilerHITL ### Ödeme mutabakatı istisnaları PSP mutabakat dosyası ana defterle eşlenir; küçük farklar otomatik düzeltilir, büyükleri finans onayına gider. 620 çalıştırma / ay %95,1 başarı 4 ajan SigortaHITL ### Sesli hasar ihbarı (FNOL) Müşteri arar ve anlatır; poliçe plakadan bulunur, teminat kontrol edilir, dosya açılır ve eksper atanır. 34.000 çalıştırma / ay %95,8 başarı 5 ajan SigortaHITL ### Poliçe yenileme teklifi Poliçeler yeniden fiyatlanır, sert artışlar onaya gider; müşteri tek dokunuşla onaylar, poliçe anında kesilir. 22.000 çalıştırma / ay %94,4 başarı 4 ajan SigortaHITL ### Underwriting belge çıkarımı Sağlık raporu ve tahliller okunur, kılavuza göre değerlendirilir; underwriter yalnızca eşik üstünü görür. 4.800 çalıştırma / ay %94,9 başarı 4 ajan SigortaHITL ### Fotoğrafla kasko hasar tespiti Fotoğraflardan hasar tespit edilir, parça/işçilikle tahmin çıkar, eksper onaylar; servis randevusu alınır. 15.500 çalıştırma / ay %93,3 başarı 4 ajan SigortaHITL ### Rücu ve tahsilat Kapanmış dosyalar rücu için taranır, kusurlu tarafın sigortacısı bulunur, talep avukat onayıyla gönderilir. 3.400 çalıştırma / ay %95,5 başarı 4 ajan Perakende & E-ticaret ### Siparişim nerede? (WISMO) Müşteri sorar, sipariş ve kargo olayları anında birleşir; gecikmede yeni tarih verilir, kargo araştırılır. 110.000 çalıştırma / ay %98,1 başarı 4 ajan Perakende & E-ticaretHITL ### İade ve para iadesi Politika ve müşteri geçmişi saniyede kontrol edilir, etiket üretilir; iade tutarı kargo taramasıyla ödenir. 48.000 çalıştırma / ay %96,7 başarı 4 ajan Perakende & E-ticaret ### Ürün soru-cevap asistanı Beden, uyumluluk ve özellik soruları katalog ve iade verisiyle yanıtlanır; stok ve fiyat doğrulanır. 120.000 çalıştırma / ay %98,7 başarı 4 ajan Perakende & E-ticaretHITL ### Katalog zenginleştirme (toplu) Tedarikçi PDF ve görsellerinden öznitelik çıkarılır, TR/EN içerik yazılır; yüzlerce SKU dakikalarda hazır. 900 çalıştırma / ay %95,8 başarı 4 ajan Perakende & E-ticaretHITL ### Yorum yanıtlama ve kalite eskalasyonu Yorumlar sınıflanır, marka tonunda yanıtlanır; tekrarlayan kusurlar iade verisiyle QA'e Jira kaydı olur. 26.000 çalıştırma / ay %97,2 başarı 4 ajan TelekomHITL ### Fatura itirazı çözümü Fatura kalemleri ve kullanım kayıtları saniyede açıklanır; iyi niyet kredisi politikaya göre uygulanır. 88.000 çalıştırma / ay %96,3 başarı 4 ajan TelekomHITL ### SIM değişimi dolandırıcılık kontrolü Her SIM değişimi provizyondan önce risk skorlanır; şüpheli talepler bekletilir ve bankalara sinyal iletilir. 41.000 çalıştırma / ay %97,7 başarı 4 ajan TelekomHITL ### Şebeke kesintisi iletişimi Kesinti kaydı açılınca etkilenen aboneler bulunur, mesaj yazılıp onaylanır; müşteri aramadan bilgilendirilir. 1.400 çalıştırma / ay %98,5 başarı 4 ajan TelekomHITL ### Tarife yükseltme danışmanı Kullanım analiz edilir, doğru tarife karşılaştırmalı sunulur; onayla değişiklik anında provizyonlanır. 52.000 çalıştırma / ay %95,8 başarı 4 ajan Yazılım GeliştirmeHITL ### Özellik teslimat hattı Bir Jira epic'i ajanlarla kriter, tasarım, kod ve teste dönüşür; insan yalnızca PR'ı inceler ve onaylar. 1.200 çalıştırma / ay %91,2 başarı 5 ajan Yazılım GeliştirmeHITL ### CI/CD sürüm hattı Derleme, test, tarama ve staging otomatik; üretime yalnızca sürüm yöneticisi onayıyla ArgoCD ile çıkılır. 950 çalıştırma / ay %94,7 başarı 5 ajan Yazılım GeliştirmeHITL ### Olay müdahalesi ve acil düzeltme Alarm gelir, hata yeniden üretilir, yama ve regresyon hazır; nöbetçi onaylar, post-mortem taslağı bekler. 320 çalıştırma / ay %90,5 başarı 5 ajan Yazılım GeliştirmeHITL ### Kod incelemesi ve güvenlik kapısı Her PR ajanlarca incelenir, taranır, kapsamı ölçülür; geliştirici düzeltir, insan yalnızca temiz PR'ı onaylar. 4.200 çalıştırma / ay %93,6 başarı 5 ajan Yazılım GeliştirmeHITL ### Backlog düzenleme ve tahminleme Talepler triyajlanır, netleştirilir, karmaşıklığı tahminlenir; sprint önerisi ürün sahibinin onayına gider. 340 çalıştırma / ay %92,8 başarı 4 ajan Yazılım GeliştirmeHITL ### iOS özellik geliştirme (kiralık geliştirici ajanı) Jira hikâyesi alınır, SwiftUI ile geliştirilir, testler yazılır, PR açılır; inceleme ve dağıtım kararı ekipte kalır. 380 çalıştırma / ay %94,4 başarı 4 ajan Yazılım GeliştirmeHITL ### Regresyon test döngüsü (kiralık test ajanı) Her sürüm adayında regresyon koşulur, yeni hatalar kanıtla açılır, kapsama raporu yazılır; “go” kararı sürüm yöneticisinde. 260 çalıştırma / ay %95,8 başarı 4 ajan SağlıkHITL ### Ön yetkilendirme (prior authorization) İstem düşer düşmez ödeyici kuralları taranır, klinik kanıt toplanır, dosya dakikalar içinde gönderime hazır. 6.200 çalıştırma / ay %94,1 başarı 4 ajan SağlıkHITL ### Hasta ön kabul ve triyaj Hasta yazar ya da anlatır; anamnez ve aciliyet skoru saniyeler sürer, son kararı her zaman hemşire verir. 24.500 çalıştırma / ay %95,8 başarı 5 ajan Sağlık ### Randevu planlama ve no-show azaltma Gelmeme riski yüksek hastalar WhatsApp'tan önceden aranır, boşalan slotlar bekleme listesinden anında dolar. 62.000 çalıştırma / ay %98,2 başarı 4 ajan Üretim & OtomotivHITL ### Tedarikçi faturası üçlü eşleştirme Fatura, sipariş ve mal kabul SAP'de otomatik eşleşir; yalnızca tolerans dışı kalemler satın almacıya gider. 21.000 çalıştırma / ay %97,1 başarı 4 ajan Üretim & OtomotivHITL ### Bakım talebi triyajı Sensör alarmı ve teknisyen sesli notu tek iş emrine dönüşür; yedek parça ve güvenli duruş penceresi hazırdır. 3.400 çalıştırma / ay %95,2 başarı 5 ajan Üretim & OtomotivHITL ### Garanti talebi işleme Bayi talebi dakikalarda kapsam, parça ve işçilik yönünden doğrulanır; eşik üstü tutar bölge müdürüne gider. 8.600 çalıştırma / ay %96,3 başarı 4 ajan LojistikHITL ### Sevkiyat istisna yönetimi Gecikme ya da hasar geldiğinde etki hesaplanır, taşıyıcı yeniden planlanır, müşteri daha sormadan bilgilenir. 12.800 çalıştırma / ay %95,5 başarı 4 ajan LojistikHITL ### Gümrük belgesi hazırlama GTİP kodu, fatura ve çeki listesi dakikalarda hazırlanır; düşük güvenli satırlar gümrük uzmanına gider. 4.300 çalıştırma / ay %94,7 başarı 4 ajan EnerjiHITL ### Kesinti bildirimi yönetimi Bildirimler fider bazında kümelenir, tek OSS kaydı açılır, etkilenen herkes ekip yola çıkmadan bilgilenir. 38.000 çalıştırma / ay %97,7 başarı 5 ajan EnerjiHITL ### Sayaç okuma itirazı Müşteri sayaç fotoğrafı yollar; endeks görüntüden okunur, tahmini fatura düzeltilir, büyük iade lidere gider. 9.200 çalıştırma / ay %94,9 başarı 4 ajan EnerjiHITL ### Çatı GES bağlantı başvurusu Başvuru belgeleri okunur, trafo kapasitesi ve mevzuat kontrol edilir, eksikler aynı gün bildirilir; bağlantı görüşü mühendisin. 5.400 çalıştırma / ay %95,3 başarı 5 ajan EnerjiHITL ### Talep yanıtı (demand response) olayı Şebeke sıkışıklığında uygun tüketiciler seçilir, teklif kanaldan gönderilir, katılım ölçülür ve ödeme hesaplanır. 140 çalıştırma / ay %96,2 başarı 4 ajan Kamu ve HükümetHITL ### Vatandaş başvurusu yönlendirme e-Devlet başvurusu saniyeler içinde sınıflandırılır, yasal süresi hesaplanır ve doğru birime SLA ile atanır. 54.000 çalıştırma / ay %98,1 başarı 4 ajan Kamu ve HükümetHITL ### Ruhsat başvurusu ön inceleme Eksik belge ve mevzuat uyumu dakikalar içinde raporlanır; kararı her zaman ruhsat memuru verir. 3.100 çalıştırma / ay %95,8 başarı 4 ajan Kamu ve HükümetHITL ### Kamu ihalesi teklif dosyası ön kontrolü Teklif dosyaları şartnameyle karşılaştırılır, eksik ve uygunsuz belgeler listelenir; değerlendirme kararı komisyonun. 380 çalıştırma / ay %94,8 başarı 4 ajan Kamu ve HükümetHITL ### Sosyal yardım başvurusu ön değerlendirme Başvuru kurum kayıtlarıyla doğrulanır, uygunluk kriterleri puanlanır, eksikler vatandaşa bildirilir; karar sosyal inceleme görevlisinin. 21.000 çalıştırma / ay %96,6 başarı 4 ajan Kurumsal Fonksiyonlar ### E-posta sınıflandırma ve ek kontrolü Ortak kutuya gelen her e-posta sınıflandırılır, ekler listeyle karşılaştırılır, eksikler anında sorulur. 86.000 çalıştırma / ay %97,6 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Çağrı merkezi kalite izleme Örneklem değil, tüm çağrılar rubriğe göre puanlanır; uyum ihlali ve koçluk fırsatı sabah liderin önünde. 90.000 çalıştırma / ay %98,8 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Sözleşme madde çıkarımı Sözleşme maddeleri çıkarılır, hukuk playbook'u ile karşılaştırılır; sapmalar redline önerisiyle avukata gider. 1.400 çalıştırma / ay %95,1 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Çalışan işe alıştırma (onboarding) İK kaydı açıldığı anda hesaplar, ekipman ve eğitim planı hazırlanır; şablon dışı erişimler yöneticiye sorulur. 1.900 çalıştırma / ay %97,2 başarı 5 ajan Kurumsal FonksiyonlarHITL ### BT yardım masası L1 Şifre sıfırlama ve erişim talepleri sohbette çözülür; finans grubu erişimi grup sahibine Teams'te sorulur. 47.000 çalıştırma / ay %98,5 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Borçlar hesabı (AP) posta kutusu ap@ kutusuna düşen her fatura yakalanır, kodlanır ve ERP'ye kaydedilir; yalnızca istisnalar analiste gider. 28.000 çalıştırma / ay %96,8 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Satış fırsatı nitelendirme Gelen talep saniyeler içinde zenginleştirilir, skorlanır ve CRM'e işlenir; SDR aramaya bağlam notuyla başlar. 15.600 çalıştırma / ay %97,9 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Ay sonu kapanış ve mutabakat (kiralık muhasebe ajanı) Faturalar işlenir, banka ve tedarikçi mutabakatı yapılır, kapanış listesi yürütülür; kayıtlar mali müşavir onayıyla deftere işlenir. 60 çalıştırma / ay %96,2 başarı 4 ajan Savunma SanayiiHITL ### İhracat kontrolü ve son kullanıcı taraması Parça, alıcı ve son kullanım tek akışta sınıflandırılır; lisans gereği tespit edilir, karar her zaman uyum yöneticisinde kalır. 2.600 çalıştırma / ay %96,1 başarı 4 ajan Savunma SanayiiHITL ### Platform bakım (MRO) iş emri hazırlığı Arıza raporu teknik yayınlarla eşlenir, görev kartı ve parça listesi çıkar, iş emri planlayıcı onayıyla açılır. 4.100 çalıştırma / ay %95,7 başarı 5 ajan Savunma SanayiiHITL ### Tedarikçi güvenlik ve uyum değerlendirmesi Tedarikçi anketleri ve belgeleri okunur, tesis güvenlik gereklilikleriyle karşılaştırılır, eksikler listelenir; nitelendirme kararı güvenlik müdürünün. 900 çalıştırma / ay %95,2 başarı 4 ajan IoT ve Bağlı CihazlarHITL ### Cihaz filosu anomali triyajı Binlerce cihazdan gelen telemetri anomalileri kümelenir, kök neden aday listesi çıkar, doğru ekibe tek olay olarak düşer. 18.000 çalıştırma / ay %96,9 başarı 4 ajan IoT ve Bağlı CihazlarHITL ### Kademeli OTA güncelleme dağıtımı Firmware dağıtımı %1 → %10 → %100 kademelerinde ilerler; her kademede sağlık kapısı, kritik filolarda insan onayı. 320 çalıştırma / ay %98,1 başarı 4 ajan IoT ve Bağlı CihazlarHITL ### Soğuk zincir sapması müdahalesi Sıcaklık sapması anında değerlendirilir, ürün riski hesaplanır, saha ekibi sevk edilir, kalite kararı kayıt altına alınır. 2.700 çalıştırma / ay %97,4 başarı 4 ajan Simülasyonlar gerçek ajan tanımlarından türetilir; sayılar temsilidir. Referans kurumlarımızın gerçek metriklerini keşif oturumunda paylaşırız. ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Borçlar hesabı (AP) posta kutusu URL: https://www.aginies.com/tr/ajanlar/accounts-payable-inbox Language: tr Summary: ap@ kutusuna düşen her fatura yakalanır, kodlanır ve ERP'ye kaydedilir; yalnızca istisnalar analiste gider. ap@ posta kutusuna PDF fatura içeren e-posta → Fatura ERP'de kaydedildi, arşivlendi, tedarikçi bilgilendirildi, istisnalar kuyrukta Kurumsal Fonksiyonlaremailbatch # Borçlar hesabı (AP) posta kutusu ap@ kutusuna düşen her fatura yakalanır, kodlanır ve ERP'ye kaydedilir; yalnızca istisnalar analiste gider. Kısaca Borçlar hesabı (AP) posta kutusu, Kurumsal Fonksiyonlar sektöründe email, batch kanalında çalışan bir ajan senaryosudur. Tetikleyici: ap@ posta kutusuna PDF fatura içeren e-posta. AP Süpervizörü 3 uzman ajanı (Yakalama Ajanı, Kodlama Ajanı, Kayıt Ajanı) 4 sistem üzerinde yönetir (Exchange / M365 Mail, Oracle EBS, Document Store (S3), Vendor Master / Procurement); 9 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Fatura ERP'de kaydedildi, arşivlendi, tedarikçi bilgilendirildi, istisnalar kuyrukta. Onay kuralı: Sipariş (SAS) olmayan fatura > 25.000 TL → masraf yeri sahibi onayı. Kurumsal Fonksiyonlar ### Borçlar hesabı (AP) posta kutusu [Borçlar hesabı (AP) posta kutusu] çalıştırma #41.287.359süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Sipariş (SAS) olmayan fatura > 25.000 TL → masraf yeri sahibi onayı Bu çalıştırmadaki ajanlar AP Süpervizörüsupervisorplanlar, yönlendirir, onaylar Yakalama Ajanıaginies-konrul-1 · on-premOCR ile alanları çıkarır, mükerreri yakalar Kodlama Ajanıaginies-tulpar-4 · on-premHesap, masraf yeri ve KDV kodlamasını önerir Kayıt Ajanıaginies-talay-2 · on-premERP'ye kaydeder, arşivler, bildirir Denetim izi00:00.00 - ◦ap@ kutusu: gece 312 e-posta; 287 PDF fatura, 25 hesap ekstresi - ∿Yıldız Lojistik faturası INV-77120: 42.600 TL + %18 KDV, SAS yok (hizmet) - ✓Mükerrer: 180 günde no/tutar/tarih eşleşmesi yok; IBAN tedarikçi kartıyla uyumlu - ⚙Tedarikçi 20418 aktif, vade 45 gün, "navlun 2026" sözleşmesi kayıtlı - ∿Hesap 6300 (navlun) · masraf yeri 2210 · KDV %18 indirilebilir · güven 0,91 - ⏸SAS'sız fatura > 25.000 TL → masraf yeri sahibi aynı gün onayladı - ⇢Fatura kaydedildi (AP-2026-118440), vade 23 Ekim, ödeme koşusu 21 Ekim - ⇢PDF ve çıkarım JSON'u arşivlendi, 10 yıl saklama bayrağı kondu - ➤Tedarikçiye teyit gönderildi; 287 faturadan 14 istisna AP analist kuyruğuna alındı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici ap@ posta kutusuna PDF fatura içeren e-posta Sonuç Fatura ERP'de kaydedildi, arşivlendi, tedarikçi bilgilendirildi, istisnalar kuyrukta ## Ajanlar SÜPERVİZÖR ### AP Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### Yakalama Ajanı OCR ile alanları çıkarır, mükerreri yakalar ocr_documentextract_invoice_fieldsdetect_duplicate aginies-tulpar-4 · on-prem ### Kodlama Ajanı Hesap, masraf yeri ve KDV kodlamasını önerir get_vendor_mastersuggest_gl_accountget_cost_centrecheck_vat aginies-talay-2 · on-prem ### Kayıt Ajanı ERP'ye kaydeder, arşivler, bildirir post_invoicepark_invoicearchive_documentssend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | AP Süpervizörü | ap@ kutusu: gece 312 e-posta; 287 PDF fatura, 25 hesap ekstresi | — | 300 | — | | 02 | akıl yürütme | Yakalama Ajanı | Yıldız Lojistik faturası INV-77120: 42.600 TL + %18 KDV, SAS yok (hizmet) | — | 860 | 680 | | 03 | doğrulama | Yakalama Ajanı | Mükerrer: 180 günde no/tutar/tarih eşleşmesi yok; IBAN tedarikçi kartıyla uyumlu | — | 380 | 220 | | 04 | araç | Kodlama Ajanı | Tedarikçi 20418 aktif, vade 45 gün, "navlun 2026" sözleşmesi kayıtlı | Vendor Master / Procurement | 400 | — | | 05 | akıl yürütme | Kodlama Ajanı | Hesap 6300 (navlun) · masraf yeri 2210 · KDV %18 indirilebilir · güven 0,91 | — | 720 | 760 | | 06 | insan onayı | AP Süpervizörü | SAS'sız fatura > 25.000 TL → masraf yeri sahibi aynı gün onayladı | — | 2.600 | — | | 07 | yazma | Kayıt Ajanı | Fatura kaydedildi (AP-2026-118440), vade 23 Ekim, ödeme koşusu 21 Ekim | Oracle EBS | 540 | — | | 08 | yazma | Kayıt Ajanı | PDF ve çıkarım JSON'u arşivlendi, 10 yıl saklama bayrağı kondu | Document Store (S3) | 260 | — | | 09 | bildirim | Kayıt Ajanı | Tedarikçiye teyit gönderildi; 287 faturadan 14 istisna AP analist kuyruğuna alındı | Exchange / M365 Mail | 240 | — | Simülasyon metrikleri (tahmini) 28.000çalıştırma / ay %96,8başarı 6,3 sp50 $0,0026tahmini maliyet / çalıştırma Onay kuralı Sipariş (SAS) olmayan fatura > 25.000 TL → masraf yeri sahibi onayı Adım profili giriş · 1akıl yürütme · 2araç · 1doğrulama · 1insan onayı · 1yazma · 2bildirim · 1 SİSTEMLER - Exchange / M365 Mailemail - Oracle EBSerp - Document Store (S3)storage - Vendor Master / Procurementdb Çalıştırmak için gerekenler - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - ERP/MES için kayıt ve sorgu API erişimi - Doküman deposuna okuma erişimi ve belge tipleri - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Küresel ortak hizmet merkezlerindeki AP posta kutusu otomasyonunu yansıtır Bu senaryonun yer aldığı paketler Fatura ve doküman veri çıkarımı →E-posta sınıflandırma ve ek doğrulama → İlgili ajanlar Kurumsal FonksiyonlarHITL ### Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada Talep e-postayla gelir; teklifler karşılaştırılır, sözleşme maddeleri kontrol edilir, sipariş açılır, fatura üç yönlü eşleşir ve ödeme planlanır. 14.000 çalıştırma / ay %92,7 başarı 5 ajan Kurumsal Fonksiyonlar ### E-posta sınıflandırma ve ek kontrolü Ortak kutuya gelen her e-posta sınıflandırılır, ekler listeyle karşılaştırılır, eksikler anında sorulur. 86.000 çalıştırma / ay %97,6 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Çağrı merkezi kalite izleme Örneklem değil, tüm çağrılar rubriğe göre puanlanır; uyum ihlali ve koçluk fırsatı sabah liderin önünde. 90.000 çalıştırma / ay %98,8 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # AML işlem izleme alarm triyajı URL: https://www.aginies.com/tr/ajanlar/aml-transaction-monitoring Language: tr Summary: Alarm zenginleştirilir, anlatı taslağı hazır gelir; kararı analist verir, bildirim dakikalarda dosyalanır. İzleme motorundan kural alarmı (yapılandırma / hızlı geçiş) → Alarm kararlı, vaka kapalı ya da STR MASAK'a dosyalandı, anlatı ve kanıt izi kayıtlı Finansal Teknolojilerapibatch # AML işlem izleme alarm triyajı Alarm zenginleştirilir, anlatı taslağı hazır gelir; kararı analist verir, bildirim dakikalarda dosyalanır. Kısaca AML işlem izleme alarm triyajı, Finansal Teknolojiler sektöründe api, batch kanalında çalışan bir ajan senaryosudur. Tetikleyici: İzleme motorundan kural alarmı (yapılandırma / hızlı geçiş). AML Süpervizörü 4 uzman ajanı (Zenginleştirme Ajanı, Tarama Ajanı, Anlatı Ajanı, Dosyalama Ajanı) 4 sistem üzerinde yönetir (Core Ledger, Sanctions Screening, Case Management, MASAK Reporting Portal); 10 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Alarm kararlı, vaka kapalı ya da STR MASAK'a dosyalandı, anlatı ve kanıt izi kayıtlı. Onay kuralı: Her alarm → uyum analisti kararı zorunlu; ajan tek başına STR dosyalamaz, vaka kapatmaz. Finansal Teknolojiler ### AML işlem izleme alarm triyajı [AML işlem izleme alarm triyajı] çalıştırma #41.287.205süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Her alarm → uyum analisti kararı zorunlu; ajan tek başına STR dosyalamaz, vaka kapatmaz Bu çalıştırmadaki ajanlar AML Süpervizörüsupervisorplanlar, yönlendirir, onaylar Zenginleştirme Ajanıaginies-talay-2 · on-premMüşteri profili, 90 günlük işlem ve karşı taraf verisini toplar Tarama Ajanıaginies-sunkar-1 · on-premKarşı tarafları yaptırım/PEP listelerinde tarar, önceki vakaları bulur Anlatı Ajanıaginies-tulpar-4 · on-premTipolojiyi değerlendirir, STR anlatı taslağını yazar Dosyalama Ajanıaginies-talay-2 · on-premVakayı günceller, MASAK bildirimini dosyalar Denetim izi00:00.00 - ◦Alarm #TM-51290: 6 günde 9 nakit yatırma, toplam 1.140.000 TL, kural R-14 - ⚙Müşteri profili ve 90 günlük işlem geçmişi çekildi: 47 işlem, 3 karşı taraf - ⌘Plan: karşı taraf taraması → tipoloji → anlatı taslağı → analist kararı → dosyalama - ⚙3 karşı taraf tarandı: yaptırım eşleşmesi yok, 1 PEP bağlantısı (akraba) - ⌕Önceki vakalar: aynı müşteri için 2024'te kapatılmış 1 alarm (yanlış pozitif) - ∿Tipoloji: eşik altı parçalama, beyan edilen gelirin 4 katı → şüphe skoru 0,82 - ✓Anlatı taslağı 5 kanıt referansıyla hazırlandı, MASAK format kontrolü geçti - ⏸Zorunlu analist kararı: uyum analisti anlatıyı düzenledi, "STR dosyala" dedi - ⇢STR MASAK portalına dosyalandı, referans #STR-2026-08812 - ⇢Vaka #AML-7734 "dosyalandı" durumuna alındı, kanıt paketi ve karar izi eklendi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici İzleme motorundan kural alarmı (yapılandırma / hızlı geçiş) Sonuç Alarm kararlı, vaka kapalı ya da STR MASAK'a dosyalandı, anlatı ve kanıt izi kayıtlı ## Ajanlar SÜPERVİZÖR ### AML Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Zenginleştirme Ajanı Müşteri profili, 90 günlük işlem ve karşı taraf verisini toplar get_alertget_customer_profileget_transactionsget_counterparties aginies-sunkar-1 · on-prem ### Tarama Ajanı Karşı tarafları yaptırım/PEP listelerinde tarar, önceki vakaları bulur screen_sanctionsscreen_pepsearch_prior_cases aginies-tulpar-4 · on-prem ### Anlatı Ajanı Tipolojiyi değerlendirir, STR anlatı taslağını yazar score_typologydraft_narrativesummarise_evidence aginies-talay-2 · on-prem ### Dosyalama Ajanı Vakayı günceller, MASAK bildirimini dosyalar update_casefile_strnotify_analyst ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | AML Süpervizörü | Alarm #TM-51290: 6 günde 9 nakit yatırma, toplam 1.140.000 TL, kural R-14 | — | 180 | — | | 02 | araç | Zenginleştirme Ajanı | Müşteri profili ve 90 günlük işlem geçmişi çekildi: 47 işlem, 3 karşı taraf | Core Ledger | 520 | — | | 03 | plan | AML Süpervizörü | Plan: karşı taraf taraması → tipoloji → anlatı taslağı → analist kararı → dosyalama | — | 300 | 220 | | 04 | araç | Tarama Ajanı | 3 karşı taraf tarandı: yaptırım eşleşmesi yok, 1 PEP bağlantısı (akraba) | Sanctions Screening | 560 | — | | 05 | bilgi | Tarama Ajanı | Önceki vakalar: aynı müşteri için 2024'te kapatılmış 1 alarm (yanlış pozitif) | Case Management | 410 | — | | 06 | akıl yürütme | Anlatı Ajanı | Tipoloji: eşik altı parçalama, beyan edilen gelirin 4 katı → şüphe skoru 0,82 | — | 1.200 | 1.600 | | 07 | doğrulama | Anlatı Ajanı | Anlatı taslağı 5 kanıt referansıyla hazırlandı, MASAK format kontrolü geçti | — | 900 | 1.100 | | 08 | insan onayı | AML Süpervizörü | Zorunlu analist kararı: uyum analisti anlatıyı düzenledi, "STR dosyala" dedi | — | 3.800 | — | | 09 | yazma | Dosyalama Ajanı | STR MASAK portalına dosyalandı, referans #STR-2026-08812 | MASAK Reporting Portal | 740 | — | | 10 | yazma | Dosyalama Ajanı | Vaka #AML-7734 "dosyalandı" durumuna alındı, kanıt paketi ve karar izi eklendi | Case Management | 460 | — | Simülasyon metrikleri (tahmini) 8.500çalıştırma / ay %95,8başarı 9,2 sp50 $0,0093tahmini maliyet / çalıştırma Onay kuralı Her alarm → uyum analisti kararı zorunlu; ajan tek başına STR dosyalamaz, vaka kapatmaz Adım profili giriş · 1plan · 1bilgi · 1akıl yürütme · 1araç · 2doğrulama · 1insan onayı · 1yazma · 2 SİSTEMLER - Core Ledgercore - Sanctions Screeningapi - Case Managementticketing - MASAK Reporting Portalapi Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Küresel banka ve fintech'lerin AML alarm zenginleştirme ve anlatı taslağı deseni, analist kararı merkezde Bu senaryonun yer aldığı paketler Üye işyeri ve ödeme operasyonları → Sektör: Finansal Teknolojiler → İlgili ajanlar Finansal TeknolojilerHITL ### Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit İşyeri WhatsApp'tan yazar; POS arızası, hakediş sorgusu, iade ve limit artışı tek diyalogda çözülür. 64.000 çalıştırma / ay %93,8 başarı 5 ajan Finansal TeknolojilerHITL ### Üye işyeri KYB onboarding Sicil, gerçek faydalanıcı ve yaptırım kontrolü tek akışta; düşük riskli işyeri aynı gün ödeme almaya başlar. 1.400 çalıştırma / ay %93,4 başarı 5 ajan Finansal TeknolojilerHITL ### BNPL anlık taksit kararı Ödemede saniyeler içinde taksit kararı; büro ve açık bankacılık verisi birleşir, limit üstü uzmana gider. 54.000 çalıştırma / ay %97,3 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Randevu planlama ve no-show azaltma URL: https://www.aginies.com/tr/ajanlar/appointment-scheduling-noshow Language: tr Summary: Gelmeme riski yüksek hastalar WhatsApp'tan önceden aranır, boşalan slotlar bekleme listesinden anında dolar. Gece toplu işi: sonraki 72 saatin randevu listesi → Randevu taşındı ya da teyit edildi, eski slot bekleme listesine açıldı, hatırlatma planlandı Sağlıkwhatsappsmsbatch # Randevu planlama ve no-show azaltma Gelmeme riski yüksek hastalar WhatsApp'tan önceden aranır, boşalan slotlar bekleme listesinden anında dolar. Kısaca Randevu planlama ve no-show azaltma, Sağlık sektöründe whatsapp, sms, batch kanalında çalışan bir ajan senaryosudur. Tetikleyici: Gece toplu işi: sonraki 72 saatin randevu listesi. Randevu Süpervizörü 3 uzman ajanı (Gelmeme Riski Ajanı, Diyalog Ajanı, Rezervasyon Ajanı) 3 sistem üzerinde yönetir (Hospital Information System (HIS), WhatsApp Business API, SMS Gateway); 9 adımın 1 tanesi sistemde işlem, 1 tanesi doğrulama veya onaydır. Sonuç: Randevu taşındı ya da teyit edildi, eski slot bekleme listesine açıldı, hatırlatma planlandı. Sağlık ### Randevu planlama ve no-show azaltma [Randevu planlama ve no-show azaltma] çalıştırma #41.287.777süre 0 mstoken 0tahmini maliyet $0,000durum — Bu çalıştırmadaki ajanlar Randevu Süpervizörüsupervisorplanlar, yönlendirir, onaylar Gelmeme Riski Ajanıaginies-sunkar-1 · on-premGeçmiş verilerle gelmeme riskini tahmin eder Diyalog Ajanıaginies-sunkar-1 · on-premWhatsApp'ta konuşur, erteleme niyetini anlar Rezervasyon Ajanıaginies-talay-2 · on-premHIS'te uygun slotu bulur ve randevuyu taşır Denetim izi00:00.00 - ◦Gece toplu işi: sonraki 72 saatteki 1.240 randevu yüklendi - ⚙Geçmiş çekildi: hasta P-77120, son 12 ayda 2 gelmeme - ∿Gelmeme riski 0,62 (eşik 0,40) → proaktif teyit kanalı seçildi - ➤WhatsApp gönderildi: "Dr. Aksoy, Perş 10:20 — teyit / erteleme?" - ∿Yanıt: "Perşembe olmaz, haftaya?" → niyet: erteleme, tercih: haftaya sabah - ⚙3 boş slot bulundu: Pzt 09:00, Sal 09:40, Çar 11:20 (aynı hekim) - ✓Kural: aynı hekim, ≤ 14 gün gecikme, hazırlık çakışması yok → uygun - ⇢Randevu Salı 09:40'a taşındı, eski slot bekleme listesine açıldı - ➤Teyit SMS'i gönderildi, 24 saat hatırlatması kuruldu, 12 slot listeden dolduruldu - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Gece toplu işi: sonraki 72 saatin randevu listesi Sonuç Randevu taşındı ya da teyit edildi, eski slot bekleme listesine açıldı, hatırlatma planlandı ## Ajanlar SÜPERVİZÖR ### Randevu Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Gelmeme Riski Ajanı Geçmiş verilerle gelmeme riskini tahmin eder get_appointment_historyscore_noshow_risk aginies-sunkar-1 · on-prem ### Diyalog Ajanı WhatsApp'ta konuşur, erteleme niyetini anlar send_whatsappparse_replypropose_slots aginies-talay-2 · on-prem ### Rezervasyon Ajanı HIS'te uygun slotu bulur ve randevuyu taşır find_free_slotsreschedule_appointmentrelease_slotsend_sms ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Randevu Süpervizörü | Gece toplu işi: sonraki 72 saatteki 1.240 randevu yüklendi | — | 300 | — | | 02 | araç | Gelmeme Riski Ajanı | Geçmiş çekildi: hasta P-77120, son 12 ayda 2 gelmeme | Hospital Information System (HIS) | 420 | — | | 03 | akıl yürütme | Gelmeme Riski Ajanı | Gelmeme riski 0,62 (eşik 0,40) → proaktif teyit kanalı seçildi | — | 540 | 380 | | 04 | bildirim | Diyalog Ajanı | WhatsApp gönderildi: "Dr. Aksoy, Perş 10:20 — teyit / erteleme?" | WhatsApp Business API | 260 | — | | 05 | akıl yürütme | Diyalog Ajanı | Yanıt: "Perşembe olmaz, haftaya?" → niyet: erteleme, tercih: haftaya sabah | — | 620 | 410 | | 06 | araç | Rezervasyon Ajanı | 3 boş slot bulundu: Pzt 09:00, Sal 09:40, Çar 11:20 (aynı hekim) | Hospital Information System (HIS) | 480 | — | | 07 | doğrulama | Rezervasyon Ajanı | Kural: aynı hekim, ≤ 14 gün gecikme, hazırlık çakışması yok → uygun | — | 380 | 240 | | 08 | yazma | Rezervasyon Ajanı | Randevu Salı 09:40'a taşındı, eski slot bekleme listesine açıldı | Hospital Information System (HIS) | 520 | — | | 09 | bildirim | Rezervasyon Ajanı | Teyit SMS'i gönderildi, 24 saat hatırlatması kuruldu, 12 slot listeden dolduruldu | SMS Gateway | 240 | — | Simülasyon metrikleri (tahmini) 62.000çalıştırma / ay %98,2başarı 3,8 sp50 $0,0008tahmini maliyet / çalıştırma Adım profili giriş · 1akıl yürütme · 2araç · 2doğrulama · 1yazma · 1bildirim · 2 SİSTEMLER - Hospital Information System (HIS)core - WhatsApp Business APIchannel - SMS Gatewaychannel Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük hastane ağlarının WhatsApp tabanlı hatırlatma ve gelmeme azaltma programlarını yansıtır Bu senaryonun yer aldığı paketler Sağlık operasyonları ve klinik yönlendirme → Sektör: Sağlık → İlgili ajanlar SağlıkHITL ### Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuç Hasta yazar ya da arar; randevu, ön hazırlık talimatı, sigorta ön onayı ve sonuç bildirimi tek koordinatörde birleşir. 58.000 çalıştırma / ay %95,1 başarı 5 ajan SağlıkHITL ### Ön yetkilendirme (prior authorization) İstem düşer düşmez ödeyici kuralları taranır, klinik kanıt toplanır, dosya dakikalar içinde gönderime hazır. 6.200 çalıştırma / ay %94,1 başarı 4 ajan SağlıkHITL ### Hasta ön kabul ve triyaj Hasta yazar ya da anlatır; anamnez ve aciliyet skoru saniyeler sürer, son kararı her zaman hemşire verir. 24.500 çalıştırma / ay %95,8 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Backlog düzenleme ve tahminleme URL: https://www.aginies.com/tr/ajanlar/backlog-grooming-estimation Language: tr Summary: Talepler triyajlanır, netleştirilir, karmaşıklığı tahminlenir; sprint önerisi ürün sahibinin onayına gider. Haftalık toplu iş: Jira'da "Triyaj" sütunundaki yeni talepler → Talepler puanlandı ve önceliklendi, sprint 42 önerisi onaylandı, Jira ve Slack güncel Yazılım Geliştirmebatchwebapi # Backlog düzenleme ve tahminleme Talepler triyajlanır, netleştirilir, karmaşıklığı tahminlenir; sprint önerisi ürün sahibinin onayına gider. Kısaca Backlog düzenleme ve tahminleme, Yazılım Geliştirme sektöründe batch, web, api kanalında çalışan bir ajan senaryosudur. Tetikleyici: Haftalık toplu iş: Jira'da "Triyaj" sütunundaki yeni talepler. Planlama Süpervizörü 3 uzman ajanı (Proje Yöneticisi Ajanı, Analist Ajanı, Mimar Ajanı) 4 sistem üzerinde yönetir (Jira, Confluence, GitHub, Slack / Teams); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Talepler puanlandı ve önceliklendi, sprint 42 önerisi onaylandı, Jira ve Slack güncel. Onay kuralı: Sprint kapsamı → ürün sahibi onayı; ajan sprint'e talep atayamaz, yalnızca önerir. Yazılım Geliştirme ### Backlog düzenleme ve tahminleme [Backlog düzenleme ve tahminleme] çalıştırma #41.286.957süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Sprint kapsamı → ürün sahibi onayı; ajan sprint'e talep atayamaz, yalnızca önerir Bu çalıştırmadaki ajanlar Planlama Süpervizörüsupervisorplanlar, yönlendirir, onaylar Proje Yöneticisi Ajanıaginies-tulpar-4 · on-premTalepleri triyajlar, mükerrerleri birleştirir, sprint önerisi hazırlar Analist Ajanıaginies-tulpar-4 · on-premBelirsiz talepleri netleştirir, kabul kriterlerini yazar Mimar Ajanıaginies-tulpar-4 · on-premKarmaşıklığı ve riski tahminler, hikâye puanı önerir Denetim izi00:00.00 - ◦Haftalık triyaj: 27 yeni talep (14 özellik, 9 hata, 4 teknik borç) - ∿3 mükerrer birleştirildi (PROJ-4901 ↔ 4887), 24 talep sıralandı; hız 46 puan/sprint - ⌘Plan: belirsiz talepleri netleştir → tahminle → sprint öner → ürün sahibi onayı - ⌕PRD ve önceki kararlar okundu; 6 talep için netleştirme sorusu hazırlandı - ➤#product kanalında 6 soru soruldu, 5 yanıt geldi; PROJ-4910 beklemeye alındı - ⚙Etkilenen modüller ve değişiklik sıklığı çekildi: 4 talep checkout'a dokunuyor - ∿Tahmin: 23 talep = 118 puan; 2 talep 13 puan üstü → parçalama önerildi - ✓Sprint 42 önerisi: 11 talep, 44 puan (hız 46), bağımlılık çakışması yok - ⏸Ürün sahibi öneriyi inceledi, 1 talebi değiştirdi, sprint 42'yi onayladı - ⇢Sprint 42 oluşturuldu, 11 talep puanlarıyla taşındı, kalanı önceliklendirildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Haftalık toplu iş: Jira'da "Triyaj" sütunundaki yeni talepler Sonuç Talepler puanlandı ve önceliklendi, sprint 42 önerisi onaylandı, Jira ve Slack güncel ## Ajanlar SÜPERVİZÖR ### Planlama Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-tulpar-4 · on-prem ### Proje Yöneticisi Ajanı Talepleri triyajlar, mükerrerleri birleştirir, sprint önerisi hazırlar list_jira_issuesdetect_duplicatesrank_by_valuepropose_sprintupdate_jira_issue aginies-tulpar-4 · on-prem ### Analist Ajanı Belirsiz talepleri netleştirir, kabul kriterlerini yazar search_confluenceask_clarifying_questionwrite_acceptance_criteria aginies-tulpar-4 · on-prem ### Mimar Ajanı Karmaşıklığı ve riski tahminler, hikâye puanı önerir search_codebaseestimate_complexityflag_dependencies ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Planlama Süpervizörü | Haftalık triyaj: 27 yeni talep (14 özellik, 9 hata, 4 teknik borç) | — | 240 | — | | 02 | akıl yürütme | Proje Yöneticisi Ajanı | 3 mükerrer birleştirildi (PROJ-4901 ↔ 4887), 24 talep sıralandı; hız 46 puan/sprint | — | 1.800 | 3.600 | | 03 | plan | Planlama Süpervizörü | Plan: belirsiz talepleri netleştir → tahminle → sprint öner → ürün sahibi onayı | — | 280 | 200 | | 04 | bilgi | Analist Ajanı | PRD ve önceki kararlar okundu; 6 talep için netleştirme sorusu hazırlandı | Confluence | 1.400 | 2.800 | | 05 | bildirim | Analist Ajanı | #product kanalında 6 soru soruldu, 5 yanıt geldi; PROJ-4910 beklemeye alındı | Slack / Teams | 3.200 | 900 | | 06 | araç | Mimar Ajanı | Etkilenen modüller ve değişiklik sıklığı çekildi: 4 talep checkout'a dokunuyor | GitHub | 620 | — | | 07 | akıl yürütme | Mimar Ajanı | Tahmin: 23 talep = 118 puan; 2 talep 13 puan üstü → parçalama önerildi | — | 2.000 | 3.900 | | 08 | doğrulama | Proje Yöneticisi Ajanı | Sprint 42 önerisi: 11 talep, 44 puan (hız 46), bağımlılık çakışması yok | — | 900 | 1.100 | | 09 | insan onayı | Planlama Süpervizörü | Ürün sahibi öneriyi inceledi, 1 talebi değiştirdi, sprint 42'yi onayladı | — | 4.000 | — | | 10 | yazma | Proje Yöneticisi Ajanı | Sprint 42 oluşturuldu, 11 talep puanlarıyla taşındı, kalanı önceliklendirildi | Jira | 560 | — | Simülasyon metrikleri (tahmini) 340çalıştırma / ay %92,8başarı 14,6 sp50 $0,0189tahmini maliyet / çalıştırma Onay kuralı Sprint kapsamı → ürün sahibi onayı; ajan sprint'e talep atayamaz, yalnızca önerir Adım profili giriş · 1plan · 1bilgi · 1akıl yürütme · 2araç · 1doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - Jiraticketing - Confluencestorage - GitHubapi - Slack / Teamschannel Çalıştırmak için gerekenler - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Doküman deposuna okuma erişimi ve belge tipleri - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Çevik ölçeklendirme yapan büyük yazılım kuruluşlarının yapay zekâ destekli backlog iyileştirme deseni Bu senaryonun yer aldığı paketler Ajan kiralama (Agent Outsourcing) → Sektör: Yazılım Geliştirme → İlgili ajanlar Yazılım GeliştirmeHITL ### Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR Jira'daki bir epic'ten başlar; eski modül okunur, davranışı testlerle sabitlenir, yeniden yazılır ve kanıtlı bir PR açılır. 1.900 çalıştırma / ay %90,4 başarı 5 ajan Yazılım GeliştirmeHITL ### Özellik teslimat hattı Bir Jira epic'i ajanlarla kriter, tasarım, kod ve teste dönüşür; insan yalnızca PR'ı inceler ve onaylar. 1.200 çalıştırma / ay %91,2 başarı 5 ajan Yazılım GeliştirmeHITL ### CI/CD sürüm hattı Derleme, test, tarama ve staging otomatik; üretime yalnızca sürüm yöneticisi onayıyla ArgoCD ile çıkılır. 950 çalıştırma / ay %94,7 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Fatura itirazı çözümü URL: https://www.aginies.com/tr/ajanlar/bill-dispute-resolution Language: tr Summary: Fatura kalemleri ve kullanım kayıtları saniyede açıklanır; iyi niyet kredisi politikaya göre uygulanır. Uygulama, WhatsApp veya çağrıda "faturam neden yüksek?" sorusu → Müşteri fatura kalemlerini anladı, kredi uygulandı, vaka kapatıldı Telekomappvoicewhatsapp # Fatura itirazı çözümü Fatura kalemleri ve kullanım kayıtları saniyede açıklanır; iyi niyet kredisi politikaya göre uygulanır. Kısaca Fatura itirazı çözümü, Telekom sektöründe app, voice, whatsapp kanalında çalışan bir ajan senaryosudur. Tetikleyici: Uygulama, WhatsApp veya çağrıda "faturam neden yüksek?" sorusu. Fatura Süpervizörü 3 uzman ajanı (Niyet Ajanı, Fatura Analiz Ajanı, Düzeltme Ajanı) 3 sistem üzerinde yönetir (Billing (Amdocs), Usage Mediation / CDR Store, SMS Gateway); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Müşteri fatura kalemlerini anladı, kredi uygulandı, vaka kapatıldı. Onay kuralı: İyi niyet kredisi > 75 TL → ekip lideri onayı. Telekom ### Fatura itirazı çözümü [Fatura itirazı çözümü] çalıştırma #41.288.477süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: İyi niyet kredisi > 75 TL → ekip lideri onayı Bu çalıştırmadaki ajanlar Fatura Süpervizörüsupervisorplanlar, yönlendirir, onaylar Niyet Ajanıaginies-sunkar-1 · on-premİtirazı ve dönemi anlar, hattı doğrular Fatura Analiz Ajanıaginies-tulpar-4 · on-premFatura kalemlerini kullanım kayıtları ve tarifeyle açıklar Düzeltme Ajanıaginies-talay-2 · on-premİyi niyet politikasını uygular, krediyi yansıtır Denetim izi00:00.00 - ◦Uygulama: "Bu ay faturam neden 640 TL? Normalde 420 TL" - ∿Niyet: fatura itirazı · dönem: Ağustos · hat 0532 *** 4471 doğrulandı - ⚙Fatura #INV-2608-99120: 640 TL = paket 420 + yurt dışı dolaşım 185 + SMS 35 - ⚙CDR: 14–16.08 arasında Bulgaristan'da 380 MB dolaşım verisi - ⌘Plan: kalem açıklaması → iyi niyet politikası → onay → kredi → bildirim - ∿Dolaşım ücreti tarifeyle uyumlu; müşteri 4 yıldır hatta, ilk itirazı - ✓İyi niyet politikası: ilk itiraz + sadık müşteri → %50 kredi (92,50 TL), eşik üstü - ⏸Kredi > 75 TL → ekip lideri 92,50 TL krediyi onayladı - ⇢92,50 TL kredi bir sonraki faturaya yansıtıldı, vaka #BD-40211 kapatıldı - ➤SMS: kalem açıklaması, kredi ve dolaşım paketi önerisi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Uygulama, WhatsApp veya çağrıda "faturam neden yüksek?" sorusu Sonuç Müşteri fatura kalemlerini anladı, kredi uygulandı, vaka kapatıldı ## Ajanlar SÜPERVİZÖR ### Fatura Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Niyet Ajanı İtirazı ve dönemi anlar, hattı doğrular detect_intentextract_bill_periodverify_customer aginies-tulpar-4 · on-prem ### Fatura Analiz Ajanı Fatura kalemlerini kullanım kayıtları ve tarifeyle açıklar get_invoiceget_usage_detailget_tariffexplain_charges aginies-talay-2 · on-prem ### Düzeltme Ajanı İyi niyet politikasını uygular, krediyi yansıtır check_goodwill_policyapply_creditclose_casesend_sms ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Fatura Süpervizörü | Uygulama: "Bu ay faturam neden 640 TL? Normalde 420 TL" | — | 180 | — | | 02 | akıl yürütme | Niyet Ajanı | Niyet: fatura itirazı · dönem: Ağustos · hat 0532 *** 4471 doğrulandı | — | 400 | 250 | | 03 | araç | Fatura Analiz Ajanı | Fatura #INV-2608-99120: 640 TL = paket 420 + yurt dışı dolaşım 185 + SMS 35 | Billing (Amdocs) | 420 | — | | 04 | araç | Fatura Analiz Ajanı | CDR: 14–16.08 arasında Bulgaristan'da 380 MB dolaşım verisi | Usage Mediation / CDR Store | 480 | — | | 05 | plan | Fatura Süpervizörü | Plan: kalem açıklaması → iyi niyet politikası → onay → kredi → bildirim | — | 240 | 170 | | 06 | akıl yürütme | Fatura Analiz Ajanı | Dolaşım ücreti tarifeyle uyumlu; müşteri 4 yıldır hatta, ilk itirazı | — | 760 | 820 | | 07 | doğrulama | Düzeltme Ajanı | İyi niyet politikası: ilk itiraz + sadık müşteri → %50 kredi (92,50 TL), eşik üstü | — | 440 | 320 | | 08 | insan onayı | Fatura Süpervizörü | Kredi > 75 TL → ekip lideri 92,50 TL krediyi onayladı | — | 2.300 | — | | 09 | yazma | Düzeltme Ajanı | 92,50 TL kredi bir sonraki faturaya yansıtıldı, vaka #BD-40211 kapatıldı | Billing (Amdocs) | 480 | — | | 10 | bildirim | Düzeltme Ajanı | SMS: kalem açıklaması, kredi ve dolaşım paketi önerisi | SMS Gateway | 260 | — | Simülasyon metrikleri (tahmini) 88.000çalıştırma / ay %96,3başarı 6 sp50 $0,0009tahmini maliyet / çalıştırma Onay kuralı İyi niyet kredisi > 75 TL → ekip lideri onayı Adım profili giriş · 1plan · 1akıl yürütme · 2araç · 2doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - Billing (Amdocs)core - Usage Mediation / CDR Storedb - SMS Gatewaychannel Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük operatörlerin "faturamı açıkla" asistanı deseninin kredi uygulayan sürümü Bu senaryonun yer aldığı paketler Çağrı merkezi kalite yönetimi →Abone hizmetleri ve şebeke desteği → Sektör: Telekom → İlgili ajanlar TelekomHITL ### Abone hizmet asistanı: fatura, paket, arıza ve taahhüt Abone arar ya da yazar; fatura itirazı, paket değişikliği ve hat arızası aynı oturumda BSS ve şebeke sistemlerinde çözülür. 410.000 çalıştırma / ay %94,3 başarı 5 ajan TelekomHITL ### SIM değişimi dolandırıcılık kontrolü Her SIM değişimi provizyondan önce risk skorlanır; şüpheli talepler bekletilir ve bankalara sinyal iletilir. 41.000 çalıştırma / ay %97,7 başarı 4 ajan TelekomHITL ### Şebeke kesintisi iletişimi Kesinti kaydı açılınca etkilenen aboneler bulunur, mesaj yazılıp onaylanır; müşteri aramadan bilgilendirilir. 1.400 çalıştırma / ay %98,5 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # BNPL anlık taksit kararı URL: https://www.aginies.com/tr/ajanlar/bnpl-credit-decision Language: tr Summary: Ödemede saniyeler içinde taksit kararı; büro ve açık bankacılık verisi birleşir, limit üstü uzmana gider. Ödeme sayfasında "3 taksitle öde" seçimi (API) → Taksit planı ana deftere yazıldı, satıcıya ödeme taahhüdü verildi, müşteri bilgilendirildi Finansal Teknolojilerapiappsms # BNPL anlık taksit kararı Ödemede saniyeler içinde taksit kararı; büro ve açık bankacılık verisi birleşir, limit üstü uzmana gider. Kısaca BNPL anlık taksit kararı, Finansal Teknolojiler sektöründe api, app, sms kanalında çalışan bir ajan senaryosudur. Tetikleyici: Ödeme sayfasında "3 taksitle öde" seçimi (API). Kredi Karar Süpervizörü 4 uzman ajanı (Kabul Ajanı, Büro Ajanı, Karar Ajanı, Kayıt Ajanı) 5 sistem üzerinde yönetir (Merchant Checkout API, Credit Bureau (KKB / Findeks), Open Banking API, Core Ledger, SMS Gateway); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Taksit planı ana deftere yazıldı, satıcıya ödeme taahhüdü verildi, müşteri bilgilendirildi. Onay kuralı: Sepet > 15.000 TL veya büro skoru < 1.200 → kredi uzmanı onayı. Finansal Teknolojiler ### BNPL anlık taksit kararı [BNPL anlık taksit kararı] çalıştırma #41.289.947süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Sepet > 15.000 TL veya büro skoru < 1.200 → kredi uzmanı onayı Bu çalıştırmadaki ajanlar Kredi Karar Süpervizörüsupervisorplanlar, yönlendirir, onaylar Kabul Ajanıaginies-sunkar-1 · on-premSepeti, müşteriyi ve dolandırıcılık sinyallerini kontrol eder Büro Ajanıaginies-talay-2 · on-premKKB/Findeks skorunu ve açık bankacılık nakit akışını çeker Karar Ajanıaginies-tulpar-4 · on-premÖdeme gücü ve politikaya göre karar ve limit verir Kayıt Ajanıaginies-talay-2 · on-premPlanı ana deftere yazar, satıcıya ve müşteriye bildirir Denetim izi00:00.00 - ◦Sepet 18.750 TL, 3 taksit talebi; müşteri E.Y., cihaz ve e-posta kayıtlı - ⚙Sepet ve satıcı verisi çekildi: MCC 5732, satıcı ters ibraz oranı %0,4 - ⌘Plan: büro skoru → açık bankacılık → ödeme gücü → politika → kayıt - ⚙Findeks skoru 1.480, açık kredi 2, gecikme yok, 30 günde 1 sorgu - ⚙Açık bankacılık: 3 aylık ortalama gelir 62.000 TL, aylık serbest nakit 14.200 TL - ✓Politika v3.0: taksit 6.250 TL ≤ serbest nakdin %50'si → uygun, 15.000 TL limit üstü - ∿Karar açıklaması üretildi: 3 × 6.250 TL, %0 faiz, ilk taksit bugün - ⏸Tutar > 15.000 TL → kredi uzmanı 90 saniyede onayladı - ⇢Taksit planı #BN-204118 ana deftere yazıldı, satıcıya T+1 ödeme taahhüdü verildi - ➤Müşteriye SMS: plan özeti ve taksit tarihleri (08.09, 08.10, 08.11) - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Ödeme sayfasında "3 taksitle öde" seçimi (API) Sonuç Taksit planı ana deftere yazıldı, satıcıya ödeme taahhüdü verildi, müşteri bilgilendirildi ## Ajanlar SÜPERVİZÖR ### Kredi Karar Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Kabul Ajanı Sepeti, müşteriyi ve dolandırıcılık sinyallerini kontrol eder get_checkoutget_customercheck_fraud_signals aginies-talay-2 · on-prem ### Büro Ajanı KKB/Findeks skorunu ve açık bankacılık nakit akışını çeker get_bureau_scorefetch_open_bankingcompute_affordability aginies-tulpar-4 · on-prem ### Karar Ajanı Ödeme gücü ve politikaya göre karar ve limit verir apply_credit_policyset_instalment_planexplain_decision aginies-talay-2 · on-prem ### Kayıt Ajanı Planı ana deftere yazar, satıcıya ve müşteriye bildirir post_journalconfirm_merchant_payoutsend_sms ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Kredi Karar Süpervizörü | Sepet 18.750 TL, 3 taksit talebi; müşteri E.Y., cihaz ve e-posta kayıtlı | — | 160 | — | | 02 | araç | Kabul Ajanı | Sepet ve satıcı verisi çekildi: MCC 5732, satıcı ters ibraz oranı %0,4 | Merchant Checkout API | 320 | — | | 03 | plan | Kredi Karar Süpervizörü | Plan: büro skoru → açık bankacılık → ödeme gücü → politika → kayıt | — | 260 | 190 | | 04 | araç | Büro Ajanı | Findeks skoru 1.480, açık kredi 2, gecikme yok, 30 günde 1 sorgu | Credit Bureau (KKB / Findeks) | 440 | — | | 05 | araç | Büro Ajanı | Açık bankacılık: 3 aylık ortalama gelir 62.000 TL, aylık serbest nakit 14.200 TL | Open Banking API | 680 | — | | 06 | doğrulama | Karar Ajanı | Politika v3.0: taksit 6.250 TL ≤ serbest nakdin %50'si → uygun, 15.000 TL limit üstü | — | 580 | 640 | | 07 | akıl yürütme | Karar Ajanı | Karar açıklaması üretildi: 3 × 6.250 TL, %0 faiz, ilk taksit bugün | — | 420 | 380 | | 08 | insan onayı | Kredi Karar Süpervizörü | Tutar > 15.000 TL → kredi uzmanı 90 saniyede onayladı | — | 2.200 | — | | 09 | yazma | Kayıt Ajanı | Taksit planı #BN-204118 ana deftere yazıldı, satıcıya T+1 ödeme taahhüdü verildi | Core Ledger | 480 | — | | 10 | bildirim | Kayıt Ajanı | Müşteriye SMS: plan özeti ve taksit tarihleri (08.09, 08.10, 08.11) | SMS Gateway | 220 | — | Simülasyon metrikleri (tahmini) 54.000çalıştırma / ay %97,3başarı 5,6 sp50 $0,0015tahmini maliyet / çalıştırma Onay kuralı Sepet > 15.000 TL veya büro skoru < 1.200 → kredi uzmanı onayı Adım profili giriş · 1plan · 1akıl yürütme · 1araç · 3doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - Merchant Checkout APIapi - Credit Bureau (KKB / Findeks)api - Open Banking APIapi - Core Ledgercore - SMS Gatewaychannel Çalıştırmak için gerekenler - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Küresel BNPL sağlayıcılarının ödeme anında açık bankacılık destekli kredi kararı deseni Bu senaryonun yer aldığı paketler Finansal analiz ve operasyonel zekâ → Sektör: Finansal Teknolojiler → İlgili ajanlar Finansal TeknolojilerHITL ### Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit İşyeri WhatsApp'tan yazar; POS arızası, hakediş sorgusu, iade ve limit artışı tek diyalogda çözülür. 64.000 çalıştırma / ay %93,8 başarı 5 ajan Finansal TeknolojilerHITL ### Üye işyeri KYB onboarding Sicil, gerçek faydalanıcı ve yaptırım kontrolü tek akışta; düşük riskli işyeri aynı gün ödeme almaya başlar. 1.400 çalıştırma / ay %93,4 başarı 5 ajan Finansal TeknolojilerHITL ### AML işlem izleme alarm triyajı Alarm zenginleştirilir, anlatı taslağı hazır gelir; kararı analist verir, bildirim dakikalarda dosyalanır. 8.500 çalıştırma / ay %95,8 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Çağrı merkezi kalite izleme URL: https://www.aginies.com/tr/ajanlar/call-center-quality-monitoring Language: tr Summary: Örneklem değil, tüm çağrılar rubriğe göre puanlanır; uyum ihlali ve koçluk fırsatı sabah liderin önünde. Gece toplu işi: Genesys'ten günün çağrı kayıtları → Her çağrının puan kartı dolduruldu, ihlaller lidere düştü, haftalık koçluk özeti Workday'de Kurumsal Fonksiyonlarvoicebatch # Çağrı merkezi kalite izleme Örneklem değil, tüm çağrılar rubriğe göre puanlanır; uyum ihlali ve koçluk fırsatı sabah liderin önünde. Kısaca Çağrı merkezi kalite izleme, Kurumsal Fonksiyonlar sektöründe voice, batch kanalında çalışan bir ajan senaryosudur. Tetikleyici: Gece toplu işi: Genesys'ten günün çağrı kayıtları. Kalite Süpervizörü 3 uzman ajanı (Transkript Ajanı, Puanlama Ajanı, Koçluk Ajanı) 4 sistem üzerinde yönetir (Genesys SIP, QM Scorecard / Case Management, QA Rubric Vector Store, Workday); 9 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Her çağrının puan kartı dolduruldu, ihlaller lidere düştü, haftalık koçluk özeti Workday'de. Onay kuralı: Puan < 60 veya uyum ihlali → takım lideri incelemesi. Kurumsal Fonksiyonlar ### Çağrı merkezi kalite izleme [Çağrı merkezi kalite izleme] çalıştırma #41.287.411süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Puan < 60 veya uyum ihlali → takım lideri incelemesi Bu çalıştırmadaki ajanlar Kalite Süpervizörüsupervisorplanlar, yönlendirir, onaylar Transkript Ajanıaginies-voice-tr-v3 · on-premKayıtları yazıya döker, konuşmacıları ayırır, KVKK maskeler Puanlama Ajanıaginies-tulpar-4 · on-premRubrik kriterlerine göre puanlar, uyum ifadelerini arar Koçluk Ajanıaginies-sunkar-1 · on-premKoçluk bayrağı açar, özet üretir Denetim izi00:00.00 - ◦Gece toplu işi: Genesys'ten 4.120 kayıt kuyruğa alındı (%100 kapsama) - ⚙C-9921834 çağrısı (6:42 dk) temsilci ve kuyruk bilgisiyle çekildi - ∿Transkript 2 konuşmacı, %94 kelime doğruluğu, kart ve TC maskelendi - ⌕Rubrik v7 getirildi: 12 kriter, 3 zorunlu uyum ifadesi - ∿Puan 78/100 · eksik: bilgilendirme ifadesi · empati 4/5 · çözüm: evet - ✓Kalibrasyon: %2 örneklem QA tarafından çift puanlandı, sapma 3,1 puan → uygun - ⇢Puan kartı güncellendi; A-2231 temsilcisi için "uyum ifadesi" koçluk bayrağı açıldı - ⏸Uyum ihlali → takım lideri dosyaya işlenmeden önce inceledi - ➤Haftalık koçluk özeti Workday'e gönderildi: 3 temsilci işaretli, 41 takdir - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Gece toplu işi: Genesys'ten günün çağrı kayıtları Sonuç Her çağrının puan kartı dolduruldu, ihlaller lidere düştü, haftalık koçluk özeti Workday'de ## Ajanlar SÜPERVİZÖR ### Kalite Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-voice-tr-v3 · on-prem ### Transkript Ajanı Kayıtları yazıya döker, konuşmacıları ayırır, KVKK maskeler get_recordingtranscribe_calldiarize_speakersredact_pii aginies-tulpar-4 · on-prem ### Puanlama Ajanı Rubrik kriterlerine göre puanlar, uyum ifadelerini arar search_rubricscore_rubricdetect_compliance_phrasesdetect_sentiment aginies-sunkar-1 · on-prem ### Koçluk Ajanı Koçluk bayrağı açar, özet üretir summarise_callcreate_coaching_flagupdate_scorecardpush_hr_summary ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Kalite Süpervizörü | Gece toplu işi: Genesys'ten 4.120 kayıt kuyruğa alındı (%100 kapsama) | — | 300 | — | | 02 | araç | Transkript Ajanı | C-9921834 çağrısı (6:42 dk) temsilci ve kuyruk bilgisiyle çekildi | Genesys SIP | 380 | — | | 03 | akıl yürütme | Transkript Ajanı | Transkript 2 konuşmacı, %94 kelime doğruluğu, kart ve TC maskelendi | — | 1.400 | 520 | | 04 | bilgi | Puanlama Ajanı | Rubrik v7 getirildi: 12 kriter, 3 zorunlu uyum ifadesi | QA Rubric Vector Store | 320 | 300 | | 05 | akıl yürütme | Puanlama Ajanı | Puan 78/100 · eksik: bilgilendirme ifadesi · empati 4/5 · çözüm: evet | — | 960 | 1.300 | | 06 | doğrulama | Puanlama Ajanı | Kalibrasyon: %2 örneklem QA tarafından çift puanlandı, sapma 3,1 puan → uygun | — | 340 | 220 | | 07 | yazma | Koçluk Ajanı | Puan kartı güncellendi; A-2231 temsilcisi için "uyum ifadesi" koçluk bayrağı açıldı | QM Scorecard / Case Management | 460 | — | | 08 | insan onayı | Kalite Süpervizörü | Uyum ihlali → takım lideri dosyaya işlenmeden önce inceledi | — | 2.200 | — | | 09 | bildirim | Koçluk Ajanı | Haftalık koçluk özeti Workday'e gönderildi: 3 temsilci işaretli, 41 takdir | Workday | 260 | — | Simülasyon metrikleri (tahmini) 90.000çalıştırma / ay %98,8başarı 6,6 sp50 $0,0016tahmini maliyet / çalıştırma Onay kuralı Puan < 60 veya uyum ihlali → takım lideri incelemesi Adım profili giriş · 1bilgi · 1akıl yürütme · 2araç · 1doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - Genesys SIPvoice - QM Scorecard / Case Managementticketing - QA Rubric Vector Storedb - Workdayhr Çalıştırmak için gerekenler - Santral ile SIP entegrasyonu ve kayıt erişimi - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - İK sistemi API erişimi ve kişisel veri sınırları - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük BPO'lar ve banka çağrı merkezlerindeki %100 kalite kapsama programlarını yansıtır Bu senaryonun yer aldığı paketler Çağrı merkezi kalite yönetimi →KVKK kişisel veri koruma →Ses analizi ve kalite → İlgili ajanlar Kurumsal FonksiyonlarHITL ### Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada Talep e-postayla gelir; teklifler karşılaştırılır, sözleşme maddeleri kontrol edilir, sipariş açılır, fatura üç yönlü eşleşir ve ödeme planlanır. 14.000 çalıştırma / ay %92,7 başarı 5 ajan Kurumsal Fonksiyonlar ### E-posta sınıflandırma ve ek kontrolü Ortak kutuya gelen her e-posta sınıflandırılır, ekler listeyle karşılaştırılır, eksikler anında sorulur. 86.000 çalıştırma / ay %97,6 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Sözleşme madde çıkarımı Sözleşme maddeleri çıkarılır, hukuk playbook'u ile karşılaştırılır; sapmalar redline önerisiyle avukata gider. 1.400 çalıştırma / ay %95,1 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Kredi kartı borcu ödeme URL: https://www.aginies.com/tr/ajanlar/card-debt-payment Language: tr Summary: Müşteri yazar ya da konuşur; borç sorgulanır, hesap seçilir, OTP ile doğrulanır, ödeme çekirdek sistemde yapılır. Mobil uygulamada ya da sesli kanalda "kart borcumu öde" talebi → Ödeme çekirdek sistemde gerçekleşti, müşteri bilgilendirildi, karar izi kaydedildi Bankacılıkappvoice # Kredi kartı borcu ödeme Müşteri yazar ya da konuşur; borç sorgulanır, hesap seçilir, OTP ile doğrulanır, ödeme çekirdek sistemde yapılır. Kısaca Kredi kartı borcu ödeme, Bankacılık sektöründe app, voice kanalında çalışan bir ajan senaryosudur. Tetikleyici: Mobil uygulamada ya da sesli kanalda "kart borcumu öde" talebi. Kart Süpervizörü 3 uzman ajanı (Niyet Ajanı, Ödeme Ajanı, Ses Ajanı) 4 sistem üzerinde yönetir (Card Management, Core Banking, Knowledge Base · kart_sss_v3, OTP / SMS Gateway); 9 adımın 1 tanesi sistemde işlem, 1 tanesi doğrulama veya onaydır. Sonuç: Ödeme çekirdek sistemde gerçekleşti, müşteri bilgilendirildi, karar izi kaydedildi. Onay kuralı: Tutar > 100.000 TL → şube yöneticisi onayı. Bankacılık ### Kredi kartı borcu ödeme [Kredi kartı borcu ödeme] çalıştırma #41.287.329süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Tutar > 100.000 TL → şube yöneticisi onayı Bu çalıştırmadaki ajanlar Kart Süpervizörüsupervisorplanlar, yönlendirir, onaylar Niyet Ajanıaginies-sunkar-1 · on-premTalebi anlar; kartı, tutarı ve ödeme türünü çıkarır Ödeme Ajanıaginies-tulpar-4 · on-premHesap bakiyesini kontrol eder, OTP ister, ödemeyi çekirdek sistemde yapar Ses Ajanıaginies-voice-tr-v3 · on-premSesli kanalda konuşmayı yazıya döker ve yanıtı seslendirir Denetim izi00:00.00 - ◦Talep alındı, oturum kimliği doğrulandı · kanal: mobil - ∿Niyet: borç ödeme · Kart: ****2210 · Tür: dönem borcunun tamamı - ⚙Kart borcu çekildi: 24.860 TL · son ödeme 14 Eylül - ⌘Plan: hesap seçimi → OTP → ödeme → bildirim - ⚙Hesaplar listelendi: vadesiz ****7741 bakiye yeterli - ✓Politika: tutar ≤ 100.000 TL → onay gerekmez; OTP zorunlu - ➤OTP gönderildi ve doğrulandı (2 deneme hakkı kaldı) - ⇢24.860 TL ****7741 hesabından karta ödendi · işlem #TX-88120 - ➤Müşteriye sesli ve yazılı teyit verildi, karar izi kaydedildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Mobil uygulamada ya da sesli kanalda "kart borcumu öde" talebi Sonuç Ödeme çekirdek sistemde gerçekleşti, müşteri bilgilendirildi, karar izi kaydedildi ## Ajanlar SÜPERVİZÖR ### Kart Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Niyet Ajanı Talebi anlar; kartı, tutarı ve ödeme türünü çıkarır detect_intentextract_entitiesget_card_debt aginies-tulpar-4 · on-prem ### Ödeme Ajanı Hesap bakiyesini kontrol eder, OTP ister, ödemeyi çekirdek sistemde yapar get_accountssend_otppay_from_account aginies-voice-tr-v3 · on-prem ### Ses Ajanı Sesli kanalda konuşmayı yazıya döker ve yanıtı seslendirir transcribe_streamspeak ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Kart Süpervizörü | Talep alındı, oturum kimliği doğrulandı · kanal: mobil | — | 160 | — | | 02 | akıl yürütme | Niyet Ajanı | Niyet: borç ödeme · Kart: ****2210 · Tür: dönem borcunun tamamı | — | 520 | 380 | | 03 | araç | Niyet Ajanı | Kart borcu çekildi: 24.860 TL · son ödeme 14 Eylül | Card Management | 340 | — | | 04 | plan | Kart Süpervizörü | Plan: hesap seçimi → OTP → ödeme → bildirim | — | 280 | 190 | | 05 | araç | Ödeme Ajanı | Hesaplar listelendi: vadesiz ****7741 bakiye yeterli | Core Banking | 410 | — | | 06 | doğrulama | Ödeme Ajanı | Politika: tutar ≤ 100.000 TL → onay gerekmez; OTP zorunlu | — | 300 | 210 | | 07 | bildirim | Ödeme Ajanı | OTP gönderildi ve doğrulandı (2 deneme hakkı kaldı) | OTP / SMS Gateway | 2.200 | — | | 08 | yazma | Ödeme Ajanı | 24.860 TL ****7741 hesabından karta ödendi · işlem #TX-88120 | Core Banking | 540 | — | | 09 | bildirim | Ses Ajanı | Müşteriye sesli ve yazılı teyit verildi, karar izi kaydedildi | OTP / SMS Gateway | 260 | 120 | Simülasyon metrikleri (tahmini) 52.000çalıştırma / ay %97,8başarı 5 sp50 $0,001tahmini maliyet / çalıştırma Onay kuralı Tutar > 100.000 TL → şube yöneticisi onayı Adım profili giriş · 1plan · 1akıl yürütme · 1araç · 2doğrulama · 1yazma · 1bildirim · 2 SİSTEMLER - Card Managementcore - Core Bankingcore - Knowledge Base · kart_sss_v3db - OTP / SMS Gatewaychannel Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Diyalog bankacılığı: ödeme işlemini konuşma içinde tamamlayan uygulama içi asistan deseni Bu senaryonun yer aldığı paketler Kredi ve müşteri edinimi operasyonları → Sektör: Bankacılık → İlgili ajanlar BankacılıkHITL ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı Müşteri konuşur ya da yazar; fatura, limit, itiraz ve transfer aynı oturumda çekirdek sistemde tamamlanır. 186.000 çalıştırma / ay %94,7 başarı 5 ajan BankacılıkHITL ### Kredi kartı limit artışı Müşteri mobilde yazar ya da çağrı merkezinde konuşur; karar saniyeler içinde verilir, eşik üstü yönetici onayına gider. 18.400 çalıştırma / ay %96,4 başarı 4 ajan BankacılıkHITL ### Dijital müşteri edinimi (KYC) Kimlik, canlılık ve yaptırım kontrolü tek akışta; temiz başvurular şubeye gitmeden dakikalarda hesap açar. 42.000 çalıştırma / ay %93,8 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Katalog zenginleştirme (toplu) URL: https://www.aginies.com/tr/ajanlar/catalogue-enrichment Language: tr Summary: Tedarikçi PDF ve görsellerinden öznitelik çıkarılır, TR/EN içerik yazılır; yüzlerce SKU dakikalarda hazır. Tedarikçi katalog dosyalarının S3 klasörüne düşmesi → SKU'lar PIM'de yayına hazır, istisnalar kuyrukta, katalog ekibi özet raporu aldı Perakende & E-ticaretbatchemail # Katalog zenginleştirme (toplu) Tedarikçi PDF ve görsellerinden öznitelik çıkarılır, TR/EN içerik yazılır; yüzlerce SKU dakikalarda hazır. Kısaca Katalog zenginleştirme (toplu), Perakende & E-ticaret sektöründe batch, email kanalında çalışan bir ajan senaryosudur. Tetikleyici: Tedarikçi katalog dosyalarının S3 klasörüne düşmesi. Katalog Süpervizörü 3 uzman ajanı (Belge Ajanı, Normalizasyon Ajanı, İçerik Ajanı) 4 sistem üzerinde yönetir (Document Store (S3), PIM / Catalogue, SAP ERP, Email (Exchange)); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: SKU'lar PIM'de yayına hazır, istisnalar kuyrukta, katalog ekibi özet raporu aldı. Onay kuralı: Yeni kategori veya tedarikçi fiyatı sapması > %15 → kategori yöneticisi onayı. Perakende & E-ticaret ### Katalog zenginleştirme (toplu) [Katalog zenginleştirme (toplu)] çalıştırma #41.287.928süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Yeni kategori veya tedarikçi fiyatı sapması > %15 → kategori yöneticisi onayı Bu çalıştırmadaki ajanlar Katalog Süpervizörüsupervisorplanlar, yönlendirir, onaylar Belge Ajanıaginies-konrul-1 · on-premPDF ve görsellerden ürün özniteliklerini çıkarır Normalizasyon Ajanıaginies-talay-2 · on-premTaksonomiye eşler, birimleri düzeltir, mükerrerleri ayıklar İçerik Ajanıaginies-tulpar-4 · on-premMarka tonunda TR/EN açıklama ve SEO başlığı yazar Denetim izi00:00.00 - ◦Tedarikçi teslimatı: 340 ürünlük PDF katalog + 1.120 görsel S3'e düştü - ⌕Dosyalar okundu; SKU başına 14 öznitelik çıkarıldı (malzeme, ölçü, renk), güven %96 - ⌘Plan: taksonomi eşleme → birim normalizasyonu → tekilleştirme → içerik → QC - ⚙SAP'tan mevcut 2.150 SKU ve tedarikçi fiyat listesi çekildi - ∿340 SKU → 312 yeni, 28 mevcut; ölçüler cm'ye çevrildi, 3 renk kodu eşlenemedi - ∿312 ürün için TR/EN açıklama ve SEO başlığı üretildi, marka ton kılavuzu uygulandı - ✓QC: zorunlu alanlar %99,1 dolu; 9 SKU eksik ölçü → insan kuyruğu - ⏸Yeni kategori veya fiyat sapması > %15 → kategori yöneticisi 12 SKU'yu onayladı - ⇢303 SKU PIM'de "yayına hazır", 9 SKU "eksik veri" olarak kaydedildi - ➤Katalog ekibine özet rapor ve istisna listesi e-posta ile gönderildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Tedarikçi katalog dosyalarının S3 klasörüne düşmesi Sonuç SKU'lar PIM'de yayına hazır, istisnalar kuyrukta, katalog ekibi özet raporu aldı ## Ajanlar SÜPERVİZÖR ### Katalog Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### Belge Ajanı PDF ve görsellerden ürün özniteliklerini çıkarır parse_supplier_pdfextract_attributesread_product_image aginies-talay-2 · on-prem ### Normalizasyon Ajanı Taksonomiye eşler, birimleri düzeltir, mükerrerleri ayıklar get_existing_skusmap_to_taxonomynormalise_unitsdedupe_skurun_qc aginies-tulpar-4 · on-prem ### İçerik Ajanı Marka tonunda TR/EN açıklama ve SEO başlığı yazar write_descriptiongenerate_seo_titletranslatesave_to_pimsend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Katalog Süpervizörü | Tedarikçi teslimatı: 340 ürünlük PDF katalog + 1.120 görsel S3'e düştü | — | 300 | — | | 02 | bilgi | Belge Ajanı | Dosyalar okundu; SKU başına 14 öznitelik çıkarıldı (malzeme, ölçü, renk), güven %96 | Document Store (S3) | 1.800 | 3.400 | | 03 | plan | Katalog Süpervizörü | Plan: taksonomi eşleme → birim normalizasyonu → tekilleştirme → içerik → QC | — | 260 | 190 | | 04 | araç | Normalizasyon Ajanı | SAP'tan mevcut 2.150 SKU ve tedarikçi fiyat listesi çekildi | SAP ERP | 520 | — | | 05 | akıl yürütme | Normalizasyon Ajanı | 340 SKU → 312 yeni, 28 mevcut; ölçüler cm'ye çevrildi, 3 renk kodu eşlenemedi | — | 740 | 680 | | 06 | akıl yürütme | İçerik Ajanı | 312 ürün için TR/EN açıklama ve SEO başlığı üretildi, marka ton kılavuzu uygulandı | — | 2.400 | 5.200 | | 07 | doğrulama | Normalizasyon Ajanı | QC: zorunlu alanlar %99,1 dolu; 9 SKU eksik ölçü → insan kuyruğu | — | 480 | 350 | | 08 | insan onayı | Katalog Süpervizörü | Yeni kategori veya fiyat sapması > %15 → kategori yöneticisi 12 SKU'yu onayladı | — | 3.200 | — | | 09 | yazma | İçerik Ajanı | 303 SKU PIM'de "yayına hazır", 9 SKU "eksik veri" olarak kaydedildi | PIM / Catalogue | 620 | — | | 10 | bildirim | İçerik Ajanı | Katalog ekibine özet rapor ve istisna listesi e-posta ile gönderildi | Email (Exchange) | 240 | — | Simülasyon metrikleri (tahmini) 900çalıştırma / ay %95,8başarı 10,5 sp50 $0,0198tahmini maliyet / çalıştırma Onay kuralı Yeni kategori veya tedarikçi fiyatı sapması > %15 → kategori yöneticisi onayı Adım profili giriş · 1plan · 1bilgi · 1akıl yürütme · 2araç · 1doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - Document Store (S3)storage - PIM / Cataloguedb - SAP ERPerp - Email (Exchange)email Çalıştırmak için gerekenler - Doküman deposuna okuma erişimi ve belge tipleri - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - ERP/MES için kayıt ve sorgu API erişimi - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Pazar yerlerinin tedarikçi verisini ölçekte otomatik zenginleştiren PIM deseni Bu senaryonun yer aldığı paketler Sektör: Perakende & E-ticaret → İlgili ajanlar Perakende & E-ticaretHITL ### Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin Müşteri bir fotoğraf ve bir cümle gönderir; ajan stokta olan, bedenine uyan ve bütçeye sığan üç kombin kurar, seçileni sepete ekler. 142.000 çalıştırma / ay %96,1 başarı 5 ajan Perakende & E-ticaret ### Siparişim nerede? (WISMO) Müşteri sorar, sipariş ve kargo olayları anında birleşir; gecikmede yeni tarih verilir, kargo araştırılır. 110.000 çalıştırma / ay %98,1 başarı 4 ajan Perakende & E-ticaretHITL ### İade ve para iadesi Politika ve müşteri geçmişi saniyede kontrol edilir, etiket üretilir; iade tutarı kargo taramasıyla ödenir. 48.000 çalıştırma / ay %96,7 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Ters ibraz (chargeback) itirazı URL: https://www.aginies.com/tr/ajanlar/chargeback-dispute Language: tr Summary: İtiraz alınır, kanıt paketi toplanır ve kart şemasına dosyalanır; eşik üstü geçici alacak onaya gider. Uygulamada veya şubede "bu işlemi tanımıyorum / ürün gelmedi" itirazı → Ters ibraz şemaya dosyalandı, geçici alacak yansıtıldı, müşteri takvimi biliyor Bankacılıkappemailbranch # Ters ibraz (chargeback) itirazı İtiraz alınır, kanıt paketi toplanır ve kart şemasına dosyalanır; eşik üstü geçici alacak onaya gider. Kısaca Ters ibraz (chargeback) itirazı, Bankacılık sektöründe app, email, branch kanalında çalışan bir ajan senaryosudur. Tetikleyici: Uygulamada veya şubede "bu işlemi tanımıyorum / ürün gelmedi" itirazı. İtiraz Süpervizörü 3 uzman ajanı (Kabul Ajanı, Kanıt Ajanı, Dosyalama Ajanı) 4 sistem üzerinde yönetir (Card Management, Document Store (S3), Visa / Mastercard Dispute API, Email (Exchange)); 10 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Ters ibraz şemaya dosyalandı, geçici alacak yansıtıldı, müşteri takvimi biliyor. Onay kuralı: Tutar > 2.500 TL → geçici alacak öncesi itiraz ekibi lideri onayı. Bankacılık ### Ters ibraz (chargeback) itirazı [Ters ibraz (chargeback) itirazı] çalıştırma #41.288.461süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Tutar > 2.500 TL → geçici alacak öncesi itiraz ekibi lideri onayı Bu çalıştırmadaki ajanlar İtiraz Süpervizörüsupervisorplanlar, yönlendirir, onaylar Kabul Ajanıaginies-sunkar-1 · on-premİtirazı sınıflar, işlemi ve neden kodunu belirler Kanıt Ajanıaginies-tulpar-4 · on-premİşlem ve üye işyeri verisiyle kanıt paketini hazırlar Dosyalama Ajanıaginies-talay-2 · on-premGeçici alacağı yansıtır, şemaya dosyalar, müşteriyi bilgilendirir Denetim izi00:00.00 - ◦Uygulamadan itiraz: 12.03 tarihli 3.200 TL işlem, "ürün teslim edilmedi" - ✓Neden kodu 13.1 (mal teslim edilmedi) · 120 günlük süre içinde → geçerli - ⌘Plan: işlem detayı → üye işyeri geçmişi → kanıt paketi → şemaya dosyala - ⚙İşlem detayı çekildi; üye işyeri MCC 5732, son 90 günde 14 itiraz - ⌕Müşterinin yüklediği sipariş e-postası ve kargo takip ekranı alındı - ∿Kanıt özeti: teslimat kaydı yok, satıcı yanıtsız → güçlü dosya - ⏸Tutar > 2.500 TL → itiraz ekibi lideri geçici alacağı onayladı - ⇢Geçici alacak 3.200 TL hesaba yansıtıldı - ⇢Ters ibraz Visa'ya dosyalandı, vaka #CB-20481 - ➤Müşteriye e-posta: vaka numarası ve 45 günlük takvim - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Uygulamada veya şubede "bu işlemi tanımıyorum / ürün gelmedi" itirazı Sonuç Ters ibraz şemaya dosyalandı, geçici alacak yansıtıldı, müşteri takvimi biliyor ## Ajanlar SÜPERVİZÖR ### İtiraz Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Kabul Ajanı İtirazı sınıflar, işlemi ve neden kodunu belirler classify_disputeextract_transactionmap_reason_code aginies-tulpar-4 · on-prem ### Kanıt Ajanı İşlem ve üye işyeri verisiyle kanıt paketini hazırlar get_transaction_detailget_merchant_historyfetch_customer_docsdraft_dispute_summary aginies-talay-2 · on-prem ### Dosyalama Ajanı Geçici alacağı yansıtır, şemaya dosyalar, müşteriyi bilgilendirir post_provisional_creditfile_chargebacksend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | İtiraz Süpervizörü | Uygulamadan itiraz: 12.03 tarihli 3.200 TL işlem, "ürün teslim edilmedi" | — | 200 | — | | 02 | doğrulama | Kabul Ajanı | Neden kodu 13.1 (mal teslim edilmedi) · 120 günlük süre içinde → geçerli | — | 560 | 520 | | 03 | plan | İtiraz Süpervizörü | Plan: işlem detayı → üye işyeri geçmişi → kanıt paketi → şemaya dosyala | — | 300 | 210 | | 04 | araç | Kanıt Ajanı | İşlem detayı çekildi; üye işyeri MCC 5732, son 90 günde 14 itiraz | Card Management | 440 | — | | 05 | bilgi | Kanıt Ajanı | Müşterinin yüklediği sipariş e-postası ve kargo takip ekranı alındı | Document Store (S3) | 380 | — | | 06 | akıl yürütme | Kanıt Ajanı | Kanıt özeti: teslimat kaydı yok, satıcı yanıtsız → güçlü dosya | — | 1.100 | 1.450 | | 07 | insan onayı | İtiraz Süpervizörü | Tutar > 2.500 TL → itiraz ekibi lideri geçici alacağı onayladı | — | 2.800 | — | | 08 | yazma | Dosyalama Ajanı | Geçici alacak 3.200 TL hesaba yansıtıldı | Card Management | 480 | — | | 09 | yazma | Dosyalama Ajanı | Ters ibraz Visa'ya dosyalandı, vaka #CB-20481 | Visa / Mastercard Dispute API | 620 | — | | 10 | bildirim | Dosyalama Ajanı | Müşteriye e-posta: vaka numarası ve 45 günlük takvim | Email (Exchange) | 240 | — | Simülasyon metrikleri (tahmini) 9.800çalıştırma / ay %95,2başarı 7,1 sp50 $0,0048tahmini maliyet / çalıştırma Onay kuralı Tutar > 2.500 TL → geçici alacak öncesi itiraz ekibi lideri onayı Adım profili giriş · 1plan · 1bilgi · 1akıl yürütme · 1araç · 1doğrulama · 1insan onayı · 1yazma · 2bildirim · 1 SİSTEMLER - Card Managementcore - Document Store (S3)storage - Visa / Mastercard Dispute APIapi - Email (Exchange)email Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Doküman deposuna okuma erişimi ve belge tipleri - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Kart ihraççılarının Visa/Mastercard itiraz kurallarına uygun uçtan uca dispute otomasyonu deseni Bu senaryonun yer aldığı paketler Dolandırıcılık ve itiraz operasyonları → Sektör: Bankacılık → İlgili ajanlar BankacılıkHITL ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı Müşteri konuşur ya da yazar; fatura, limit, itiraz ve transfer aynı oturumda çekirdek sistemde tamamlanır. 186.000 çalıştırma / ay %94,7 başarı 5 ajan BankacılıkHITL ### Kredi kartı limit artışı Müşteri mobilde yazar ya da çağrı merkezinde konuşur; karar saniyeler içinde verilir, eşik üstü yönetici onayına gider. 18.400 çalıştırma / ay %96,4 başarı 4 ajan BankacılıkHITL ### Dijital müşteri edinimi (KYC) Kimlik, canlılık ve yaptırım kontrolü tek akışta; temiz başvurular şubeye gitmeden dakikalarda hesap açar. 42.000 çalıştırma / ay %93,8 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # CI/CD sürüm hattı URL: https://www.aginies.com/tr/ajanlar/cicd-release-pipeline Language: tr Summary: Derleme, test, tarama ve staging otomatik; üretime yalnızca sürüm yöneticisi onayıyla ArgoCD ile çıkılır. main dalına birleştirme veya v2.14.0 sürüm etiketi → v2.14.0 üretimde, sağlık kontrolleri yeşil, sürüm notları yayımlandı, geri alma planı hazır Yazılım Geliştirmeapibatch # CI/CD sürüm hattı Derleme, test, tarama ve staging otomatik; üretime yalnızca sürüm yöneticisi onayıyla ArgoCD ile çıkılır. Kısaca CI/CD sürüm hattı, Yazılım Geliştirme sektöründe api, batch kanalında çalışan bir ajan senaryosudur. Tetikleyici: main dalına birleştirme veya v2.14.0 sürüm etiketi. Sürüm Süpervizörü 4 uzman ajanı (Derleme Ajanı, Güvenlik Ajanı, Test Ajanı, Sürüm Ajanı) 4 sistem üzerinde yönetir (CI Runner (GitHub Actions), SonarQube + Snyk, Kubernetes / ArgoCD, Slack); 10 adımın 2 tanesi sistemde işlem, 3 tanesi doğrulama veya onaydır. Sonuç: v2.14.0 üretimde, sağlık kontrolleri yeşil, sürüm notları yayımlandı, geri alma planı hazır. Onay kuralı: Her üretim dağıtımı → sürüm yöneticisi onayı zorunlu; kritik CVE varsa kapı otomatik kapanır. Yazılım Geliştirme ### CI/CD sürüm hattı [CI/CD sürüm hattı] çalıştırma #41.290.163süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Her üretim dağıtımı → sürüm yöneticisi onayı zorunlu; kritik CVE varsa kapı otomatik kapanır Bu çalıştırmadaki ajanlar Sürüm Süpervizörüsupervisorplanlar, yönlendirir, onaylar Derleme Ajanıaginies-sunkar-1 · on-premDerler, birim/entegrasyon testlerini koşturur, imajı kaydeder Güvenlik Ajanıaginies-tulpar-4 · on-premSAST ve bağımlılık taramasını değerlendirir, politikaya göre kapı kararı verir Test Ajanıaginies-tulpar-4 · on-premStaging'de duman testlerini koşturur, sonuçları taban çizgisiyle karşılaştırır Sürüm Ajanıaginies-talay-2 · on-premArgoCD ile dağıtır, sürüm notlarını yazar, ekibi bilgilendirir Denetim izi00:00.00 - ◦Etiket v2.14.0 main'de: 23 commit, 4 PR, son sürümden bu yana 6 gün - ⚙CI: derleme 4 dk 12 sn, 1.284 birim + 96 entegrasyon testi geçti, imaj kaydedildi - ⚙SAST: 0 kritik, 2 orta; bağımlılık: 1 yüksek CVE (lodash 4.17.20) - ✓Yüksek CVE erişilemez kod yolunda → politika: geç, takip görevi PROJ-4877 açıldı - ⌘Plan: staging dağıtımı → duman testleri → sürüm yöneticisi onayı → üretim → notlar - ⇢Staging'e dağıtıldı (argocd sync billing-api-staging), 3/3 pod hazır - ✓Duman testleri 42/42 geçti; p95 gecikme 184 ms, taban çizgisinden %3 iyi - ⏸Üretim dağıtımı → sürüm yöneticisi değişiklik kaydı CHG-2107'yi onayladı - ⇢Üretime canary %10 → %100 dağıtıldı, sağlık kontrolleri yeşil, geri alma hazır - ➤Sürüm notları yayımlandı; #releases kanalına özet ve 4 PR bağlantısı gönderildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici main dalına birleştirme veya v2.14.0 sürüm etiketi Sonuç v2.14.0 üretimde, sağlık kontrolleri yeşil, sürüm notları yayımlandı, geri alma planı hazır ## Ajanlar SÜPERVİZÖR ### Sürüm Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Derleme Ajanı Derler, birim/entegrasyon testlerini koşturur, imajı kaydeder run_buildrun_testspush_image aginies-tulpar-4 · on-prem ### Güvenlik Ajanı SAST ve bağımlılık taramasını değerlendirir, politikaya göre kapı kararı verir run_sastscan_dependenciestriage_findings aginies-tulpar-4 · on-prem ### Test Ajanı Staging'de duman testlerini koşturur, sonuçları taban çizgisiyle karşılaştırır run_smoke_testscheck_healthcompare_baseline aginies-talay-2 · on-prem ### Sürüm Ajanı ArgoCD ile dağıtır, sürüm notlarını yazar, ekibi bilgilendirir deploy_argocddraft_release_notespost_slack ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Sürüm Süpervizörü | Etiket v2.14.0 main'de: 23 commit, 4 PR, son sürümden bu yana 6 gün | — | 180 | — | | 02 | araç | Derleme Ajanı | CI: derleme 4 dk 12 sn, 1.284 birim + 96 entegrasyon testi geçti, imaj kaydedildi | CI Runner (GitHub Actions) | 3.600 | — | | 03 | araç | Güvenlik Ajanı | SAST: 0 kritik, 2 orta; bağımlılık: 1 yüksek CVE (lodash 4.17.20) | SonarQube + Snyk | 2.400 | — | | 04 | doğrulama | Güvenlik Ajanı | Yüksek CVE erişilemez kod yolunda → politika: geç, takip görevi PROJ-4877 açıldı | — | 900 | 1.100 | | 05 | plan | Sürüm Süpervizörü | Plan: staging dağıtımı → duman testleri → sürüm yöneticisi onayı → üretim → notlar | — | 280 | 200 | | 06 | yazma | Sürüm Ajanı | Staging'e dağıtıldı (argocd sync billing-api-staging), 3/3 pod hazır | Kubernetes / ArgoCD | 1.800 | — | | 07 | doğrulama | Test Ajanı | Duman testleri 42/42 geçti; p95 gecikme 184 ms, taban çizgisinden %3 iyi | — | 2.600 | 600 | | 08 | insan onayı | Sürüm Süpervizörü | Üretim dağıtımı → sürüm yöneticisi değişiklik kaydı CHG-2107'yi onayladı | — | 4.000 | — | | 09 | yazma | Sürüm Ajanı | Üretime canary %10 → %100 dağıtıldı, sağlık kontrolleri yeşil, geri alma hazır | Kubernetes / ArgoCD | 3.400 | — | | 10 | bildirim | Sürüm Ajanı | Sürüm notları yayımlandı; #releases kanalına özet ve 4 PR bağlantısı gönderildi | Slack | 320 | 700 | Simülasyon metrikleri (tahmini) 950çalıştırma / ay %94,7başarı 18,9 sp50 $0,0075tahmini maliyet / çalıştırma Onay kuralı Her üretim dağıtımı → sürüm yöneticisi onayı zorunlu; kritik CVE varsa kapı otomatik kapanır Adım profili giriş · 1plan · 1araç · 2doğrulama · 2insan onayı · 1yazma · 2bildirim · 1 SİSTEMLER - CI Runner (GitHub Actions)api - SonarQube + Snykapi - Kubernetes / ArgoCDapi - Slackchannel Çalıştırmak için gerekenler - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek GitOps ve ilerici dağıtım uygulayan büyük yazılım ekiplerinin ajanla güçlendirilmiş sürüm hattı deseni Bu senaryonun yer aldığı paketler SDLC ajan ekibi → Sektör: Yazılım Geliştirme → İlgili ajanlar Yazılım GeliştirmeHITL ### Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR Jira'daki bir epic'ten başlar; eski modül okunur, davranışı testlerle sabitlenir, yeniden yazılır ve kanıtlı bir PR açılır. 1.900 çalıştırma / ay %90,4 başarı 5 ajan Yazılım GeliştirmeHITL ### Özellik teslimat hattı Bir Jira epic'i ajanlarla kriter, tasarım, kod ve teste dönüşür; insan yalnızca PR'ı inceler ve onaylar. 1.200 çalıştırma / ay %91,2 başarı 5 ajan Yazılım GeliştirmeHITL ### Olay müdahalesi ve acil düzeltme Alarm gelir, hata yeniden üretilir, yama ve regresyon hazır; nöbetçi onaylar, post-mortem taslağı bekler. 320 çalıştırma / ay %90,5 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Vatandaş başvurusu yönlendirme URL: https://www.aginies.com/tr/ajanlar/citizen-request-routing Language: tr Summary: e-Devlet başvurusu saniyeler içinde sınıflandırılır, yasal süresi hesaplanır ve doğru birime SLA ile atanır. e-Devlet Kapısı üzerinden yeni başvuru → Dosya doğru birimde SLA ile açıldı, vatandaş dosya numarası ve süreyle bilgilendirildi Kamu ve Hükümetwebappemailvoice # Vatandaş başvurusu yönlendirme e-Devlet başvurusu saniyeler içinde sınıflandırılır, yasal süresi hesaplanır ve doğru birime SLA ile atanır. Kısaca Vatandaş başvurusu yönlendirme, Kamu ve Hükümet sektöründe web, app, email, voice kanalında çalışan bir ajan senaryosudur. Tetikleyici: e-Devlet Kapısı üzerinden yeni başvuru. Başvuru Süpervizörü 3 uzman ajanı (Sınıflandırma Ajanı, SLA Ajanı, Dosya Ajanı) 4 sistem üzerinde yönetir (e-Devlet Gateway, Case Management, Legislation Vector Store, SMS Gateway); 8 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Dosya doğru birimde SLA ile açıldı, vatandaş dosya numarası ve süreyle bilgilendirildi. Onay kuralı: Sınıflandırma güveni < 0,85 veya personel şikâyeti → evrak kayıt memuru onayı. Kamu ve Hükümet ### Vatandaş başvurusu yönlendirme [Vatandaş başvurusu yönlendirme] çalıştırma #41.287.179süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Sınıflandırma güveni < 0,85 veya personel şikâyeti → evrak kayıt memuru onayı Bu çalıştırmadaki ajanlar Başvuru Süpervizörüsupervisorplanlar, yönlendirir, onaylar Sınıflandırma Ajanıaginies-sunkar-1 · on-premKonuyu ve birimi belirler, kişisel veriyi maskeler SLA Ajanıaginies-talay-2 · on-premİlgili mevzuattan yasal süreyi hesaplar Dosya Ajanıaginies-sunkar-1 · on-premDosyayı açar, atar ve vatandaşı bilgilendirir Denetim izi00:00.00 - ◦e-Devlet üzerinden başvuru alındı, kimlik kapıdan doğrulandı (TC) - ∿Konu: "su faturası itirazı" · birim: Su ve Kanalizasyon · güven 0,81 · KVKK maskeli - ⏸Güven 0,81 < 0,85 → evrak kayıt memuru birimi teyit etti - ⌕3071 sayılı Dilekçe Kanunu ve 4982 sayılı Kanun → 15 iş günü - ∿Süre: 29 Eylül 2026; 10. gün hatırlatma, 13. gün yükseltme - ✓Mükerrer kontrol: aynı vatandaştan 30 günde aynı konuda açık dosya yok - ⇢CM-2026-114882 dosyası açıldı, Su ve Kanalizasyon'a atandı, SLA 15 gün - ➤Vatandaşa SMS: dosya numarası ve 15 günlük süre; durum e-Devlet'te görünür - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici e-Devlet Kapısı üzerinden yeni başvuru Sonuç Dosya doğru birimde SLA ile açıldı, vatandaş dosya numarası ve süreyle bilgilendirildi ## Ajanlar SÜPERVİZÖR ### Başvuru Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Sınıflandırma Ajanı Konuyu ve birimi belirler, kişisel veriyi maskeler classify_requestextract_entitiesmask_pii aginies-talay-2 · on-prem ### SLA Ajanı İlgili mevzuattan yasal süreyi hesaplar search_regulationscompute_deadline aginies-sunkar-1 · on-prem ### Dosya Ajanı Dosyayı açar, atar ve vatandaşı bilgilendirir check_duplicatecreate_caseassign_unitsend_sms ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Başvuru Süpervizörü | e-Devlet üzerinden başvuru alındı, kimlik kapıdan doğrulandı (TC) | — | 200 | — | | 02 | akıl yürütme | Sınıflandırma Ajanı | Konu: "su faturası itirazı" · birim: Su ve Kanalizasyon · güven 0,81 · KVKK maskeli | — | 620 | 540 | | 03 | insan onayı | Başvuru Süpervizörü | Güven 0,81 Simülasyon metrikleri (tahmini) 54.000çalıştırma / ay %98,1başarı 4,9 sp50 $0,0009tahmini maliyet / çalıştırma Onay kuralı Sınıflandırma güveni < 0,85 veya personel şikâyeti → evrak kayıt memuru onayı Adım profili giriş · 1bilgi · 1akıl yürütme · 2doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - e-Devlet Gatewayapi - Case Managementticketing - Legislation Vector Storedb - SMS Gatewaychannel Çalıştırmak için gerekenler - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Estonya ve GOV.UK gibi tek durak dijital devlet hizmetlerinin başvuru yönlendirme modelini yansıtır Bu senaryonun yer aldığı paketler E-posta sınıflandırma ve ek doğrulama →Vatandaş hizmetleri ve dosya ön inceleme → Sektör: Kamu ve Hükümet → İlgili ajanlar Kamu ve HükümetHITL ### Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi Vatandaş yazar ya da arar; uygunluk kontrol edilir, belgeler okunur, başvuru açılır, randevu verilir, her adım izlenebilir. 230.000 çalıştırma / ay %93,8 başarı 5 ajan Kamu ve HükümetHITL ### Ruhsat başvurusu ön inceleme Eksik belge ve mevzuat uyumu dakikalar içinde raporlanır; kararı her zaman ruhsat memuru verir. 3.100 çalıştırma / ay %95,8 başarı 4 ajan Kamu ve HükümetHITL ### Kamu ihalesi teklif dosyası ön kontrolü Teklif dosyaları şartnameyle karşılaştırılır, eksik ve uygunsuz belgeler listelenir; değerlendirme kararı komisyonun. 380 çalıştırma / ay %94,8 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi URL: https://www.aginies.com/tr/ajanlar/citizen-services-assistant Language: tr Summary: Vatandaş yazar ya da arar; uygunluk kontrol edilir, belgeler okunur, başvuru açılır, randevu verilir, her adım izlenebilir. Belediye çağrı merkezi, WhatsApp veya e-Devlet yönlendirmeli başvuru talebi → Başvuru dosyası açıldı, eksik belge tamamlandı, randevu verildi, takip numarası iletildi Kamu ve Hükümetwhatsappvoiceweb # Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi Vatandaş yazar ya da arar; uygunluk kontrol edilir, belgeler okunur, başvuru açılır, randevu verilir, her adım izlenebilir. Kısaca Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi, Kamu ve Hükümet sektöründe whatsapp, voice, web kanalında çalışan bir ajan senaryosudur. Tetikleyici: Belediye çağrı merkezi, WhatsApp veya e-Devlet yönlendirmeli başvuru talebi. Vatandaş Hizmetleri Süpervizörü 4 uzman ajanı (Diyalog Ajanı, Uygunluk Ajanı, Belge Ajanı, Dosya Ajanı) 5 sistem üzerinde yönetir (WhatsApp Business, e-Government Gateway, Case Management, Document Store, Appointment System); 10 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Başvuru dosyası açıldı, eksik belge tamamlandı, randevu verildi, takip numarası iletildi. Onay kuralı: Uygunluk sınırda, belge tutarsız veya aynı hanede başka başvuru → memur incelemesi. Kamu ve Hükümet ### Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi [Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi] çalıştırma #41.289.038süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Uygunluk sınırda, belge tutarsız veya aynı hanede başka başvuru → memur incelemesi Bu çalıştırmadaki ajanlar Vatandaş Hizmetleri Süpervizörüsupervisorplanlar, yönlendirir, onaylar Diyalog Ajanıaginies-talay-2 · on-premVatandaşla sade dille konuşur, talebi ve kimliği netleştirir Uygunluk Ajanıaginies-tulpar-4 · on-premMevzuat ve kriterlere göre uygunluğu değerlendirir, gerekçe yazar Belge Ajanıaginies-konrul-1 · on-premYüklenen belgeleri okur, alanları çıkarır, eksikleri listeler Dosya Ajanıaginies-sunkar-1 · on-premBaşvuruyu açar, randevuyu verir, memura atar Denetim izi00:00.00 - ◦WhatsApp: "Doğalgaz yardımı için başvurmak istiyorum, kira kontratımı ve maaş bordromu gönderiyorum" - ✓Kimlik e-Devlet doğrulama bağlantısıyla teyit edildi; hane bilgisi nüfus kaydından alındı - ⌘Plan: belgeleri oku → kriterler → uygunluk → eksikler → dosya → randevu → takip no - ∿Kira kontratı: 9.500 TL/ay, 12 ay; bordro: net 24.300 TL; ikisi de okunaklı ve imzalı - ⌕Kriterler: hane geliri eşiği, kira oranı, aynı adreste başka yardım; sosyal yardım kaydı sorgulandı - ∿Uygunluk: gelir eşiğin %8 altında, kira oranı %39 → uygun; gerekçe metni yazıldı - ✓Eksik: ikametgâh belgesi; e-Devlet'ten otomatik alındı, vatandaştan ek belge istenmedi - ⇢Başvuru #SA-2026-48811 açıldı, belgeler dosyaya bağlandı, memur ataması yapıldı - ⇢Sosyal inceleme randevusu: Salı 10:30, ev ziyareti; takvime işlendi - ➤Takip numarası, randevu ve süreç adımları sade dille WhatsApp'ta; SMS yedeği gönderildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Belediye çağrı merkezi, WhatsApp veya e-Devlet yönlendirmeli başvuru talebi Sonuç Başvuru dosyası açıldı, eksik belge tamamlandı, randevu verildi, takip numarası iletildi ## Ajanlar SÜPERVİZÖR ### Vatandaş Hizmetleri Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Diyalog Ajanı Vatandaşla sade dille konuşur, talebi ve kimliği netleştirir verify_identitydetect_intentcompose_reply aginies-tulpar-4 · on-prem ### Uygunluk Ajanı Mevzuat ve kriterlere göre uygunluğu değerlendirir, gerekçe yazar get_criteriaquery_registryassess_eligibility aginies-konrul-1 · on-prem ### Belge Ajanı Yüklenen belgeleri okur, alanları çıkarır, eksikleri listeler extract_fieldsvalidate_documentlist_missing aginies-sunkar-1 · on-prem ### Dosya Ajanı Başvuruyu açar, randevuyu verir, memura atar open_casebook_appointmentassign_officersend_sms ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Vatandaş Hizmetleri Süpervizörü | WhatsApp: "Doğalgaz yardımı için başvurmak istiyorum, kira kontratımı ve maaş bordromu gönderiyorum" | WhatsApp Business | 180 | — | | 02 | doğrulama | Diyalog Ajanı | Kimlik e-Devlet doğrulama bağlantısıyla teyit edildi; hane bilgisi nüfus kaydından alındı | e-Government Gateway | 1.600 | 160 | | 03 | plan | Vatandaş Hizmetleri Süpervizörü | Plan: belgeleri oku → kriterler → uygunluk → eksikler → dosya → randevu → takip no | — | 240 | 220 | | 04 | akıl yürütme | Belge Ajanı | Kira kontratı: 9.500 TL/ay, 12 ay; bordro: net 24.300 TL; ikisi de okunaklı ve imzalı | Document Store | 2.100 | 1.200 | | 05 | bilgi | Uygunluk Ajanı | Kriterler: hane geliri eşiği, kira oranı, aynı adreste başka yardım; sosyal yardım kaydı sorgulandı | e-Government Gateway | 900 | 420 | | 06 | akıl yürütme | Uygunluk Ajanı | Uygunluk: gelir eşiğin %8 altında, kira oranı %39 → uygun; gerekçe metni yazıldı | — | 1.300 | 700 | | 07 | doğrulama | Belge Ajanı | Eksik: ikametgâh belgesi; e-Devlet'ten otomatik alındı, vatandaştan ek belge istenmedi | e-Government Gateway | 800 | — | | 08 | yazma | Dosya Ajanı | Başvuru #SA-2026-48811 açıldı, belgeler dosyaya bağlandı, memur ataması yapıldı | Case Management | 560 | — | | 09 | yazma | Dosya Ajanı | Sosyal inceleme randevusu: Salı 10:30, ev ziyareti; takvime işlendi | Appointment System | 480 | — | | 10 | bildirim | Diyalog Ajanı | Takip numarası, randevu ve süreç adımları sade dille WhatsApp'ta; SMS yedeği gönderildi | WhatsApp Business | 440 | 260 | Simülasyon metrikleri (tahmini) 230.000çalıştırma / ay %93,8başarı 8,2 sp50 $0,0021tahmini maliyet / çalıştırma Onay kuralı Uygunluk sınırda, belge tutarsız veya aynı hanede başka başvuru → memur incelemesi Adım profili giriş · 1plan · 1bilgi · 1akıl yürütme · 2doğrulama · 2yazma · 2bildirim · 1 SİSTEMLER - WhatsApp Businesschannel - e-Government Gatewayapi - Case Managementticketing - Document Storestorage - Appointment Systemapi Çalıştırmak için gerekenler - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Doküman deposuna okuma erişimi ve belge tipleri - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Dijital kamu hizmeti programlarının "tek başvuru, sıfır tekrar belge" deseni Bu senaryonun yer aldığı paketler Vatandaş hizmetleri ve dosya ön inceleme → Sektör: Kamu ve Hükümet → İlgili ajanlar Kamu ve HükümetHITL ### Vatandaş başvurusu yönlendirme e-Devlet başvurusu saniyeler içinde sınıflandırılır, yasal süresi hesaplanır ve doğru birime SLA ile atanır. 54.000 çalıştırma / ay %98,1 başarı 4 ajan Kamu ve HükümetHITL ### Ruhsat başvurusu ön inceleme Eksik belge ve mevzuat uyumu dakikalar içinde raporlanır; kararı her zaman ruhsat memuru verir. 3.100 çalıştırma / ay %95,8 başarı 4 ajan Kamu ve HükümetHITL ### Kamu ihalesi teklif dosyası ön kontrolü Teklif dosyaları şartnameyle karşılaştırılır, eksik ve uygunsuz belgeler listelenir; değerlendirme kararı komisyonun. 380 çalıştırma / ay %94,8 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Kod incelemesi ve güvenlik kapısı URL: https://www.aginies.com/tr/ajanlar/code-review-security-gate Language: tr Summary: Her PR ajanlarca incelenir, taranır, kapsamı ölçülür; geliştirici düzeltir, insan yalnızca temiz PR'ı onaylar. GitHub'da PR açıldı veya güncellendi (webhook) → PR bulguları düzeltildi, güvenlik ve kapsam kapıları yeşil, insan onayıyla birleşmeye hazır Yazılım Geliştirmeapiweb # Kod incelemesi ve güvenlik kapısı Her PR ajanlarca incelenir, taranır, kapsamı ölçülür; geliştirici düzeltir, insan yalnızca temiz PR'ı onaylar. Kısaca Kod incelemesi ve güvenlik kapısı, Yazılım Geliştirme sektöründe api, web kanalında çalışan bir ajan senaryosudur. Tetikleyici: GitHub'da PR açıldı veya güncellendi (webhook). İnceleme Süpervizörü 4 uzman ajanı (İnceleyici Ajanı, Güvenlik Ajanı, Test Ajanı, Geliştirici Ajanı) 4 sistem üzerinde yönetir (GitHub, SonarQube, Dependency Scanner (Snyk), CI Runner (GitHub Actions)); 10 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: PR bulguları düzeltildi, güvenlik ve kapsam kapıları yeşil, insan onayıyla birleşmeye hazır. Onay kuralı: Her PR → CODEOWNERS insan onayı; yüksek SAST bulgusu veya kapsam < %80 kapıyı kapatır. Yazılım Geliştirme ### Kod incelemesi ve güvenlik kapısı [Kod incelemesi ve güvenlik kapısı] çalıştırma #41.287.209süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Her PR → CODEOWNERS insan onayı; yüksek SAST bulgusu veya kapsam < %80 kapıyı kapatır Bu çalıştırmadaki ajanlar İnceleme Süpervizörüsupervisorplanlar, yönlendirir, onaylar İnceleyici Ajanıaginies-tulpar-4 · on-premDeğişiklikleri okur; hata, tasarım ve stil bulgularını yazar Güvenlik Ajanıaginies-sunkar-1 · on-premSAST, gizli anahtar ve bağımlılık bulgularını triyajlar Test Ajanıaginies-tulpar-4 · on-premKapsam farkını ölçer, eksik testleri önerir Geliştirici Ajanıaginies-tulpar-4 · on-premBulguları düzeltir, testleri ekler, PR'ı günceller Denetim izi00:00.00 - ◦PR #2318 açıldı: "Kupon doğrulama servisi", 9 dosya, +684 / −120, yazar M.A. - ⌘Plan: inceleme + güvenlik + kapsam paralel → düzeltme → yeniden kontrol → insan onayı - ∿Diff ve 4 çağıran modül okundu; 3 bulgu: yarış durumu, null kontrolü, N+1 sorgu - ⚙SAST: 1 yüksek (SQL dize birleştirme, satır 212), gizli anahtar yok - ⚙Bağımlılık: yeni eklenen jsonwebtoken 8.5.1 → 1 orta CVE, 9.0.2 önerildi - ⚙Kapsam: yeni kod %64 (eşik %80), CouponRules için 6 test eksik - ⇢5 bulgu düzeltildi, jsonwebtoken 9.0.2, 6 test eklendi; commit 4e1b9c gönderildi - ✓Yeniden kontrol: SAST 0 yüksek, kapsam %86, 1.312 test geçti → kapılar yeşil - ⏸Kod sahibi (CODEOWNERS) PR #2318'i inceledi, 1 yorum ekledi, onayladı - ⇢PR #2318 main'e birleştirildi, inceleme izi ve tarama raporları PR'a eklendi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici GitHub'da PR açıldı veya güncellendi (webhook) Sonuç PR bulguları düzeltildi, güvenlik ve kapsam kapıları yeşil, insan onayıyla birleşmeye hazır ## Ajanlar SÜPERVİZÖR ### İnceleme Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-tulpar-4 · on-prem ### İnceleyici Ajanı Değişiklikleri okur; hata, tasarım ve stil bulgularını yazar read_diffsearch_codebasepost_review_comment aginies-sunkar-1 · on-prem ### Güvenlik Ajanı SAST, gizli anahtar ve bağımlılık bulgularını triyajlar run_sastscan_secretsscan_dependenciestriage_findings aginies-tulpar-4 · on-prem ### Test Ajanı Kapsam farkını ölçer, eksik testleri önerir run_testsreport_coveragesuggest_tests aginies-tulpar-4 · on-prem ### Geliştirici Ajanı Bulguları düzeltir, testleri ekler, PR'ı günceller edit_filesrun_testspush_commit ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | İnceleme Süpervizörü | PR #2318 açıldı: "Kupon doğrulama servisi", 9 dosya, +684 / −120, yazar M.A. | — | 160 | — | | 02 | plan | İnceleme Süpervizörü | Plan: inceleme + güvenlik + kapsam paralel → düzeltme → yeniden kontrol → insan onayı | — | 280 | 210 | | 03 | akıl yürütme | İnceleyici Ajanı | Diff ve 4 çağıran modül okundu; 3 bulgu: yarış durumu, null kontrolü, N+1 sorgu | — | 2.200 | 4.800 | | 04 | araç | Güvenlik Ajanı | SAST: 1 yüksek (SQL dize birleştirme, satır 212), gizli anahtar yok | SonarQube | 1.800 | — | | 05 | araç | Güvenlik Ajanı | Bağımlılık: yeni eklenen jsonwebtoken 8.5.1 → 1 orta CVE, 9.0.2 önerildi | Dependency Scanner (Snyk) | 900 | — | | 06 | araç | Test Ajanı | Kapsam: yeni kod %64 (eşik %80), CouponRules için 6 test eksik | CI Runner (GitHub Actions) | 2.400 | — | | 07 | yazma | Geliştirici Ajanı | 5 bulgu düzeltildi, jsonwebtoken 9.0.2, 6 test eklendi; commit 4e1b9c gönderildi | GitHub | 3.600 | 7.400 | | 08 | doğrulama | Test Ajanı | Yeniden kontrol: SAST 0 yüksek, kapsam %86, 1.312 test geçti → kapılar yeşil | — | 2.600 | 500 | | 09 | insan onayı | İnceleme Süpervizörü | Kod sahibi (CODEOWNERS) PR #2318'i inceledi, 1 yorum ekledi, onayladı | — | 4.000 | — | | 10 | yazma | Geliştirici Ajanı | PR #2318 main'e birleştirildi, inceleme izi ve tarama raporları PR'a eklendi | GitHub | 420 | — | Simülasyon metrikleri (tahmini) 4.200çalıştırma / ay %93,6başarı 17,9 sp50 $0,0129tahmini maliyet / çalıştırma Onay kuralı Her PR → CODEOWNERS insan onayı; yüksek SAST bulgusu veya kapsam < %80 kapıyı kapatır Adım profili giriş · 1plan · 1akıl yürütme · 1araç · 3doğrulama · 1insan onayı · 1yazma · 2 SİSTEMLER - GitHubapi - SonarQubeapi - Dependency Scanner (Snyk)api - CI Runner (GitHub Actions)api Çalıştırmak için gerekenler - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük teknoloji şirketlerinin yapay zekâ destekli kod incelemesi ve sola kaydırılmış güvenlik kapısı deseni Bu senaryonun yer aldığı paketler SDLC ajan ekibi → Sektör: Yazılım Geliştirme → İlgili ajanlar Yazılım GeliştirmeHITL ### Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR Jira'daki bir epic'ten başlar; eski modül okunur, davranışı testlerle sabitlenir, yeniden yazılır ve kanıtlı bir PR açılır. 1.900 çalıştırma / ay %90,4 başarı 5 ajan Yazılım GeliştirmeHITL ### Özellik teslimat hattı Bir Jira epic'i ajanlarla kriter, tasarım, kod ve teste dönüşür; insan yalnızca PR'ı inceler ve onaylar. 1.200 çalıştırma / ay %91,2 başarı 5 ajan Yazılım GeliştirmeHITL ### CI/CD sürüm hattı Derleme, test, tarama ve staging otomatik; üretime yalnızca sürüm yöneticisi onayıyla ArgoCD ile çıkılır. 950 çalıştırma / ay %94,7 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Soğuk zincir sapması müdahalesi URL: https://www.aginies.com/tr/ajanlar/cold-chain-excursion-response Language: tr Summary: Sıcaklık sapması anında değerlendirilir, ürün riski hesaplanır, saha ekibi sevk edilir, kalite kararı kayıt altına alınır. Soğuk oda ya da araç sensöründen eşik ihlali → Etkilenen partiler belirlendi, sevk yapıldı, kalite kararı ve müşteri bilgilendirmesi tamamlandı IoT ve Bağlı Cihazlarapiappwhatsapp # Soğuk zincir sapması müdahalesi Sıcaklık sapması anında değerlendirilir, ürün riski hesaplanır, saha ekibi sevk edilir, kalite kararı kayıt altına alınır. Kısaca Soğuk zincir sapması müdahalesi, IoT ve Bağlı Cihazlar sektöründe api, app, whatsapp kanalında çalışan bir ajan senaryosudur. Tetikleyici: Soğuk oda ya da araç sensöründen eşik ihlali. Soğuk Zincir Süpervizörü 3 uzman ajanı (Sensör Ajanı, Kalite Ajanı, Saha Ajanı) 5 sistem üzerinde yönetir (IoT Platform, WMS / Lot Tracking, Stability Rules (vector), Field Service, WhatsApp Business API); 9 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Etkilenen partiler belirlendi, sevk yapıldı, kalite kararı ve müşteri bilgilendirmesi tamamlandı. Onay kuralı: Ürün taşıma ya da imha kararı → kalite yöneticisi onayı. IoT ve Bağlı Cihazlar ### Soğuk zincir sapması müdahalesi [Soğuk zincir sapması müdahalesi] çalıştırma #41.287.519süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Ürün taşıma ya da imha kararı → kalite yöneticisi onayı Bu çalıştırmadaki ajanlar Soğuk Zincir Süpervizörüsupervisorplanlar, yönlendirir, onaylar Sensör Ajanıaginies-sunkar-1 · on-premSapmayı doğrular ve süresini hesaplar Kalite Ajanıaginies-tulpar-4 · on-premÜrün stabilite kurallarıyla riski değerlendirir Saha Ajanıaginies-talay-2 · on-premTeknisyen sevk eder ve müşteriyi bilgilendirir Denetim izi00:00.00 - ◦Soğuk oda CR-4: 8,6 °C, eşik 8 °C; 14 dakikadır sapmada - ⚙Son 2 saat serisi çekildi; ikinci sensör doğruluyor, kompresör akımı düşük - ∿Sapma gerçek, eğilim +0,3 °C / 10 dk; 10 °C’ye ulaşma tahmini 45 dk - ⚙CR-4’teki partiler: 18 parti, 3 ürün ailesi, en hassas: aşı partisi L-2291 - ⌕Stabilite kuralı: 2–8 °C, kümülatif sapma toleransı 4 saat @ ≤ 12 °C - ✓Kümülatif sapma 14 dk + tahmin 45 dk < 4 saat → ürün riski düşük, izleme sürecek - ⇢Servis emri FS-2026-8812: kompresör kontrolü, öncelik acil, en yakın teknisyen 25 dk - ⏸Aşı partisi etkilenirse taşıma kararı → kalite yöneticisi bekleme kararını onayladı - ➤Depo sorumlusuna WhatsApp: durum, teknisyen ETA, izleme penceresi; kayıt kalite dosyasına yazıldı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Soğuk oda ya da araç sensöründen eşik ihlali Sonuç Etkilenen partiler belirlendi, sevk yapıldı, kalite kararı ve müşteri bilgilendirmesi tamamlandı ## Ajanlar SÜPERVİZÖR ### Soğuk Zincir Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Sensör Ajanı Sapmayı doğrular ve süresini hesaplar get_sensor_seriesvalidate_readingcompute_excursion aginies-tulpar-4 · on-prem ### Kalite Ajanı Ürün stabilite kurallarıyla riski değerlendirir get_lots_in_locationsearch_stability_rulesscore_product_risk aginies-talay-2 · on-prem ### Saha Ajanı Teknisyen sevk eder ve müşteriyi bilgilendirir create_service_orderdispatch_techniciansend_whatsapp ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Soğuk Zincir Süpervizörü | Soğuk oda CR-4: 8,6 °C, eşik 8 °C; 14 dakikadır sapmada | IoT Platform | 200 | — | | 02 | araç | Sensör Ajanı | Son 2 saat serisi çekildi; ikinci sensör doğruluyor, kompresör akımı düşük | IoT Platform | 420 | — | | 03 | akıl yürütme | Sensör Ajanı | Sapma gerçek, eğilim +0,3 °C / 10 dk; 10 °C’ye ulaşma tahmini 45 dk | — | 600 | 520 | | 04 | araç | Kalite Ajanı | CR-4’teki partiler: 18 parti, 3 ürün ailesi, en hassas: aşı partisi L-2291 | WMS / Lot Tracking | 440 | — | | 05 | bilgi | Kalite Ajanı | Stabilite kuralı: 2–8 °C, kümülatif sapma toleransı 4 saat @ ≤ 12 °C | Stability Rules (vector) | 480 | 460 | | 06 | doğrulama | Kalite Ajanı | Kümülatif sapma 14 dk + tahmin 45 dk Simülasyon metrikleri (tahmini) 2.700çalıştırma / ay %97,4başarı 5,9 sp50 $0,0026tahmini maliyet / çalıştırma Onay kuralı Ürün taşıma ya da imha kararı → kalite yöneticisi onayı Adım profili giriş · 1bilgi · 1akıl yürütme · 1araç · 2doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - IoT Platformcore - WMS / Lot Trackingerp - Stability Rules (vector)db - Field Serviceticketing - WhatsApp Business APIchannel Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - ERP/MES için kayıt ve sorgu API erişimi - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek İlaç ve gıda lojistiğinde GDP uyumlu soğuk zincir izleme ve sapma yönetimi pratiklerini yansıtır Bu senaryonun yer aldığı paketler Saha servis ve sevk otomasyonu →IoT cihaz filosu operasyonu →Lojistik kontrol kulesi → Sektör: IoT ve Bağlı Cihazlar → İlgili ajanlar IoT ve Bağlı CihazlarHITL ### Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyen Müşteri "cihaz çalışmıyor" der; telemetri okunur, uzaktan tanı konur, güncelleme gönderilir, gerekirse teknisyen doğru parçayla atanır. 37.000 çalıştırma / ay %94,1 başarı 5 ajan IoT ve Bağlı CihazlarHITL ### Cihaz filosu anomali triyajı Binlerce cihazdan gelen telemetri anomalileri kümelenir, kök neden aday listesi çıkar, doğru ekibe tek olay olarak düşer. 18.000 çalıştırma / ay %96,9 başarı 4 ajan IoT ve Bağlı CihazlarHITL ### Kademeli OTA güncelleme dağıtımı Firmware dağıtımı %1 → %10 → %100 kademelerinde ilerler; her kademede sağlık kapısı, kritik filolarda insan onayı. 320 çalıştırma / ay %98,1 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Erken dönem tahsilat iletişimi URL: https://www.aginies.com/tr/ajanlar/collections-outreach Language: tr Summary: Gecikmiş hesaplar segmentlenir, doğru kanaldan nazikçe hatırlatılır; müşteri ödeme planını kendisi kurar. Gece toplu işi: 5–30 gün gecikmeli hesap listesi → Ödeme planı kayıtlı, gecikme faizi donduruldu, müşteri taksit tarihlerini biliyor Bankacılıkbatchwhatsappsmsvoice # Erken dönem tahsilat iletişimi Gecikmiş hesaplar segmentlenir, doğru kanaldan nazikçe hatırlatılır; müşteri ödeme planını kendisi kurar. Kısaca Erken dönem tahsilat iletişimi, Bankacılık sektöründe batch, whatsapp, sms, voice kanalında çalışan bir ajan senaryosudur. Tetikleyici: Gece toplu işi: 5–30 gün gecikmeli hesap listesi. Tahsilat Süpervizörü 3 uzman ajanı (Segment Ajanı, Diyalog Ajanı, Ödeme Planı Ajanı) 4 sistem üzerinde yönetir (Core Banking, Collections Platform, WhatsApp Business API, SMS Gateway); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Ödeme planı kayıtlı, gecikme faizi donduruldu, müşteri taksit tarihlerini biliyor. Onay kuralı: Plan bakiyesi > 25.000 TL veya > 3 taksit → tahsilat ekip lideri onayı. Bankacılık ### Erken dönem tahsilat iletişimi [Erken dönem tahsilat iletişimi] çalıştırma #41.287.984süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Plan bakiyesi > 25.000 TL veya > 3 taksit → tahsilat ekip lideri onayı Bu çalıştırmadaki ajanlar Tahsilat Süpervizörüsupervisorplanlar, yönlendirir, onaylar Segment Ajanıaginies-talay-2 · on-premÖdeme eğilimini skorlar, kanal ve tonu seçer Diyalog Ajanıaginies-tulpar-4 · on-premMüşteriyle konuşur, ödeme niyetini ve plan talebini anlar Ödeme Planı Ajanıaginies-sunkar-1 · on-premPolitikaya uygun planı hesaplar ve kaydeder Denetim izi00:00.00 - ◦Gece toplu işi: 5–30 gün gecikmeli 12.400 hesap listelendi - ⚙Bakiye, gecikme günü ve son 6 ay ödeme davranışı çekildi - ∿Müşteri A.K.: 22 gün gecikme, 32.600 TL, ödeme eğilimi 0,68 → WhatsApp - ⌘Plan: nazik hatırlatma → yanıt → plan teklifi → onay → kayıt - ➤WhatsApp: hatırlatma ve "bugün öderim / plan istiyorum" seçenekleri - ∿Müşteri: "maaş 25'inde, üçe bölebilir miyiz?" → plan talebi - ✓Politika: ≤ 3 taksit, ilk taksit ≤ 15 gün → 3 × 10.867 TL uygun, eşik üstü - ⏸Bakiye > 25.000 TL → tahsilat ekip lideri planı onayladı - ⇢Ödeme planı #PP-33120 oluşturuldu, gecikme faizi donduruldu - ➤SMS ile plan özeti ve ilk taksit tarihi (25.09) gönderildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Gece toplu işi: 5–30 gün gecikmeli hesap listesi Sonuç Ödeme planı kayıtlı, gecikme faizi donduruldu, müşteri taksit tarihlerini biliyor ## Ajanlar SÜPERVİZÖR ### Tahsilat Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Segment Ajanı Ödeme eğilimini skorlar, kanal ve tonu seçer get_delinquent_accountsscore_propensitypick_channel aginies-tulpar-4 · on-prem ### Diyalog Ajanı Müşteriyle konuşur, ödeme niyetini ve plan talebini anlar send_whatsappparse_replypropose_plan aginies-sunkar-1 · on-prem ### Ödeme Planı Ajanı Politikaya uygun planı hesaplar ve kaydeder calculate_plancreate_payment_planfreeze_late_interestsend_sms ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Tahsilat Süpervizörü | Gece toplu işi: 5–30 gün gecikmeli 12.400 hesap listelendi | — | 300 | — | | 02 | araç | Segment Ajanı | Bakiye, gecikme günü ve son 6 ay ödeme davranışı çekildi | Core Banking | 520 | — | | 03 | akıl yürütme | Segment Ajanı | Müşteri A.K.: 22 gün gecikme, 32.600 TL, ödeme eğilimi 0,68 → WhatsApp | — | 680 | 640 | | 04 | plan | Tahsilat Süpervizörü | Plan: nazik hatırlatma → yanıt → plan teklifi → onay → kayıt | — | 260 | 180 | | 05 | bildirim | Diyalog Ajanı | WhatsApp: hatırlatma ve "bugün öderim / plan istiyorum" seçenekleri | WhatsApp Business API | 250 | — | | 06 | akıl yürütme | Diyalog Ajanı | Müşteri: "maaş 25'inde, üçe bölebilir miyiz?" → plan talebi | — | 540 | 470 | | 07 | doğrulama | Ödeme Planı Ajanı | Politika: ≤ 3 taksit, ilk taksit ≤ 15 gün → 3 × 10.867 TL uygun, eşik üstü | — | 610 | 520 | | 08 | insan onayı | Tahsilat Süpervizörü | Bakiye > 25.000 TL → tahsilat ekip lideri planı onayladı | — | 2.700 | — | | 09 | yazma | Ödeme Planı Ajanı | Ödeme planı #PP-33120 oluşturuldu, gecikme faizi donduruldu | Collections Platform | 480 | — | | 10 | bildirim | Ödeme Planı Ajanı | SMS ile plan özeti ve ilk taksit tarihi (25.09) gönderildi | SMS Gateway | 220 | — | Simülasyon metrikleri (tahmini) 96.000çalıştırma / ay %91,2başarı 6,5 sp50 $0,0008tahmini maliyet / çalıştırma Onay kuralı Plan bakiyesi > 25.000 TL veya > 3 taksit → tahsilat ekip lideri onayı Adım profili giriş · 1plan · 1akıl yürütme · 2araç · 1doğrulama · 1insan onayı · 1yazma · 1bildirim · 2 SİSTEMLER - Core Bankingcore - Collections Platformdb - WhatsApp Business APIchannel - SMS Gatewaychannel Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Dijital öncelikli tahsilat: ABD/AB tahsilat platformlarının self-servis ödeme planı deseni Bu senaryonun yer aldığı paketler Dolandırıcılık ve itiraz operasyonları → Sektör: Bankacılık → İlgili ajanlar BankacılıkHITL ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı Müşteri konuşur ya da yazar; fatura, limit, itiraz ve transfer aynı oturumda çekirdek sistemde tamamlanır. 186.000 çalıştırma / ay %94,7 başarı 5 ajan BankacılıkHITL ### Kredi kartı limit artışı Müşteri mobilde yazar ya da çağrı merkezinde konuşur; karar saniyeler içinde verilir, eşik üstü yönetici onayına gider. 18.400 çalıştırma / ay %96,4 başarı 4 ajan BankacılıkHITL ### Dijital müşteri edinimi (KYC) Kimlik, canlılık ve yaptırım kontrolü tek akışta; temiz başvurular şubeye gitmeden dakikalarda hesap açar. 42.000 çalıştırma / ay %93,8 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Sözleşme madde çıkarımı URL: https://www.aginies.com/tr/ajanlar/contract-clause-extraction Language: tr Summary: Sözleşme maddeleri çıkarılır, hukuk playbook'u ile karşılaştırılır; sapmalar redline önerisiyle avukata gider. Satın almadan inceleme için tedarikçi sözleşmesi gelmesi → Sapma kaydı ve risk skoru CLM'de, redline v1 hazır, avukat ve satın alma bilgilendirildi Kurumsal Fonksiyonlaremailweb # Sözleşme madde çıkarımı Sözleşme maddeleri çıkarılır, hukuk playbook'u ile karşılaştırılır; sapmalar redline önerisiyle avukata gider. Kısaca Sözleşme madde çıkarımı, Kurumsal Fonksiyonlar sektöründe email, web kanalında çalışan bir ajan senaryosudur. Tetikleyici: Satın almadan inceleme için tedarikçi sözleşmesi gelmesi. Sözleşme Süpervizörü 3 uzman ajanı (Madde Ajanı, Playbook Ajanı, CLM Ajanı) 4 sistem üzerinde yönetir (CLM (Contract Lifecycle Management), Legal Playbook Vector Store, Document Store (S3), Exchange / M365 Mail); 9 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Sapma kaydı ve risk skoru CLM'de, redline v1 hazır, avukat ve satın alma bilgilendirildi. Onay kuralı: Yüksek riskli sapma veya sözleşme değeri > 500.000 € → hukuk müşaviri onayı. Kurumsal Fonksiyonlar ### Sözleşme madde çıkarımı [Sözleşme madde çıkarımı] çalıştırma #41.286.687süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Yüksek riskli sapma veya sözleşme değeri > 500.000 € → hukuk müşaviri onayı Bu çalıştırmadaki ajanlar Sözleşme Süpervizörüsupervisorplanlar, yönlendirir, onaylar Madde Ajanıaginies-tulpar-4 · on-premMaddeleri çıkarır ve türüne göre sınıflandırır Playbook Ajanıaginies-tulpar-4 · on-premPlaybook pozisyonlarıyla karşılaştırır, redline önerir CLM Ajanıaginies-talay-2 · on-premCLM kaydını günceller ve tarafları bilgilendirir Denetim izi00:00.00 - ◦Satın almadan 38 sayfalık tedarikçi MSA'sı (DOCX) inceleme için alındı - ∿27 madde çıkarıldı: sorumluluk, tazminat, fesih, veri koruma, fikri mülkiyet… - ⌕Playbook v2026.2: 27 madde türü için standart pozisyon ve alternatifler getirildi - ∿4 sapma: sorumluluk tavanı 1× (std 2×), 30 gün fesih (std 90), DPA yok, oto-yenileme - ✓Risk: 2 yüksek, 2 orta; yetkili mahkeme ve uygulanacak hukuk playbook ile uyumlu - ⏸Yüksek riskli sapma → hukuk müşaviri inceledi; redline eklendi, karşı tarafa gönderilmedi - ⇢Sözleşme kaydı güncellendi: sapma listesi, risk skoru 62/100, redline v1 - ⇢Kaynak belge, madde haritası ve redline arşivlendi - ➤Satın alma ve hukuk özetle bilgilendirildi, inceleme görevi atandı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Satın almadan inceleme için tedarikçi sözleşmesi gelmesi Sonuç Sapma kaydı ve risk skoru CLM'de, redline v1 hazır, avukat ve satın alma bilgilendirildi ## Ajanlar SÜPERVİZÖR ### Sözleşme Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-tulpar-4 · on-prem ### Madde Ajanı Maddeleri çıkarır ve türüne göre sınıflandırır extract_clausesclassify_clause_type aginies-tulpar-4 · on-prem ### Playbook Ajanı Playbook pozisyonlarıyla karşılaştırır, redline önerir search_playbookcompare_positionsdraft_redline aginies-talay-2 · on-prem ### CLM Ajanı CLM kaydını günceller ve tarafları bilgilendirir update_contract_recordarchive_documentscreate_review_tasksend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Sözleşme Süpervizörü | Satın almadan 38 sayfalık tedarikçi MSA'sı (DOCX) inceleme için alındı | — | 200 | — | | 02 | akıl yürütme | Madde Ajanı | 27 madde çıkarıldı: sorumluluk, tazminat, fesih, veri koruma, fikri mülkiyet… | — | 1.600 | 2.400 | | 03 | bilgi | Playbook Ajanı | Playbook v2026.2: 27 madde türü için standart pozisyon ve alternatifler getirildi | Legal Playbook Vector Store | 480 | 900 | | 04 | akıl yürütme | Playbook Ajanı | 4 sapma: sorumluluk tavanı 1× (std 2×), 30 gün fesih (std 90), DPA yok, oto-yenileme | — | 1.300 | 1.800 | | 05 | doğrulama | Playbook Ajanı | Risk: 2 yüksek, 2 orta; yetkili mahkeme ve uygulanacak hukuk playbook ile uyumlu | — | 420 | 380 | | 06 | insan onayı | Sözleşme Süpervizörü | Yüksek riskli sapma → hukuk müşaviri inceledi; redline eklendi, karşı tarafa gönderilmedi | — | 3.000 | — | | 07 | yazma | CLM Ajanı | Sözleşme kaydı güncellendi: sapma listesi, risk skoru 62/100, redline v1 | CLM (Contract Lifecycle Management) | 520 | — | | 08 | yazma | CLM Ajanı | Kaynak belge, madde haritası ve redline arşivlendi | Document Store (S3) | 260 | — | | 09 | bildirim | CLM Ajanı | Satın alma ve hukuk özetle bilgilendirildi, inceleme görevi atandı | Exchange / M365 Mail | 240 | — | Simülasyon metrikleri (tahmini) 1.400çalıştırma / ay %95,1başarı 8 sp50 $0,0189tahmini maliyet / çalıştırma Onay kuralı Yüksek riskli sapma veya sözleşme değeri > 500.000 € → hukuk müşaviri onayı Adım profili giriş · 1bilgi · 1akıl yürütme · 2doğrulama · 1insan onayı · 1yazma · 2bildirim · 1 SİSTEMLER - CLM (Contract Lifecycle Management)core - Legal Playbook Vector Storedb - Document Store (S3)storage - Exchange / M365 Mailemail Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Doküman deposuna okuma erişimi ve belge tipleri - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Küresel hukuk bürolarının ve şirket içi legal ops ekiplerinin playbook tabanlı sözleşme incelemesini yansıtır Bu senaryonun yer aldığı paketler Bilgi yönetimi ve karar desteği →Kamu alımı ve ihale uyumu → İlgili ajanlar Kurumsal FonksiyonlarHITL ### Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada Talep e-postayla gelir; teklifler karşılaştırılır, sözleşme maddeleri kontrol edilir, sipariş açılır, fatura üç yönlü eşleşir ve ödeme planlanır. 14.000 çalıştırma / ay %92,7 başarı 5 ajan Kurumsal Fonksiyonlar ### E-posta sınıflandırma ve ek kontrolü Ortak kutuya gelen her e-posta sınıflandırılır, ekler listeyle karşılaştırılır, eksikler anında sorulur. 86.000 çalıştırma / ay %97,6 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Çağrı merkezi kalite izleme Örneklem değil, tüm çağrılar rubriğe göre puanlanır; uyum ihlali ve koçluk fırsatı sabah liderin önünde. 90.000 çalıştırma / ay %98,8 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA URL: https://www.aginies.com/tr/ajanlar/control-tower-exception-desk Language: tr Summary: Telematik sinyali gecikmeyi erken yakalar; alternatif rota ve taşıyıcı seçilir, müşteri bilgilendirilir, SLA tazmini hesaplanır. Telematik veya taşıyıcı API'sinden gelen sapma olayı (ETA +4 saat) → Rota değişti, müşteri yeni ETA ile bilgilendirildi, SLA kaydı ve tazmin dosyası hazır Lojistikapiemailwhatsapp # Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA Telematik sinyali gecikmeyi erken yakalar; alternatif rota ve taşıyıcı seçilir, müşteri bilgilendirilir, SLA tazmini hesaplanır. Kısaca Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA, Lojistik sektöründe api, email, whatsapp kanalında çalışan bir ajan senaryosudur. Tetikleyici: Telematik veya taşıyıcı API'sinden gelen sapma olayı (ETA +4 saat). Kontrol Kulesi Süpervizörü 4 uzman ajanı (Tahmin Ajanı, Rota Ajanı, Müşteri Ajanı, Finans Ajanı) 5 sistem üzerinde yönetir (TMS, Telematics, Carrier API, Customer Portal, ERP Billing); 10 adımın 2 tanesi sistemde işlem, 1 tanesi doğrulama veya onaydır. Sonuç: Rota değişti, müşteri yeni ETA ile bilgilendirildi, SLA kaydı ve tazmin dosyası hazır. Onay kuralı: Rota maliyet artışı > %15 veya tazmin > 10.000 TL → operasyon müdürü onayı. Lojistik ### Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA [Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA] çalıştırma #41.290.313süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Rota maliyet artışı > %15 veya tazmin > 10.000 TL → operasyon müdürü onayı Bu çalıştırmadaki ajanlar Kontrol Kulesi Süpervizörüsupervisorplanlar, yönlendirir, onaylar Tahmin Ajanıaginies-tulpar-4 · on-premSapmanın sevkiyat ve SLA etkisini tahmin eder Rota Ajanıaginies-talay-2 · on-premAlternatif rota, aktarma ve taşıyıcı seçeneklerini maliyetle karşılaştırır Müşteri Ajanıaginies-sunkar-1 · on-premAlıcıyı ve göndereni yeni ETA ve seçeneklerle bilgilendirir Finans Ajanıaginies-sunkar-1 · on-premSLA ihlalini kaydeder, tazmin ve taşıyıcı rücusunu hesaplar Denetim izi00:00.00 - ◦Telematik: araç 34 TR 7781 otoyolda 90 dakikadır duruyor; 3 sevkiyat, 2'si soğuk zincir - ∿Tahmin: ETA +4,5 saat; sevkiyat 2 SLA ihlali kesin, sevkiyat 3 sınırda; soğuk zincir 6 saat tolerans - ⌘Plan: rota/aktarma seçenekleri → maliyet → yeniden rezervasyon → müşteri → SLA kaydı - ⚙3 seçenek: yedek araçla aktarma (+2.400 TL, ETA +1,5 s), partner taşıyıcı (+3.900 TL), bekle - ✓Seçenek 1 maliyet artışı %11 < %15 eşiği; soğuk zincir toleransı içinde → otomatik onay - ⇢Yedek araç atandı, aktarma noktası ve zamanı TMS'e işlendi, sürücülere talimat gitti - ➤Alıcılara yeni ETA ve canlı takip bağlantısı; sevkiyat 2 için teslim penceresi seçimi sunuldu - ⇢SLA ihlali kaydedildi; sözleşmeye göre tazmin 1.850 TL, taşıyıcı rücu dosyası açıldı - ⚙Aktarma tamamlandı, yeni ETA teyit edildi; sıcaklık kaydı sürekli, sapma yok - ➤Gönderene olay özeti ve maliyet etkisi e-posta ile; müşteri portalında olay kapatıldı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Telematik veya taşıyıcı API'sinden gelen sapma olayı (ETA +4 saat) Sonuç Rota değişti, müşteri yeni ETA ile bilgilendirildi, SLA kaydı ve tazmin dosyası hazır ## Ajanlar SÜPERVİZÖR ### Kontrol Kulesi Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-tulpar-4 · on-prem ### Tahmin Ajanı Sapmanın sevkiyat ve SLA etkisini tahmin eder get_telematicspredict_etasla_impact aginies-talay-2 · on-prem ### Rota Ajanı Alternatif rota, aktarma ve taşıyıcı seçeneklerini maliyetle karşılaştırır route_optionscarrier_quoterebook_leg aginies-sunkar-1 · on-prem ### Müşteri Ajanı Alıcıyı ve göndereni yeni ETA ve seçeneklerle bilgilendirir notify_customeroffer_optionslog_contact aginies-sunkar-1 · on-prem ### Finans Ajanı SLA ihlalini kaydeder, tazmin ve taşıyıcı rücusunu hesaplar record_slacalc_penaltycreate_claim ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Kontrol Kulesi Süpervizörü | Telematik: araç 34 TR 7781 otoyolda 90 dakikadır duruyor; 3 sevkiyat, 2'si soğuk zincir | Telematics | 160 | — | | 02 | akıl yürütme | Tahmin Ajanı | Tahmin: ETA +4,5 saat; sevkiyat 2 SLA ihlali kesin, sevkiyat 3 sınırda; soğuk zincir 6 saat tolerans | — | 1.600 | 800 | | 03 | plan | Kontrol Kulesi Süpervizörü | Plan: rota/aktarma seçenekleri → maliyet → yeniden rezervasyon → müşteri → SLA kaydı | — | 240 | 220 | | 04 | araç | Rota Ajanı | 3 seçenek: yedek araçla aktarma (+2.400 TL, ETA +1,5 s), partner taşıyıcı (+3.900 TL), bekle | Carrier API | 1.200 | — | | 05 | doğrulama | Rota Ajanı | Seçenek 1 maliyet artışı %11 Simülasyon metrikleri (tahmini) 96.000çalıştırma / ay %93,4başarı 7,9 sp50 $0,0028tahmini maliyet / çalıştırma Onay kuralı Rota maliyet artışı > %15 veya tazmin > 10.000 TL → operasyon müdürü onayı Adım profili giriş · 1plan · 1akıl yürütme · 1araç · 2doğrulama · 1yazma · 2bildirim · 2 SİSTEMLER - TMSerp - Telematicsapi - Carrier APIapi - Customer Portalchannel - ERP Billingerp Çalıştırmak için gerekenler - ERP/MES için kayıt ve sorgu API erişimi - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Küresel lojistik sağlayıcıların kontrol kulesi deseni; otomatik yeniden rezervasyon ve SLA finansıyla Bu senaryonun yer aldığı paketler Operasyonel otomasyon ve talep yönetimi →Saha servis ve sevk otomasyonu →Lojistik kontrol kulesi → Sektör: Lojistik → İlgili ajanlar LojistikHITL ### Sevkiyat istisna yönetimi Gecikme ya da hasar geldiğinde etki hesaplanır, taşıyıcı yeniden planlanır, müşteri daha sormadan bilgilenir. 12.800 çalıştırma / ay %95,5 başarı 4 ajan LojistikHITL ### Gümrük belgesi hazırlama GTİP kodu, fatura ve çeki listesi dakikalarda hazırlanır; düşük güvenli satırlar gümrük uzmanına gider. 4.300 çalıştırma / ay %94,7 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Kredi kartı limit artışı URL: https://www.aginies.com/tr/ajanlar/credit-limit-increase Language: tr Summary: Müşteri mobilde yazar ya da çağrı merkezinde konuşur; karar saniyeler içinde verilir, eşik üstü yönetici onayına gider. Mobil uygulamada "limitimi artırmak istiyorum" mesajı → Yeni limit çekirdek sistemde güncellendi, müşteri bilgilendirildi, karar izi kaydedildi Bankacılıkappvoice # Kredi kartı limit artışı Müşteri mobilde yazar ya da çağrı merkezinde konuşur; karar saniyeler içinde verilir, eşik üstü yönetici onayına gider. Kısaca Kredi kartı limit artışı, Bankacılık sektöründe app, voice kanalında çalışan bir ajan senaryosudur. Tetikleyici: Mobil uygulamada "limitimi artırmak istiyorum" mesajı. Kart Süpervizörü 3 uzman ajanı (Niyet Ajanı, Risk Ajanı, İşlem Ajanı) 3 sistem üzerinde yönetir (Core Banking, KYC / Customer 360, Push & SMS Gateway); 8 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Yeni limit çekirdek sistemde güncellendi, müşteri bilgilendirildi, karar izi kaydedildi. Onay kuralı: Limit > 100.000 TL → şube yöneticisi onayı. Bankacılık ### Kredi kartı limit artışı [Kredi kartı limit artışı] çalıştırma #41.289.786süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Limit > 100.000 TL → şube yöneticisi onayı Bu çalıştırmadaki ajanlar Kart Süpervizörüsupervisorplanlar, yönlendirir, onaylar Niyet Ajanıaginies-sunkar-1 · on-premTalebi anlar, tutarı ve kartı çıkarır Risk Ajanıaginies-tulpar-4 · on-premSkor, gelir ve kullanım verisiyle politika kontrolü yapar İşlem Ajanıaginies-talay-2 · on-premOnaylı limiti çekirdek sisteme yazar Denetim izi00:00.00 - ◦Mesaj alındı, oturum kimliği doğrulandı - ∿Niyet: limit artışı · Tutar: 150.000 TL · Kart: ****4412 - ⌘Plan: risk kontrolü → politika → yazma → bildirim - ⚙Müşteri profili ve 12 aylık kullanım verisi çekildi - ✓Politika v1.3: skor 742, kullanım %38 → uygun, eşik üstü - ⏸Limit > 100.000 TL → şube yöneticisi onayına gönderildi - ⇢Çekirdek sistemde limit 150.000 TL olarak güncellendi - ➤Müşteriye push ve SMS gönderildi, karar izi kaydedildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Mobil uygulamada "limitimi artırmak istiyorum" mesajı Sonuç Yeni limit çekirdek sistemde güncellendi, müşteri bilgilendirildi, karar izi kaydedildi ## Ajanlar SÜPERVİZÖR ### Kart Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Niyet Ajanı Talebi anlar, tutarı ve kartı çıkarır detect_intentextract_entities aginies-tulpar-4 · on-prem ### Risk Ajanı Skor, gelir ve kullanım verisiyle politika kontrolü yapar get_customer_profileget_utilisationpolicy_check aginies-talay-2 · on-prem ### İşlem Ajanı Onaylı limiti çekirdek sisteme yazar update_card_limitsend_notification ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Kart Süpervizörü | Mesaj alındı, oturum kimliği doğrulandı | — | 180 | — | | 02 | akıl yürütme | Niyet Ajanı | Niyet: limit artışı · Tutar: 150.000 TL · Kart: ****4412 | — | 640 | 410 | | 03 | plan | Kart Süpervizörü | Plan: risk kontrolü → politika → yazma → bildirim | — | 320 | 220 | | 04 | araç | Risk Ajanı | Müşteri profili ve 12 aylık kullanım verisi çekildi | KYC / Customer 360 | 410 | — | | 05 | doğrulama | Risk Ajanı | Politika v1.3: skor 742, kullanım %38 → uygun, eşik üstü | — | 780 | 960 | | 06 | insan onayı | Kart Süpervizörü | Limit > 100.000 TL → şube yöneticisi onayına gönderildi | — | 2.400 | — | | 07 | yazma | İşlem Ajanı | Çekirdek sistemde limit 150.000 TL olarak güncellendi | Core Banking | 520 | — | | 08 | bildirim | İşlem Ajanı | Müşteriye push ve SMS gönderildi, karar izi kaydedildi | Push & SMS Gateway | 260 | — | Simülasyon metrikleri (tahmini) 18.400çalıştırma / ay %96,4başarı 5,2 sp50 $0,0019tahmini maliyet / çalıştırma Onay kuralı Limit > 100.000 TL → şube yöneticisi onayı Adım profili giriş · 1plan · 1akıl yürütme · 1araç · 1doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - Core Bankingcore - KYC / Customer 360db - Push & SMS Gatewaychannel Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Diyalog bankacılığı: küresel bankaların uygulama içi asistanlarının aksiyon alan sürümü Bu senaryonun yer aldığı paketler Diyalog bankacılığı →Kredi ve müşteri edinimi operasyonları → Sektör: Bankacılık → İlgili ajanlar BankacılıkHITL ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı Müşteri konuşur ya da yazar; fatura, limit, itiraz ve transfer aynı oturumda çekirdek sistemde tamamlanır. 186.000 çalıştırma / ay %94,7 başarı 5 ajan BankacılıkHITL ### Dijital müşteri edinimi (KYC) Kimlik, canlılık ve yaptırım kontrolü tek akışta; temiz başvurular şubeye gitmeden dakikalarda hesap açar. 42.000 çalıştırma / ay %93,8 başarı 5 ajan BankacılıkHITL ### Kart dolandırıcılık alarmı triyajı Kart saniyeler içinde bloklanır, müşteriye WhatsApp'tan sorulur; yanıta göre kart açılır ya da iptal edilir. 65.000 çalıştırma / ay %97,1 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Gümrük belgesi hazırlama URL: https://www.aginies.com/tr/ajanlar/customs-document-preparation Language: tr Summary: GTİP kodu, fatura ve çeki listesi dakikalarda hazırlanır; düşük güvenli satırlar gümrük uzmanına gider. WMS'te ihracat siparişinin sevke hazır olarak serbest bırakılması → Belge seti oluşturuldu, gümrük müşavirine iletildi, sevkiyat "belgeler hazır" durumunda Lojistikapiemailbatch # Gümrük belgesi hazırlama GTİP kodu, fatura ve çeki listesi dakikalarda hazırlanır; düşük güvenli satırlar gümrük uzmanına gider. Kısaca Gümrük belgesi hazırlama, Lojistik sektöründe api, email, batch kanalında çalışan bir ajan senaryosudur. Tetikleyici: WMS'te ihracat siparişinin sevke hazır olarak serbest bırakılması. Gümrük Süpervizörü 3 uzman ajanı (Tarife Ajanı, Belge Ajanı, Müşavir Ajanı) 5 sistem üzerinde yönetir (SAP S/4HANA, WMS, Tariff & Regulation Vector Store, Customs Broker Portal, Document Store (S3)); 9 adımın 3 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Belge seti oluşturuldu, gümrük müşavirine iletildi, sevkiyat "belgeler hazır" durumunda. Onay kuralı: GTİP güven skoru < 0,90 veya vergi farkı > 2.000 € → gümrük uzmanı onayı. Lojistik ### Gümrük belgesi hazırlama [Gümrük belgesi hazırlama] çalıştırma #41.290.018süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: GTİP güven skoru < 0,90 veya vergi farkı > 2.000 € → gümrük uzmanı onayı Bu çalıştırmadaki ajanlar Gümrük Süpervizörüsupervisorplanlar, yönlendirir, onaylar Tarife Ajanıaginies-tulpar-4 · on-premTarife cetveliyle GTİP kodunu belirler Belge Ajanıaginies-sunkar-1 · on-premFatura, çeki listesi ve menşe belgelerini hazırlar Müşavir Ajanıaginies-talay-2 · on-premBelgeleri müşavire iletir, sevkiyatı günceller Denetim izi00:00.00 - ◦Almanya'ya EX-20931 ihracat siparişi WMS'ten serbest bırakıldı, 22 kalem - ⚙Sipariş kalemleri, Incoterm DAP ve 148.600 € fatura değeri çekildi - ⌕Tarife fasıl 84 ve 85 ile 2026 AB açıklama notları getirildi - ∿22 kalem sınıflandırıldı; 19 yüksek güven, 3 işaretli (örn. 8479.89 / 8428.90) - ✓Menşe: 22/22 kalem AB–TR gümrük birliği kurallarını sağlıyor, A.TR uygun - ⏸3 düşük güvenli GTİP satırı → gümrük uzmanı kodları teyit etti - ⇢Ticari fatura, çeki listesi ve A.TR oluşturuldu (PDF, 9 sayfa) - ⇢Belge seti müşavire gönderildi, beyanname ön kontrol referansı BR-55120 - ⇢Sevkiyat durumu "gümrük belgeleri hazır", alım slotu Perş 14:00 teyitli - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici WMS'te ihracat siparişinin sevke hazır olarak serbest bırakılması Sonuç Belge seti oluşturuldu, gümrük müşavirine iletildi, sevkiyat "belgeler hazır" durumunda ## Ajanlar SÜPERVİZÖR ### Gümrük Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-tulpar-4 · on-prem ### Tarife Ajanı Tarife cetveliyle GTİP kodunu belirler search_tariff_scheduleclassify_hs_codeget_duty_rate aginies-sunkar-1 · on-prem ### Belge Ajanı Fatura, çeki listesi ve menşe belgelerini hazırlar get_order_linesbuild_commercial_invoicebuild_packing_listcheck_origin_rules aginies-talay-2 · on-prem ### Müşavir Ajanı Belgeleri müşavire iletir, sevkiyatı günceller submit_to_brokerupdate_shipmentsend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Gümrük Süpervizörü | Almanya'ya EX-20931 ihracat siparişi WMS'ten serbest bırakıldı, 22 kalem | — | 200 | — | | 02 | araç | Belge Ajanı | Sipariş kalemleri, Incoterm DAP ve 148.600 € fatura değeri çekildi | SAP S/4HANA | 420 | — | | 03 | bilgi | Tarife Ajanı | Tarife fasıl 84 ve 85 ile 2026 AB açıklama notları getirildi | Tariff & Regulation Vector Store | 480 | 720 | | 04 | akıl yürütme | Tarife Ajanı | 22 kalem sınıflandırıldı; 19 yüksek güven, 3 işaretli (örn. 8479.89 / 8428.90) | — | 1.100 | 1.450 | | 05 | doğrulama | Belge Ajanı | Menşe: 22/22 kalem AB–TR gümrük birliği kurallarını sağlıyor, A.TR uygun | — | 520 | 480 | | 06 | insan onayı | Gümrük Süpervizörü | 3 düşük güvenli GTİP satırı → gümrük uzmanı kodları teyit etti | — | 2.600 | — | | 07 | yazma | Belge Ajanı | Ticari fatura, çeki listesi ve A.TR oluşturuldu (PDF, 9 sayfa) | Document Store (S3) | 640 | — | | 08 | yazma | Müşavir Ajanı | Belge seti müşavire gönderildi, beyanname ön kontrol referansı BR-55120 | Customs Broker Portal | 560 | — | | 09 | yazma | Müşavir Ajanı | Sevkiyat durumu "gümrük belgeleri hazır", alım slotu Perş 14:00 teyitli | WMS | 380 | — | Simülasyon metrikleri (tahmini) 4.300çalıştırma / ay %94,7başarı 6,9 sp50 $0,0105tahmini maliyet / çalıştırma Onay kuralı GTİP güven skoru < 0,90 veya vergi farkı > 2.000 € → gümrük uzmanı onayı Adım profili giriş · 1bilgi · 1akıl yürütme · 1araç · 1doğrulama · 1insan onayı · 1yazma · 3 SİSTEMLER - SAP S/4HANAerp - WMScore - Tariff & Regulation Vector Storedb - Customs Broker Portalapi - Document Store (S3)storage Çalıştırmak için gerekenler - ERP/MES için kayıt ve sorgu API erişimi - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Doküman deposuna okuma erişimi ve belge tipleri - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Küresel forwarder ve gümrük müşavirlerinin ihracat belge otomasyonunu yansıtır Bu senaryonun yer aldığı paketler Bilgi yönetimi ve karar desteği →Lojistik kontrol kulesi → Sektör: Lojistik → İlgili ajanlar LojistikHITL ### Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA Telematik sinyali gecikmeyi erken yakalar; alternatif rota ve taşıyıcı seçilir, müşteri bilgilendirilir, SLA tazmini hesaplanır. 96.000 çalıştırma / ay %93,4 başarı 5 ajan LojistikHITL ### Sevkiyat istisna yönetimi Gecikme ya da hasar geldiğinde etki hesaplanır, taşıyıcı yeniden planlanır, müşteri daha sormadan bilgilenir. 12.800 çalıştırma / ay %95,5 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Talep yanıtı (demand response) olayı URL: https://www.aginies.com/tr/ajanlar/demand-response-event Language: tr Summary: Şebeke sıkışıklığında uygun tüketiciler seçilir, teklif kanaldan gönderilir, katılım ölçülür ve ödeme hesaplanır. Piyasa işletmecisinden ya da SCADA’dan yük azaltma çağrısı → Hedef MW için katılımcılar seçildi ve bilgilendirildi; gerçekleşen azaltım ölçüldü, ödeme hesaplandı Enerjiapiappwhatsappbatch # Talep yanıtı (demand response) olayı Şebeke sıkışıklığında uygun tüketiciler seçilir, teklif kanaldan gönderilir, katılım ölçülür ve ödeme hesaplanır. Kısaca Talep yanıtı (demand response) olayı, Enerji sektöründe api, app, whatsapp, batch kanalında çalışan bir ajan senaryosudur. Tetikleyici: Piyasa işletmecisinden ya da SCADA’dan yük azaltma çağrısı. Talep Yanıtı Süpervizörü 3 uzman ajanı (Seçim Ajanı, Teklif Ajanı, Uzlaştırma Ajanı) 4 sistem üzerinde yönetir (SCADA / DMS, Meter Data Management, Billing / Settlement, WhatsApp Business API); 9 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Hedef MW için katılımcılar seçildi ve bilgilendirildi; gerçekleşen azaltım ölçüldü, ödeme hesaplandı. Onay kuralı: Olay > 10 MW → şebeke operatörü onayı; ödeme hesabı denetim iziyle. Enerji ### Talep yanıtı (demand response) olayı [Talep yanıtı (demand response) olayı] çalıştırma #41.287.154süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Olay > 10 MW → şebeke operatörü onayı; ödeme hesabı denetim iziyle Bu çalıştırmadaki ajanlar Talep Yanıtı Süpervizörüsupervisorplanlar, yönlendirir, onaylar Seçim Ajanıaginies-tulpar-4 · on-premEsneklik potansiyeline göre tüketicileri seçer Teklif Ajanıaginies-talay-2 · on-premTeklifi kanaldan gönderir ve cevapları toplar Uzlaştırma Ajanıaginies-sunkar-1 · on-premSayaç verisiyle azaltımı ölçer ve ödemeyi hesaplar Denetim izi00:00.00 - ◦Çağrı: 17:00–19:00 arası 12 MW azaltım, bölge: Ege sanayi fiderleri - ⚙Esneklik portföyü çekildi: 214 tesis, sözleşmeli potansiyel 31 MW - ∿Tahmin ve geçmiş katılıma göre 63 tesis seçildi; beklenen azaltım 14,8 MW - ⏸Olay > 10 MW → şebeke operatörü katılımcı listesini onayladı - ➤63 tesise WhatsApp ve uygulama bildirimi: saat, hedef kW, tarife - ⚙20 dakikada 57 kabul, 4 ret, 2 cevapsız; taahhüt 13,6 MW - ⚙Olay sonrası 15 dakikalık sayaç verisi çekildi; taban çizgi hesaplandı - ✓Gerçekleşen azaltım 12,9 MW; 51 tesis taahhüdü karşıladı, 6 kısmi - ⇢Ödemeler hesaplandı ve uzlaştırmaya yazıldı; olay raporu SCADA kaydına eklendi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Piyasa işletmecisinden ya da SCADA’dan yük azaltma çağrısı Sonuç Hedef MW için katılımcılar seçildi ve bilgilendirildi; gerçekleşen azaltım ölçüldü, ödeme hesaplandı ## Ajanlar SÜPERVİZÖR ### Talep Yanıtı Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-tulpar-4 · on-prem ### Seçim Ajanı Esneklik potansiyeline göre tüketicileri seçer get_flex_portfolioforecast_reductionrank_participants aginies-talay-2 · on-prem ### Teklif Ajanı Teklifi kanaldan gönderir ve cevapları toplar send_whatsappsend_pushcollect_responses aginies-sunkar-1 · on-prem ### Uzlaştırma Ajanı Sayaç verisiyle azaltımı ölçer ve ödemeyi hesaplar get_interval_datacompute_baselinecompute_settlementpost_settlement ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Talep Yanıtı Süpervizörü | Çağrı: 17:00–19:00 arası 12 MW azaltım, bölge: Ege sanayi fiderleri | SCADA / DMS | 200 | — | | 02 | araç | Seçim Ajanı | Esneklik portföyü çekildi: 214 tesis, sözleşmeli potansiyel 31 MW | Meter Data Management | 480 | — | | 03 | akıl yürütme | Seçim Ajanı | Tahmin ve geçmiş katılıma göre 63 tesis seçildi; beklenen azaltım 14,8 MW | — | 820 | 1.100 | | 04 | insan onayı | Talep Yanıtı Süpervizörü | Olay > 10 MW → şebeke operatörü katılımcı listesini onayladı | — | 2.400 | — | | 05 | bildirim | Teklif Ajanı | 63 tesise WhatsApp ve uygulama bildirimi: saat, hedef kW, tarife | WhatsApp Business API | 300 | — | | 06 | araç | Teklif Ajanı | 20 dakikada 57 kabul, 4 ret, 2 cevapsız; taahhüt 13,6 MW | WhatsApp Business API | 420 | — | | 07 | araç | Uzlaştırma Ajanı | Olay sonrası 15 dakikalık sayaç verisi çekildi; taban çizgi hesaplandı | Meter Data Management | 520 | — | | 08 | doğrulama | Uzlaştırma Ajanı | Gerçekleşen azaltım 12,9 MW; 51 tesis taahhüdü karşıladı, 6 kısmi | — | 560 | 480 | | 09 | yazma | Uzlaştırma Ajanı | Ödemeler hesaplandı ve uzlaştırmaya yazıldı; olay raporu SCADA kaydına eklendi | Billing / Settlement | 480 | — | Simülasyon metrikleri (tahmini) 140çalıştırma / ay %96,2başarı 6,2 sp50 $0,0169tahmini maliyet / çalıştırma Onay kuralı Olay > 10 MW → şebeke operatörü onayı; ödeme hesabı denetim iziyle Adım profili giriş · 1akıl yürütme · 1araç · 3doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - SCADA / DMScore - Meter Data Managementdb - Billing / Settlementerp - WhatsApp Business APIchannel Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - ERP/MES için kayıt ve sorgu API erişimi - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Avrupa esneklik piyasalarındaki toplayıcı (aggregator) ve dağıtım şirketi talep yanıtı akışlarını yansıtır Bu senaryonun yer aldığı paketler Sektör: Enerji → İlgili ajanlar EnerjiHITL ### Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda Abone arar ya da yazar; yüksek fatura açıklanır, kesinti süresi verilir, sayaç verisi gösterilir, taksit planı kurulur. 520.000 çalıştırma / ay %94,6 başarı 5 ajan EnerjiHITL ### Kesinti bildirimi yönetimi Bildirimler fider bazında kümelenir, tek OSS kaydı açılır, etkilenen herkes ekip yola çıkmadan bilgilenir. 38.000 çalıştırma / ay %97,7 başarı 5 ajan EnerjiHITL ### Sayaç okuma itirazı Müşteri sayaç fotoğrafı yollar; endeks görüntüden okunur, tahmini fatura düzeltilir, büyük iade lidere gider. 9.200 çalıştırma / ay %94,9 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Cihaz filosu anomali triyajı URL: https://www.aginies.com/tr/ajanlar/device-fleet-anomaly-triage Language: tr Summary: Binlerce cihazdan gelen telemetri anomalileri kümelenir, kök neden aday listesi çıkar, doğru ekibe tek olay olarak düşer. IoT platformundan anomali alarmı ya da eşik ihlali olayı → Tek olay kaydı, etkilenen cihaz listesi ve önerilen aksiyon; kritik komutlar onaylı IoT ve Bağlı Cihazlarapibatchapp # Cihaz filosu anomali triyajı Binlerce cihazdan gelen telemetri anomalileri kümelenir, kök neden aday listesi çıkar, doğru ekibe tek olay olarak düşer. Kısaca Cihaz filosu anomali triyajı, IoT ve Bağlı Cihazlar sektöründe api, batch, app kanalında çalışan bir ajan senaryosudur. Tetikleyici: IoT platformundan anomali alarmı ya da eşik ihlali olayı. Filo Sağlık Süpervizörü 3 uzman ajanı (Sinyal Ajanı, Tanı Ajanı, Operasyon Ajanı) 5 sistem üzerinde yönetir (IoT Platform (MQTT / Digital Twin), Time-series DB, Known Issues (vector), ITSM / Incident, Slack); 9 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Tek olay kaydı, etkilenen cihaz listesi ve önerilen aksiyon; kritik komutlar onaylı. Onay kuralı: Toplu cihaz komutu > 100 cihaz → filo operasyon lideri onayı. IoT ve Bağlı Cihazlar ### Cihaz filosu anomali triyajı [Cihaz filosu anomali triyajı] çalıştırma #41.287.511süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Toplu cihaz komutu > 100 cihaz → filo operasyon lideri onayı Bu çalıştırmadaki ajanlar Filo Sağlık Süpervizörüsupervisorplanlar, yönlendirir, onaylar Sinyal Ajanıaginies-sunkar-1 · on-premTelemetriyi normalleştirir ve kümeler Tanı Ajanıaginies-tulpar-4 · on-premFirmware, konum ve olay geçmişiyle kök nedeni değerlendirir Operasyon Ajanıaginies-talay-2 · on-premOlay açar, komut hazırlar, ekibi bilgilendirir Denetim izi00:00.00 - ◦Alarm: 20 dakikada 312 cihazdan sıcaklık sapması, bölge: Marmara depoları - ⚙Son 6 saat telemetri çekildi; 312 cihaz 3 kümeye ayrıldı, en büyüğü 284 cihaz - ∿Ana küme: aynı sensör modeli, aynı firmware 4.2.1, aynı ağ geçidi grubu - ⚙Firmware geçmişi: 4.2.1 dün 02:00’de dağıtıldı; önceki sürümde sapma yok - ⌕Bilinen sorunlar: 4.2.1 kalibrasyon ofseti hatası (KI-0788), geçici çözüm: geri alma - ✓Kök neden güveni 0,91: firmware ofseti; fiziksel soğutma arızası dışlandı - ⇢Olay INC-40917 açıldı, öncelik P2, 284 cihaz iliştirildi, kök neden notu eklendi - ⏸Toplu geri alma komutu > 100 cihaz → filo operasyon lideri onayladı - ➤Ekibe özet: küme, kök neden, onaylı geri alma planı ve izleme penceresi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici IoT platformundan anomali alarmı ya da eşik ihlali olayı Sonuç Tek olay kaydı, etkilenen cihaz listesi ve önerilen aksiyon; kritik komutlar onaylı ## Ajanlar SÜPERVİZÖR ### Filo Sağlık Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Sinyal Ajanı Telemetriyi normalleştirir ve kümeler get_telemetrycluster_anomaliesget_device_metadata aginies-tulpar-4 · on-prem ### Tanı Ajanı Firmware, konum ve olay geçmişiyle kök nedeni değerlendirir get_firmware_historysearch_known_issuescorrelate_events aginies-talay-2 · on-prem ### Operasyon Ajanı Olay açar, komut hazırlar, ekibi bilgilendirir create_incidentprepare_device_commandsend_slack ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Filo Sağlık Süpervizörü | Alarm: 20 dakikada 312 cihazdan sıcaklık sapması, bölge: Marmara depoları | — | 200 | — | | 02 | araç | Sinyal Ajanı | Son 6 saat telemetri çekildi; 312 cihaz 3 kümeye ayrıldı, en büyüğü 284 cihaz | Time-series DB | 520 | — | | 03 | akıl yürütme | Sinyal Ajanı | Ana küme: aynı sensör modeli, aynı firmware 4.2.1, aynı ağ geçidi grubu | — | 640 | 700 | | 04 | araç | Tanı Ajanı | Firmware geçmişi: 4.2.1 dün 02:00’de dağıtıldı; önceki sürümde sapma yok | IoT Platform (MQTT / Digital Twin) | 440 | — | | 05 | bilgi | Tanı Ajanı | Bilinen sorunlar: 4.2.1 kalibrasyon ofseti hatası (KI-0788), geçici çözüm: geri alma | Known Issues (vector) | 480 | 520 | | 06 | doğrulama | Tanı Ajanı | Kök neden güveni 0,91: firmware ofseti; fiziksel soğutma arızası dışlandı | — | 560 | 480 | | 07 | yazma | Operasyon Ajanı | Olay INC-40917 açıldı, öncelik P2, 284 cihaz iliştirildi, kök neden notu eklendi | ITSM / Incident | 480 | — | | 08 | insan onayı | Filo Sağlık Süpervizörü | Toplu geri alma komutu > 100 cihaz → filo operasyon lideri onayladı | — | 2.600 | — | | 09 | bildirim | Operasyon Ajanı | Ekibe özet: küme, kök neden, onaylı geri alma planı ve izleme penceresi | Slack | 240 | — | Simülasyon metrikleri (tahmini) 18.000çalıştırma / ay %96,9başarı 6,2 sp50 $0,0019tahmini maliyet / çalıştırma Onay kuralı Toplu cihaz komutu > 100 cihaz → filo operasyon lideri onayı Adım profili giriş · 1bilgi · 1akıl yürütme · 1araç · 2doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - IoT Platform (MQTT / Digital Twin)core - Time-series DBdb - Known Issues (vector)db - ITSM / Incidentticketing - Slackchannel Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Endüstriyel IoT platformlarının AIOps tarzı alarm kümeleme ve kök neden akışlarını yansıtır Bu senaryonun yer aldığı paketler IoT cihaz filosu operasyonu → Sektör: IoT ve Bağlı Cihazlar → İlgili ajanlar IoT ve Bağlı CihazlarHITL ### Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyen Müşteri "cihaz çalışmıyor" der; telemetri okunur, uzaktan tanı konur, güncelleme gönderilir, gerekirse teknisyen doğru parçayla atanır. 37.000 çalıştırma / ay %94,1 başarı 5 ajan IoT ve Bağlı CihazlarHITL ### Kademeli OTA güncelleme dağıtımı Firmware dağıtımı %1 → %10 → %100 kademelerinde ilerler; her kademede sağlık kapısı, kritik filolarda insan onayı. 320 çalıştırma / ay %98,1 başarı 4 ajan IoT ve Bağlı CihazlarHITL ### Soğuk zincir sapması müdahalesi Sıcaklık sapması anında değerlendirilir, ürün riski hesaplanır, saha ekibi sevk edilir, kalite kararı kayıt altına alınır. 2.700 çalıştırma / ay %97,4 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Diyalog bankacılığı: sesli ve yazılı işlem asistanı URL: https://www.aginies.com/tr/ajanlar/dialog-banking-assistant Language: tr Summary: Müşteri konuşur ya da yazar; fatura, limit, itiraz ve transfer aynı oturumda çekirdek sistemde tamamlanır. Mobil uygulama sohbeti, WhatsApp ya da çağrı merkezi üzerinden gelen çok işlemli talep → Üç işlem gerçekleşti, her biri ayrı referansla kayıtlı; müşteri tek özetle bilgilendirildi Bankacılıkappvoicewhatsapp # Diyalog bankacılığı: sesli ve yazılı işlem asistanı Müşteri konuşur ya da yazar; fatura, limit, itiraz ve transfer aynı oturumda çekirdek sistemde tamamlanır. Kısaca Diyalog bankacılığı: sesli ve yazılı işlem asistanı, Bankacılık sektöründe app, voice, whatsapp kanalında çalışan bir ajan senaryosudur. Tetikleyici: Mobil uygulama sohbeti, WhatsApp ya da çağrı merkezi üzerinden gelen çok işlemli talep. Diyalog Bankacılığı Süpervizörü 4 uzman ajanı (Diyalog Ajanı, Kimlik Ajanı, İşlem Ajanı, Risk Ajanı) 5 sistem üzerinde yönetir (Mobile App Chat, Core Banking, Card Management, Fraud Engine, SMS Gateway); 11 adımın 3 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Üç işlem gerçekleşti, her biri ayrı referansla kayıtlı; müşteri tek özetle bilgilendirildi. Onay kuralı: Limit artışı > %60, itiraz tutarı > 5.000 TL veya yeni alıcıya transfer → operasyon onayı. Bankacılık ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı [Diyalog bankacılığı: sesli ve yazılı işlem asistanı] çalıştırma #41.289.133süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Limit artışı > %60, itiraz tutarı > 5.000 TL veya yeni alıcıya transfer → operasyon onayı Bu çalıştırmadaki ajanlar Diyalog Bankacılığı Süpervizörüsupervisorplanlar, yönlendirir, onaylar Diyalog Ajanıaginies-talay-2 · on-premÇok turlu konuşmayı yürütür, birden fazla niyeti ayrıştırır, sesli ve yazılı yanıt üretir Kimlik Ajanıaginies-sunkar-1 · on-premOturumu, biyometrik durumu ve işlem bazlı OTP eşiklerini doğrular İşlem Ajanıaginies-tulpar-4 · on-premFatura, limit, transfer ve kart işlemlerini çekirdek sistem araçlarıyla yapar Risk Ajanıaginies-sunkar-1 · on-premDolandırıcılık skoru, politika eşikleri ve onay kurallarını uygular Denetim izi00:00.00 - ◦Uygulama sohbeti: "Elektrik faturamı öde, limitimi 40 bine çıkar, dünkü 1.250 TL'yi tanımıyorum" - ∿Üç niyet ayrıştırıldı: fatura ödeme · limit artışı · işlem itirazı; sıra ve bağımlılık belirlendi - ✓Oturum biyometrik doğrulamalı, cihaz kayıtlı; 10.000 TL üstü için OTP zorunlu - ⌘Plan: fatura sorgula → öde → limit skoru → politika → itiraz kaydı → tek özet - ⚙Elektrik faturası: 842,60 TL, son ödeme 3 gün sonra, otomatik talimat yok - ⇢Fatura vadesiz hesaptan ödendi, ref #BL-2048816; dekont sohbete eklendi - ✓Limit 25.000 → 40.000: artış %60, skor 742, gelir doğrulandı → otomatik onay sınırında - ⇢Kart limiti 40.000 TL olarak güncellendi, ref #CL-77310; anında geçerli - ⚙İtiraz edilen işlem: yurt dışı e-ticaret, skor 81/100; kart geçici bloke, yeni kart önerildi - ⇢İtiraz dosyası #DS-10944 açıldı, kart bloke edildi, yeni kart siparişi verildi - ➤Tek özet: üç işlem ve referanslar; sesli kanalda okundu, SMS ve bildirim gönderildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Mobil uygulama sohbeti, WhatsApp ya da çağrı merkezi üzerinden gelen çok işlemli talep Sonuç Üç işlem gerçekleşti, her biri ayrı referansla kayıtlı; müşteri tek özetle bilgilendirildi ## Ajanlar SÜPERVİZÖR ### Diyalog Bankacılığı Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Diyalog Ajanı Çok turlu konuşmayı yürütür, birden fazla niyeti ayrıştırır, sesli ve yazılı yanıt üretir detect_intentscompose_replytts_speak aginies-sunkar-1 · on-prem ### Kimlik Ajanı Oturumu, biyometrik durumu ve işlem bazlı OTP eşiklerini doğrular check_sessionvoiceprint_matchsend_otpverify_otp aginies-tulpar-4 · on-prem ### İşlem Ajanı Fatura, limit, transfer ve kart işlemlerini çekirdek sistem araçlarıyla yapar query_billpay_billget_card_limitupdate_card_limitopen_disputeblock_card aginies-sunkar-1 · on-prem ### Risk Ajanı Dolandırıcılık skoru, politika eşikleri ve onay kurallarını uygular fraud_scorepolicy_checkroute_approval ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Diyalog Bankacılığı Süpervizörü | Uygulama sohbeti: "Elektrik faturamı öde, limitimi 40 bine çıkar, dünkü 1.250 TL'yi tanımıyorum" | Mobile App Chat | 180 | — | | 02 | akıl yürütme | Diyalog Ajanı | Üç niyet ayrıştırıldı: fatura ödeme · limit artışı · işlem itirazı; sıra ve bağımlılık belirlendi | — | 620 | 410 | | 03 | doğrulama | Kimlik Ajanı | Oturum biyometrik doğrulamalı, cihaz kayıtlı; 10.000 TL üstü için OTP zorunlu | — | 340 | 120 | | 04 | plan | Diyalog Bankacılığı Süpervizörü | Plan: fatura sorgula → öde → limit skoru → politika → itiraz kaydı → tek özet | — | 260 | 240 | | 05 | araç | İşlem Ajanı | Elektrik faturası: 842,60 TL, son ödeme 3 gün sonra, otomatik talimat yok | Core Banking | 420 | — | | 06 | yazma | İşlem Ajanı | Fatura vadesiz hesaptan ödendi, ref #BL-2048816; dekont sohbete eklendi | Core Banking | 560 | — | | 07 | doğrulama | Risk Ajanı | Limit 25.000 → 40.000: artış %60, skor 742, gelir doğrulandı → otomatik onay sınırında | Card Management | 700 | 330 | | 08 | yazma | İşlem Ajanı | Kart limiti 40.000 TL olarak güncellendi, ref #CL-77310; anında geçerli | Card Management | 480 | — | | 09 | araç | Risk Ajanı | İtiraz edilen işlem: yurt dışı e-ticaret, skor 81/100; kart geçici bloke, yeni kart önerildi | Fraud Engine | 640 | 280 | | 10 | yazma | İşlem Ajanı | İtiraz dosyası #DS-10944 açıldı, kart bloke edildi, yeni kart siparişi verildi | Card Management | 520 | — | | 11 | bildirim | Diyalog Ajanı | Tek özet: üç işlem ve referanslar; sesli kanalda okundu, SMS ve bildirim gönderildi | SMS Gateway | 460 | 220 | Simülasyon metrikleri (tahmini) 186.000çalıştırma / ay %94,7başarı 7,8 sp50 $0,0037tahmini maliyet / çalıştırma Onay kuralı Limit artışı > %60, itiraz tutarı > 5.000 TL veya yeni alıcıya transfer → operasyon onayı Adım profili giriş · 1plan · 1akıl yürütme · 1araç · 2doğrulama · 2yazma · 3bildirim · 1 SİSTEMLER - Mobile App Chatchannel - Core Bankingcore - Card Managementapi - Fraud Engineapi - SMS Gatewaychannel Çalıştırmak için gerekenler - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük bankaların sohbet ve sesli asistan deseninin işlem yapan sürümü; birden fazla işlemi tek oturumda bitirir Bu senaryonun yer aldığı paketler Sesli AI ile hasar ve talep kaydı →Diyalog bankacılığı →Ses analizi ve kalite →Dolandırıcılık ve itiraz operasyonları → Sektör: Bankacılık → İlgili ajanlar BankacılıkHITL ### Kredi kartı limit artışı Müşteri mobilde yazar ya da çağrı merkezinde konuşur; karar saniyeler içinde verilir, eşik üstü yönetici onayına gider. 18.400 çalıştırma / ay %96,4 başarı 4 ajan BankacılıkHITL ### Dijital müşteri edinimi (KYC) Kimlik, canlılık ve yaptırım kontrolü tek akışta; temiz başvurular şubeye gitmeden dakikalarda hesap açar. 42.000 çalıştırma / ay %93,8 başarı 5 ajan BankacılıkHITL ### Kart dolandırıcılık alarmı triyajı Kart saniyeler içinde bloklanır, müşteriye WhatsApp'tan sorulur; yanıta göre kart açılır ya da iptal edilir. 65.000 çalıştırma / ay %97,1 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # E-posta sınıflandırma ve ek kontrolü URL: https://www.aginies.com/tr/ajanlar/email-classification-attachment-check Language: tr Summary: Ortak kutuya gelen her e-posta sınıflandırılır, ekler listeyle karşılaştırılır, eksikler anında sorulur. operations@ ortak posta kutusuna yeni e-posta → Talep doğru kuyrukta, ekler doğrulanıp arşivlendi, eksikler gönderene soruldu Kurumsal Fonksiyonlaremail # E-posta sınıflandırma ve ek kontrolü Ortak kutuya gelen her e-posta sınıflandırılır, ekler listeyle karşılaştırılır, eksikler anında sorulur. Kısaca E-posta sınıflandırma ve ek kontrolü, Kurumsal Fonksiyonlar sektöründe email kanalında çalışan bir ajan senaryosudur. Tetikleyici: operations@ ortak posta kutusuna yeni e-posta. Posta Süpervizörü 3 uzman ajanı (Sınıflandırma Ajanı, Ek Kontrol Ajanı, Yönlendirme Ajanı) 4 sistem üzerinde yönetir (Exchange / M365 Mail, Mailbox Rules & Checklists Vector Store, ServiceNow, Document Store (S3)); 8 adımın 2 tanesi sistemde işlem, 1 tanesi doğrulama veya onaydır. Sonuç: Talep doğru kuyrukta, ekler doğrulanıp arşivlendi, eksikler gönderene soruldu. Kurumsal Fonksiyonlar ### E-posta sınıflandırma ve ek kontrolü [E-posta sınıflandırma ve ek kontrolü] çalıştırma #41.290.368süre 0 mstoken 0tahmini maliyet $0,000durum — Bu çalıştırmadaki ajanlar Posta Süpervizörüsupervisorplanlar, yönlendirir, onaylar Sınıflandırma Ajanıaginies-sunkar-1 · on-premE-postayı konu ve dile göre sınıflandırır Ek Kontrol Ajanıaginies-konrul-1 · on-premEkleri okur, türünü ve geçerliliğini kontrol eder Yönlendirme Ajanıaginies-talay-2 · on-premKayıt açar, gönderene yanıt verir Denetim izi00:00.00 - ◦operations@ kutusuna 2 ekli (PDF, XLSX) yeni e-posta düştü - ∿Sınıf: "tedarikçi kayıt talebi" · güven 0,94 · dil TR - ⌕Kontrol listesi "tedarikçi kaydı" v4: vergi levhası, imza sirküleri, IBAN yazısı - ∿PDF = vergi levhası (2025, süresi dolmuş); XLSX = banka formu; imza sirküleri yok - ✓Sonuç: 1 eksik, 1 süresi dolmuş → ilerlenemez; otomatik yanıt politika gereği serbest - ⇢SN-INC-771230 kaydı "gönderen bekleniyor" durumunda açıldı, SLA duraklatıldı - ⇢Ekler kayıt altında arşivlendi, özet değerleri (hash) kaydedildi - ➤Gönderene 2 eksik/süresi dolmuş belge için yanıt verildi, e-posta "Beklemede"ye taşındı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici operations@ ortak posta kutusuna yeni e-posta Sonuç Talep doğru kuyrukta, ekler doğrulanıp arşivlendi, eksikler gönderene soruldu ## Ajanlar SÜPERVİZÖR ### Posta Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Sınıflandırma Ajanı E-postayı konu ve dile göre sınıflandırır classify_emailextract_entities aginies-konrul-1 · on-prem ### Ek Kontrol Ajanı Ekleri okur, türünü ve geçerliliğini kontrol eder parse_attachmentdetect_document_typesearch_checklistscheck_required_attachments aginies-talay-2 · on-prem ### Yönlendirme Ajanı Kayıt açar, gönderene yanıt verir create_ticketarchive_attachmentsreply_emailmove_to_folder ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Posta Süpervizörü | operations@ kutusuna 2 ekli (PDF, XLSX) yeni e-posta düştü | — | 180 | — | | 02 | akıl yürütme | Sınıflandırma Ajanı | Sınıf: "tedarikçi kayıt talebi" · güven 0,94 · dil TR | — | 520 | 430 | | 03 | bilgi | Ek Kontrol Ajanı | Kontrol listesi "tedarikçi kaydı" v4: vergi levhası, imza sirküleri, IBAN yazısı | Mailbox Rules & Checklists Vector Store | 340 | 280 | | 04 | akıl yürütme | Ek Kontrol Ajanı | PDF = vergi levhası (2025, süresi dolmuş); XLSX = banka formu; imza sirküleri yok | — | 880 | 720 | | 05 | doğrulama | Ek Kontrol Ajanı | Sonuç: 1 eksik, 1 süresi dolmuş → ilerlenemez; otomatik yanıt politika gereği serbest | — | 300 | 200 | | 06 | yazma | Yönlendirme Ajanı | SN-INC-771230 kaydı "gönderen bekleniyor" durumunda açıldı, SLA duraklatıldı | ServiceNow | 480 | — | | 07 | yazma | Yönlendirme Ajanı | Ekler kayıt altında arşivlendi, özet değerleri (hash) kaydedildi | Document Store (S3) | 260 | — | | 08 | bildirim | Yönlendirme Ajanı | Gönderene 2 eksik/süresi dolmuş belge için yanıt verildi, e-posta "Beklemede"ye taşındı | Exchange / M365 Mail | 240 | — | Simülasyon metrikleri (tahmini) 86.000çalıştırma / ay %97,6başarı 3,2 sp50 $0,0006tahmini maliyet / çalıştırma Adım profili giriş · 1bilgi · 1akıl yürütme · 2doğrulama · 1yazma · 2bildirim · 1 SİSTEMLER - Exchange / M365 Mailemail - Mailbox Rules & Checklists Vector Storedb - ServiceNowticketing - Document Store (S3)storage Çalıştırmak için gerekenler - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Doküman deposuna okuma erişimi ve belge tipleri - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Sigorta şirketleri ve ortak hizmet merkezlerindeki akıllı posta odası otomasyonunu yansıtır Bu senaryonun yer aldığı paketler E-posta sınıflandırma ve ek doğrulama →İK ve çalışan deneyimi →KVKK kişisel veri koruma → İlgili ajanlar Kurumsal FonksiyonlarHITL ### Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada Talep e-postayla gelir; teklifler karşılaştırılır, sözleşme maddeleri kontrol edilir, sipariş açılır, fatura üç yönlü eşleşir ve ödeme planlanır. 14.000 çalıştırma / ay %92,7 başarı 5 ajan Kurumsal FonksiyonlarHITL ### Çağrı merkezi kalite izleme Örneklem değil, tüm çağrılar rubriğe göre puanlanır; uyum ihlali ve koçluk fırsatı sabah liderin önünde. 90.000 çalıştırma / ay %98,8 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Sözleşme madde çıkarımı Sözleşme maddeleri çıkarılır, hukuk playbook'u ile karşılaştırılır; sapmalar redline önerisiyle avukata gider. 1.400 çalıştırma / ay %95,1 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Çalışan işe alıştırma (onboarding) URL: https://www.aginies.com/tr/ajanlar/employee-onboarding Language: tr Summary: İK kaydı açıldığı anda hesaplar, ekipman ve eğitim planı hazırlanır; şablon dışı erişimler yöneticiye sorulur. SuccessFactors'ta yeni işe alım kaydının onaylanması → Hesaplar açıldı, ekipman sipariş edildi, eğitim planı atandı, hoş geldin e-postası gitti Kurumsal Fonksiyonlarapiemailapp # Çalışan işe alıştırma (onboarding) İK kaydı açıldığı anda hesaplar, ekipman ve eğitim planı hazırlanır; şablon dışı erişimler yöneticiye sorulur. Kısaca Çalışan işe alıştırma (onboarding), Kurumsal Fonksiyonlar sektöründe api, email, app kanalında çalışan bir ajan senaryosudur. Tetikleyici: SuccessFactors'ta yeni işe alım kaydının onaylanması. İşe Alıştırma Süpervizörü 4 uzman ajanı (İK Ajanı, Kimlik Ajanı, BT Ajanı, Eğitim Ajanı) 5 sistem üzerinde yönetir (SAP SuccessFactors, Active Directory / Entra ID, ServiceNow, LMS, Exchange / M365 Mail); 9 adımın 3 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Hesaplar açıldı, ekipman sipariş edildi, eğitim planı atandı, hoş geldin e-postası gitti. Onay kuralı: Şablon dışı erişim veya yönetici yetkisi → bölüm yöneticisi onayı. Kurumsal Fonksiyonlar ### Çalışan işe alıştırma (onboarding) [Çalışan işe alıştırma (onboarding)] çalıştırma #41.286.759süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Şablon dışı erişim veya yönetici yetkisi → bölüm yöneticisi onayı Bu çalıştırmadaki ajanlar İşe Alıştırma Süpervizörüsupervisorplanlar, yönlendirir, onaylar İK Ajanıaginies-sunkar-1 · on-premİşe alım kaydını ve rol profilini okur Kimlik Ajanıaginies-talay-2 · on-premHesap, grup ve posta kutusunu oluşturur BT Ajanıaginies-sunkar-1 · on-premEkipmanı stoktan sipariş eder Eğitim Ajanıaginies-tulpar-4 · on-premRole göre eğitim planı kurar, hoş geldin iletişimini yapar Denetim izi00:00.00 - ◦SuccessFactors olayı: yeni çalışan E-40718, Veri Analisti, başlangıç 15 Eylül, İstanbul - ⚙Rol profili: masraf yeri 4410, yönetici M-2210, erişim şablonu "Analytics-L2" - ⌘Plan: kimlik → ekipman → eğitim → hoş geldin (mümkün olanlar paralel) - ✓Görev ayrılığı: "Analytics-L2" + finans okuma → çakışma yok; MFA zorunlu - ⇢Hesap açıldı, 6 grup atandı, posta kutusu oluşturuldu, MFA kayıt bağlantısı üretildi - ⇢Dizüstü (14", 32 GB) stoktan sipariş edildi, 12 Eylül İstanbul ofisine teslim - ⏸Şablon dışı "finans okuma" erişimi → bölüm yöneticisi onayladı - ⇢Eğitim planı: 6 kurs (KVKK, güvenlik, SQL temelleri), 30 gün içinde tamamlanacak - ➤Çalışana hoş geldin, yöneticiye ve buddy'ye kontrol listesi, 1. gün ajandası eklendi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici SuccessFactors'ta yeni işe alım kaydının onaylanması Sonuç Hesaplar açıldı, ekipman sipariş edildi, eğitim planı atandı, hoş geldin e-postası gitti ## Ajanlar SÜPERVİZÖR ### İşe Alıştırma Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### İK Ajanı İşe alım kaydını ve rol profilini okur get_new_hireget_role_profile aginies-talay-2 · on-prem ### Kimlik Ajanı Hesap, grup ve posta kutusunu oluşturur check_sod_rulescreate_userassign_groupscreate_mailbox aginies-sunkar-1 · on-prem ### BT Ajanı Ekipmanı stoktan sipariş eder check_stockorder_equipmentcreate_ticket aginies-tulpar-4 · on-prem ### Eğitim Ajanı Role göre eğitim planı kurar, hoş geldin iletişimini yapar build_training_planenrol_coursessend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | İşe Alıştırma Süpervizörü | SuccessFactors olayı: yeni çalışan E-40718, Veri Analisti, başlangıç 15 Eylül, İstanbul | — | 200 | — | | 02 | araç | İK Ajanı | Rol profili: masraf yeri 4410, yönetici M-2210, erişim şablonu "Analytics-L2" | SAP SuccessFactors | 420 | — | | 03 | plan | İşe Alıştırma Süpervizörü | Plan: kimlik → ekipman → eğitim → hoş geldin (mümkün olanlar paralel) | — | 280 | 200 | | 04 | doğrulama | Kimlik Ajanı | Görev ayrılığı: "Analytics-L2" + finans okuma → çakışma yok; MFA zorunlu | — | 360 | 240 | | 05 | yazma | Kimlik Ajanı | Hesap açıldı, 6 grup atandı, posta kutusu oluşturuldu, MFA kayıt bağlantısı üretildi | Active Directory / Entra ID | 620 | — | | 06 | yazma | BT Ajanı | Dizüstü (14", 32 GB) stoktan sipariş edildi, 12 Eylül İstanbul ofisine teslim | ServiceNow | 480 | — | | 07 | insan onayı | İşe Alıştırma Süpervizörü | Şablon dışı "finans okuma" erişimi → bölüm yöneticisi onayladı | — | 2.400 | — | | 08 | yazma | Eğitim Ajanı | Eğitim planı: 6 kurs (KVKK, güvenlik, SQL temelleri), 30 gün içinde tamamlanacak | LMS | 440 | — | | 09 | bildirim | Eğitim Ajanı | Çalışana hoş geldin, yöneticiye ve buddy'ye kontrol listesi, 1. gün ajandası eklendi | Exchange / M365 Mail | 260 | — | Simülasyon metrikleri (tahmini) 1.900çalıştırma / ay %97,2başarı 5,5 sp50 $0,0028tahmini maliyet / çalıştırma Onay kuralı Şablon dışı erişim veya yönetici yetkisi → bölüm yöneticisi onayı Adım profili giriş · 1plan · 1araç · 1doğrulama · 1insan onayı · 1yazma · 3bildirim · 1 SİSTEMLER - SAP SuccessFactorshr - Active Directory / Entra IDcore - ServiceNowticketing - LMShr - Exchange / M365 Mailemail Çalıştırmak için gerekenler - İK sistemi API erişimi ve kişisel veri sınırları - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük kurumlardaki sıfır dokunuşlu işe giriş / rol değişimi / ayrılış otomasyonunu yansıtır Bu senaryonun yer aldığı paketler İK ve çalışan deneyimi → İlgili ajanlar Kurumsal FonksiyonlarHITL ### Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada Talep e-postayla gelir; teklifler karşılaştırılır, sözleşme maddeleri kontrol edilir, sipariş açılır, fatura üç yönlü eşleşir ve ödeme planlanır. 14.000 çalıştırma / ay %92,7 başarı 5 ajan Kurumsal Fonksiyonlar ### E-posta sınıflandırma ve ek kontrolü Ortak kutuya gelen her e-posta sınıflandırılır, ekler listeyle karşılaştırılır, eksikler anında sorulur. 86.000 çalıştırma / ay %97,6 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Çağrı merkezi kalite izleme Örneklem değil, tüm çağrılar rubriğe göre puanlanır; uyum ihlali ve koçluk fırsatı sabah liderin önünde. 90.000 çalıştırma / ay %98,8 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # İhracat kontrolü ve son kullanıcı taraması URL: https://www.aginies.com/tr/ajanlar/export-control-screening Language: tr Summary: Parça, alıcı ve son kullanım tek akışta sınıflandırılır; lisans gereği tespit edilir, karar her zaman uyum yöneticisinde kalır. ERP’de yeni ihracat siparişi ya da teklif talebi → Kontrol sınıfı, taraf taraması ve lisans ihtiyacı belgelendi; sipariş uyum kararıyla açıldı ya da bekletildi Savunma Sanayiiapiemailbatch # İhracat kontrolü ve son kullanıcı taraması Parça, alıcı ve son kullanım tek akışta sınıflandırılır; lisans gereği tespit edilir, karar her zaman uyum yöneticisinde kalır. Kısaca İhracat kontrolü ve son kullanıcı taraması, Savunma Sanayii sektöründe api, email, batch kanalında çalışan bir ajan senaryosudur. Tetikleyici: ERP’de yeni ihracat siparişi ya da teklif talebi. İhracat Uyum Süpervizörü 3 uzman ajanı (Sınıflandırma Ajanı, Taraf Tarama Ajanı, Lisans Ajanı) 4 sistem üzerinde yönetir (ERP (SAP GTS), Control & Sanctions Lists (vector), Trade Compliance Case System, E-mail); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Kontrol sınıfı, taraf taraması ve lisans ihtiyacı belgelendi; sipariş uyum kararıyla açıldı ya da bekletildi. Onay kuralı: Kontrole tabi her kalem → ihracat uyum yöneticisi onayı; ajan lisans kararı veremez. Savunma Sanayii ### İhracat kontrolü ve son kullanıcı taraması [İhracat kontrolü ve son kullanıcı taraması] çalıştırma #41.289.683süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Kontrole tabi her kalem → ihracat uyum yöneticisi onayı; ajan lisans kararı veremez Bu çalıştırmadaki ajanlar İhracat Uyum Süpervizörüsupervisorplanlar, yönlendirir, onaylar Sınıflandırma Ajanıaginies-tulpar-4 · on-premParçayı kontrol listeleriyle eşler Taraf Tarama Ajanıaginies-talay-2 · on-premAlıcı, aracı ve son kullanıcıyı yaptırım listeleriyle tarar Lisans Ajanıaginies-tulpar-4 · on-premLisans gereğini belirler ve dosyayı hazırlar Denetim izi00:00.00 - ◦Sipariş SO-88120: 40 adet RF modülü, alıcı ülke dışı bir sistem entegratörü - ⚙Malzeme kartı çekildi: frekans aralığı, çıkış gücü, teknik veri sayfası - ⌕Kontrol listeleri tarandı: çift kullanımlı kategori 5A001 adayı, ulusal liste eki B - ∿Parametreler eşiğin üstünde → kontrole tabi; sınıflandırma gerekçesi ve kaynak maddeler yazıldı - ⚙Alıcı, nihai kullanıcı ve nakliyeci tarandı: 0 eşleşme; sahiplik zinciri 3 seviye çözüldü - ✓Son kullanım beyanı mevcut, imzalı; belirtilen kullanım sınıflandırmayla tutarlı - ∿Lisans gerekli: bireysel ihracat lisansı; başvuru taslağı ve ek listesi hazırlandı - ⏸Kontrole tabi kalem → ihracat uyum yöneticisi inceledi, başvuruyu onayladı - ⇢Uyum dosyası TC-2026-0419 açıldı; sipariş “lisans bekliyor” durumuna alındı - ➤Satış ve lojistiğe bildirim: sevkiyat lisans gelene kadar bloke, tahmini süre eklendi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici ERP’de yeni ihracat siparişi ya da teklif talebi Sonuç Kontrol sınıfı, taraf taraması ve lisans ihtiyacı belgelendi; sipariş uyum kararıyla açıldı ya da bekletildi ## Ajanlar SÜPERVİZÖR ### İhracat Uyum Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-tulpar-4 · on-prem ### Sınıflandırma Ajanı Parçayı kontrol listeleriyle eşler get_item_mastersearch_control_listsclassify_dual_use aginies-talay-2 · on-prem ### Taraf Tarama Ajanı Alıcı, aracı ve son kullanıcıyı yaptırım listeleriyle tarar screen_partyresolve_ownershipcheck_end_use_statement aginies-tulpar-4 · on-prem ### Lisans Ajanı Lisans gereğini belirler ve dosyayı hazırlar search_regulationdraft_licence_applicationcreate_compliance_case ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | İhracat Uyum Süpervizörü | Sipariş SO-88120: 40 adet RF modülü, alıcı ülke dışı bir sistem entegratörü | — | 200 | — | | 02 | araç | Sınıflandırma Ajanı | Malzeme kartı çekildi: frekans aralığı, çıkış gücü, teknik veri sayfası | ERP (SAP GTS) | 420 | — | | 03 | bilgi | Sınıflandırma Ajanı | Kontrol listeleri tarandı: çift kullanımlı kategori 5A001 adayı, ulusal liste eki B | Control & Sanctions Lists (vector) | 640 | 900 | | 04 | akıl yürütme | Sınıflandırma Ajanı | Parametreler eşiğin üstünde → kontrole tabi; sınıflandırma gerekçesi ve kaynak maddeler yazıldı | — | 760 | 1.100 | | 05 | araç | Taraf Tarama Ajanı | Alıcı, nihai kullanıcı ve nakliyeci tarandı: 0 eşleşme; sahiplik zinciri 3 seviye çözüldü | Control & Sanctions Lists (vector) | 520 | — | | 06 | doğrulama | Taraf Tarama Ajanı | Son kullanım beyanı mevcut, imzalı; belirtilen kullanım sınıflandırmayla tutarlı | — | 480 | 420 | | 07 | akıl yürütme | Lisans Ajanı | Lisans gerekli: bireysel ihracat lisansı; başvuru taslağı ve ek listesi hazırlandı | — | 820 | 1.300 | | 08 | insan onayı | İhracat Uyum Süpervizörü | Kontrole tabi kalem → ihracat uyum yöneticisi inceledi, başvuruyu onayladı | — | 3.000 | — | | 09 | yazma | Lisans Ajanı | Uyum dosyası TC-2026-0419 açıldı; sipariş “lisans bekliyor” durumuna alındı | Trade Compliance Case System | 480 | — | | 10 | bildirim | Lisans Ajanı | Satış ve lojistiğe bildirim: sevkiyat lisans gelene kadar bloke, tahmini süre eklendi | E-mail | 240 | — | Simülasyon metrikleri (tahmini) 2.600çalıştırma / ay %96,1başarı 7,6 sp50 $0,0105tahmini maliyet / çalıştırma Onay kuralı Kontrole tabi her kalem → ihracat uyum yöneticisi onayı; ajan lisans kararı veremez Adım profili giriş · 1bilgi · 1akıl yürütme · 2araç · 2doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - ERP (SAP GTS)erp - Control & Sanctions Lists (vector)db - Trade Compliance Case Systemticketing - E-mailemail Çalıştırmak için gerekenler - ERP/MES için kayıt ve sorgu API erişimi - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Avrupa ve ABD savunma tedarikçilerinin otomatik sınıflandırma + insan kararı temelli ihracat uyum akışlarını yansıtır Bu senaryonun yer aldığı paketler Savunma ihracat kontrolü ve tedarik zinciri güvenliği → Sektör: Savunma Sanayii → İlgili ajanlar Savunma SanayiiHITL ### Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolü Teknisyen hangarda sorar; doğru revizyondaki prosedür, parça numarası ve emniyet notu yetki kontrolüyle gelir, her erişim iz bırakır. 9.800 çalıştırma / ay %95,7 başarı 5 ajan Savunma SanayiiHITL ### Platform bakım (MRO) iş emri hazırlığı Arıza raporu teknik yayınlarla eşlenir, görev kartı ve parça listesi çıkar, iş emri planlayıcı onayıyla açılır. 4.100 çalıştırma / ay %95,7 başarı 5 ajan Savunma SanayiiHITL ### Tedarikçi güvenlik ve uyum değerlendirmesi Tedarikçi anketleri ve belgeleri okunur, tesis güvenlik gereklilikleriyle karşılaştırılır, eksikler listelenir; nitelendirme kararı güvenlik müdürünün. 900 çalıştırma / ay %95,2 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Özellik teslimat hattı URL: https://www.aginies.com/tr/ajanlar/feature-delivery-pipeline Language: tr Summary: Bir Jira epic'i ajanlarla kriter, tasarım, kod ve teste dönüşür; insan yalnızca PR'ı inceler ve onaylar. Jira'da epic "Hazır" sütununa taşındı (webhook) → PR insan onayıyla main'e birleşti, testler yeşil, Jira ve Confluence güncel Yazılım Geliştirmeapiweb # Özellik teslimat hattı Bir Jira epic'i ajanlarla kriter, tasarım, kod ve teste dönüşür; insan yalnızca PR'ı inceler ve onaylar. Kısaca Özellik teslimat hattı, Yazılım Geliştirme sektöründe api, web kanalında çalışan bir ajan senaryosudur. Tetikleyici: Jira'da epic "Hazır" sütununa taşındı (webhook). Teslimat Süpervizörü 4 uzman ajanı (Analist Ajanı, Mimar Ajanı, Geliştirici Ajanı, Test Ajanı) 4 sistem üzerinde yönetir (Jira, GitHub, CI Runner (GitHub Actions), Confluence); 10 adımın 3 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: PR insan onayıyla main'e birleşti, testler yeşil, Jira ve Confluence güncel. Onay kuralı: Her PR → en az 1 insan kod incelemesi onayı; ajan main'e doğrudan yazamaz. Yazılım Geliştirme ### Özellik teslimat hattı [Özellik teslimat hattı] çalıştırma #41.287.697süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Her PR → en az 1 insan kod incelemesi onayı; ajan main'e doğrudan yazamaz Bu çalıştırmadaki ajanlar Teslimat Süpervizörüsupervisorplanlar, yönlendirir, onaylar Analist Ajanıaginies-tulpar-4 · on-premEpic'i kullanıcı hikâyelerine ve kabul kriterlerine ayırır Mimar Ajanıaginies-tulpar-4 · on-premTeknik tasarımı yazar, etkilenen servisleri ve API sözleşmesini belirler Geliştirici Ajanıaginies-tulpar-4 · on-premDal açar, kodu yazar, derler, PR oluşturur Test Ajanıaginies-sunkar-1 · on-premKabul kriterlerinden test yazar, koşturur, kapsamı raporlar Denetim izi00:00.00 - ◦Epic PROJ-4821 "Fatura PDF dışa aktarma" Hazır sütununa taşındı, 3 alt görev - ⌕Epic açıklaması, yorumlar ve bağlı 2 tasarım dosyası okundu - ∿3 kullanıcı hikâyesi ve 11 kabul kriteri yazıldı (Gherkin), PROJ-4821 güncellendi - ⌘Plan: mimari tasarım → dal + kod → testler → PR → insan incelemesi → birleştirme - ⇢Tasarım notu yayımlandı: billing-service + yeni /invoices/{id}/pdf uç noktası - ⇢Dal feature/PROJ-4821 açıldı, 6 dosya değişti (+412 / −38), derleme yeşil - ⚙CI: 14 birim + 4 entegrasyon testi yazıldı ve koşuldu, 18/18 geçti, kapsam %87 - ✓Kabul kriterleri 11/11 karşılandı, kapsam %80 eşiğinin üstünde → PR #2291 hazır - ⏸PR #2291 → kıdemli geliştirici inceledi, 2 yorum düzeltildi, onaylandı - ⇢PR #2291 main'e squash-merge edildi; PROJ-4821 "Bitti" durumuna alındı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Jira'da epic "Hazır" sütununa taşındı (webhook) Sonuç PR insan onayıyla main'e birleşti, testler yeşil, Jira ve Confluence güncel ## Ajanlar SÜPERVİZÖR ### Teslimat Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-tulpar-4 · on-prem ### Analist Ajanı Epic'i kullanıcı hikâyelerine ve kabul kriterlerine ayırır read_jira_issuewrite_acceptance_criteriaupdate_jira_issue aginies-tulpar-4 · on-prem ### Mimar Ajanı Teknik tasarımı yazar, etkilenen servisleri ve API sözleşmesini belirler search_codebasedraft_design_docpublish_confluence_page aginies-tulpar-4 · on-prem ### Geliştirici Ajanı Dal açar, kodu yazar, derler, PR oluşturur create_branchedit_filesrun_buildopen_pull_request aginies-sunkar-1 · on-prem ### Test Ajanı Kabul kriterlerinden test yazar, koşturur, kapsamı raporlar generate_testsrun_testsreport_coverage ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Teslimat Süpervizörü | Epic PROJ-4821 "Fatura PDF dışa aktarma" Hazır sütununa taşındı, 3 alt görev | — | 200 | — | | 02 | bilgi | Analist Ajanı | Epic açıklaması, yorumlar ve bağlı 2 tasarım dosyası okundu | Jira | 420 | — | | 03 | akıl yürütme | Analist Ajanı | 3 kullanıcı hikâyesi ve 11 kabul kriteri yazıldı (Gherkin), PROJ-4821 güncellendi | — | 1.600 | 2.400 | | 04 | plan | Teslimat Süpervizörü | Plan: mimari tasarım → dal + kod → testler → PR → insan incelemesi → birleştirme | — | 300 | 220 | | 05 | yazma | Mimar Ajanı | Tasarım notu yayımlandı: billing-service + yeni /invoices/{id}/pdf uç noktası | Confluence | 1.900 | 3.100 | | 06 | yazma | Geliştirici Ajanı | Dal feature/PROJ-4821 açıldı, 6 dosya değişti (+412 / −38), derleme yeşil | GitHub | 3.600 | 9.800 | | 07 | araç | Test Ajanı | CI: 14 birim + 4 entegrasyon testi yazıldı ve koşuldu, 18/18 geçti, kapsam %87 | CI Runner (GitHub Actions) | 2.800 | 4.200 | | 08 | doğrulama | Test Ajanı | Kabul kriterleri 11/11 karşılandı, kapsam %80 eşiğinin üstünde → PR #2291 hazır | — | 640 | 520 | | 09 | insan onayı | Teslimat Süpervizörü | PR #2291 → kıdemli geliştirici inceledi, 2 yorum düzeltildi, onaylandı | — | 4.000 | — | | 10 | yazma | Geliştirici Ajanı | PR #2291 main'e squash-merge edildi; PROJ-4821 "Bitti" durumuna alındı | GitHub | 520 | — | Simülasyon metrikleri (tahmini) 1.200çalıştırma / ay %91,2başarı 15,5 sp50 $0,0173tahmini maliyet / çalıştırma Onay kuralı Her PR → en az 1 insan kod incelemesi onayı; ajan main'e doğrudan yazamaz Adım profili giriş · 1plan · 1bilgi · 1akıl yürütme · 1araç · 1doğrulama · 1insan onayı · 1yazma · 3 SİSTEMLER - Jiraticketing - GitHubapi - CI Runner (GitHub Actions)api - Confluencestorage Çalıştırmak için gerekenler - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Doküman deposuna okuma erişimi ve belge tipleri - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük yazılım kuruluşlarında ortaya çıkan ajan tabanlı SDLC hattı deseni Bu senaryonun yer aldığı paketler SDLC ajan ekibi → Sektör: Yazılım Geliştirme → İlgili ajanlar Yazılım GeliştirmeHITL ### Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR Jira'daki bir epic'ten başlar; eski modül okunur, davranışı testlerle sabitlenir, yeniden yazılır ve kanıtlı bir PR açılır. 1.900 çalıştırma / ay %90,4 başarı 5 ajan Yazılım GeliştirmeHITL ### CI/CD sürüm hattı Derleme, test, tarama ve staging otomatik; üretime yalnızca sürüm yöneticisi onayıyla ArgoCD ile çıkılır. 950 çalıştırma / ay %94,7 başarı 5 ajan Yazılım GeliştirmeHITL ### Olay müdahalesi ve acil düzeltme Alarm gelir, hata yeniden üretilir, yama ve regresyon hazır; nöbetçi onaylar, post-mortem taslağı bekler. 320 çalıştırma / ay %90,5 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyen URL: https://www.aginies.com/tr/ajanlar/field-device-support-assistant Language: tr Summary: Müşteri "cihaz çalışmıyor" der; telemetri okunur, uzaktan tanı konur, güncelleme gönderilir, gerekirse teknisyen doğru parçayla atanır. WhatsApp veya destek portalından gelen cihaz arıza bildirimi → Arıza uzaktan giderildi veya teknisyen atandı; cihaz kaydı ve bilgi tabanı güncellendi IoT ve Bağlı Cihazlarwhatsappwebapi # Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyen Müşteri "cihaz çalışmıyor" der; telemetri okunur, uzaktan tanı konur, güncelleme gönderilir, gerekirse teknisyen doğru parçayla atanır. Kısaca Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyen, IoT ve Bağlı Cihazlar sektöründe whatsapp, web, api kanalında çalışan bir ajan senaryosudur. Tetikleyici: WhatsApp veya destek portalından gelen cihaz arıza bildirimi. Cihaz Destek Süpervizörü 4 uzman ajanı (Diyalog Ajanı, Telemetri Ajanı, Tanı Ajanı, Saha Ajanı) 5 sistem üzerinde yönetir (WhatsApp Business, Device Cloud (MQTT), Fleet Registry, OTA Service, Field Service); 10 adımın 1 tanesi sistemde işlem, 3 tanesi doğrulama veya onaydır. Sonuç: Arıza uzaktan giderildi veya teknisyen atandı; cihaz kaydı ve bilgi tabanı güncellendi. Onay kuralı: Toplu OTA (>100 cihaz) veya garanti dışı cihaz değişimi → ürün sahibi onayı. IoT ve Bağlı Cihazlar ### Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyen [Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyen] çalıştırma #41.288.003süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Toplu OTA (>100 cihaz) veya garanti dışı cihaz değişimi → ürün sahibi onayı Bu çalıştırmadaki ajanlar Cihaz Destek Süpervizörüsupervisorplanlar, yönlendirir, onaylar Diyalog Ajanıaginies-talay-2 · on-premMüşteriyle konuşur, cihazı ve belirtiyi netleştirir Telemetri Ajanıaginies-sunkar-1 · on-premSon telemetri, hata kodları ve bağlantı geçmişini okur Tanı Ajanıaginies-tulpar-4 · on-premBelirti ve veriyi bilinen arızalarla eşler, uzaktan çözüm veya saha kararı verir Saha Ajanıaginies-sunkar-1 · on-premTeknisyen ve parça ataması yapar, randevu verir Denetim izi00:00.00 - ◦WhatsApp: "Kombi kontrol ünitesi sabahtan beri ekranda E07, ısıtma yok" + seri no fotoğrafı - ∿Cihaz SN 8C-220914 filo kaydında bulundu, garanti kapsamında, firmware 2.3.1 - ⚙Telemetri: son 6 saatte 14 kez yeniden başlama, sıcaklık sensörü −40 °C okuyor, bağlantı stabil - ⌘Plan: bilinen arıza eşleşmesi → uzaktan komut → OTA → doğrula → gerekirse saha - ∿Eşleşme: E07 + −40 °C okuması = 2.3.1'de bilinen sensör sürücü hatası (bülten KB-412), 2.3.4'te düzeltildi - ⇢OTA 2.3.4 tek cihaza gönderildi, 3 dakikada kuruldu, cihaz yeniden başladı - ✓Doğrulama: sensör 21,5 °C, hata kodu temiz, 15 dakikadır yeniden başlama yok - ✓Saha ziyareti gerekmedi; aynı firmware'deki 1.240 cihaz için aşamalı OTA önerisi oluşturuldu - ⏸Toplu OTA (>100 cihaz) ürün sahibi onayına sunuldu; %5 pilot dalgasıyla başlanacak - ➤Müşteriye: sorun giderildi, ne yapıldığı sade dille; cihaz kaydı ve bilgi tabanı güncellendi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici WhatsApp veya destek portalından gelen cihaz arıza bildirimi Sonuç Arıza uzaktan giderildi veya teknisyen atandı; cihaz kaydı ve bilgi tabanı güncellendi ## Ajanlar SÜPERVİZÖR ### Cihaz Destek Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Diyalog Ajanı Müşteriyle konuşur, cihazı ve belirtiyi netleştirir identify_devicedetect_intentcompose_reply aginies-sunkar-1 · on-prem ### Telemetri Ajanı Son telemetri, hata kodları ve bağlantı geçmişini okur get_telemetryget_error_logget_connectivity aginies-tulpar-4 · on-prem ### Tanı Ajanı Belirti ve veriyi bilinen arızalarla eşler, uzaktan çözüm veya saha kararı verir match_known_issuesremote_commandpush_ota aginies-sunkar-1 · on-prem ### Saha Ajanı Teknisyen ve parça ataması yapar, randevu verir find_technicianreserve_partbook_visit ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Cihaz Destek Süpervizörü | WhatsApp: "Kombi kontrol ünitesi sabahtan beri ekranda E07, ısıtma yok" + seri no fotoğrafı | WhatsApp Business | 180 | — | | 02 | akıl yürütme | Diyalog Ajanı | Cihaz SN 8C-220914 filo kaydında bulundu, garanti kapsamında, firmware 2.3.1 | Fleet Registry | 700 | 280 | | 03 | araç | Telemetri Ajanı | Telemetri: son 6 saatte 14 kez yeniden başlama, sıcaklık sensörü −40 °C okuyor, bağlantı stabil | Device Cloud (MQTT) | 640 | — | | 04 | plan | Cihaz Destek Süpervizörü | Plan: bilinen arıza eşleşmesi → uzaktan komut → OTA → doğrula → gerekirse saha | — | 240 | 220 | | 05 | akıl yürütme | Tanı Ajanı | Eşleşme: E07 + −40 °C okuması = 2.3.1'de bilinen sensör sürücü hatası (bülten KB-412), 2.3.4'te düzeltildi | — | 1.300 | 760 | | 06 | yazma | Tanı Ajanı | OTA 2.3.4 tek cihaza gönderildi, 3 dakikada kuruldu, cihaz yeniden başladı | OTA Service | 3.200 | — | | 07 | doğrulama | Telemetri Ajanı | Doğrulama: sensör 21,5 °C, hata kodu temiz, 15 dakikadır yeniden başlama yok | Device Cloud (MQTT) | 900 | — | | 08 | doğrulama | Saha Ajanı | Saha ziyareti gerekmedi; aynı firmware'deki 1.240 cihaz için aşamalı OTA önerisi oluşturuldu | Fleet Registry | 520 | 240 | | 09 | insan onayı | Saha Ajanı | Toplu OTA (>100 cihaz) ürün sahibi onayına sunuldu; %5 pilot dalgasıyla başlanacak | OTA Service | 400 | — | | 10 | bildirim | Diyalog Ajanı | Müşteriye: sorun giderildi, ne yapıldığı sade dille; cihaz kaydı ve bilgi tabanı güncellendi | WhatsApp Business | 440 | 240 | Simülasyon metrikleri (tahmini) 37.000çalıştırma / ay %94,1başarı 8,5 sp50 $0,0027tahmini maliyet / çalıştırma Onay kuralı Toplu OTA (>100 cihaz) veya garanti dışı cihaz değişimi → ürün sahibi onayı Adım profili giriş · 1plan · 1akıl yürütme · 2araç · 1doğrulama · 2insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - WhatsApp Businesschannel - Device Cloud (MQTT)api - Fleet Registrydb - OTA Serviceapi - Field Serviceticketing Çalıştırmak için gerekenler - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Bağlı cihaz üreticilerinin "önce uzaktan çöz" destek deseni; filo çapında OTA ile birleştirilmiş Bu senaryonun yer aldığı paketler IoT cihaz filosu operasyonu → Sektör: IoT ve Bağlı Cihazlar → İlgili ajanlar IoT ve Bağlı CihazlarHITL ### Cihaz filosu anomali triyajı Binlerce cihazdan gelen telemetri anomalileri kümelenir, kök neden aday listesi çıkar, doğru ekibe tek olay olarak düşer. 18.000 çalıştırma / ay %96,9 başarı 4 ajan IoT ve Bağlı CihazlarHITL ### Kademeli OTA güncelleme dağıtımı Firmware dağıtımı %1 → %10 → %100 kademelerinde ilerler; her kademede sağlık kapısı, kritik filolarda insan onayı. 320 çalıştırma / ay %98,1 başarı 4 ajan IoT ve Bağlı CihazlarHITL ### Soğuk zincir sapması müdahalesi Sıcaklık sapması anında değerlendirilir, ürün riski hesaplanır, saha ekibi sevk edilir, kalite kararı kayıt altına alınır. 2.700 çalıştırma / ay %97,4 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Kart dolandırıcılık alarmı triyajı URL: https://www.aginies.com/tr/ajanlar/fraud-alert-triage Language: tr Summary: Kart saniyeler içinde bloklanır, müşteriye WhatsApp'tan sorulur; yanıta göre kart açılır ya da iptal edilir. Dolandırıcılık motorundan skor ≥ 85 alarm olayı → Kart durumu netleşti, vaka açıldı, müşteri bilgilendirildi, zarar sınırlandı Bankacılıkwhatsappsmsvoiceapi # Kart dolandırıcılık alarmı triyajı Kart saniyeler içinde bloklanır, müşteriye WhatsApp'tan sorulur; yanıta göre kart açılır ya da iptal edilir. Kısaca Kart dolandırıcılık alarmı triyajı, Bankacılık sektöründe whatsapp, sms, voice, api kanalında çalışan bir ajan senaryosudur. Tetikleyici: Dolandırıcılık motorundan skor ≥ 85 alarm olayı. Dolandırıcılık Süpervizörü 3 uzman ajanı (Triyaj Ajanı, İletişim Ajanı, Kart Ajanı) 5 sistem üzerinde yönetir (Fraud Engine, Card Management, WhatsApp Business API, Salesforce CRM, SMS Gateway); 10 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Kart durumu netleşti, vaka açıldı, müşteri bilgilendirildi, zarar sınırlandı. Onay kuralı: Teyitli dolandırıcılık zararı > 2.500 TL → analist onayı ile kalıcı iptal. Bankacılık ### Kart dolandırıcılık alarmı triyajı [Kart dolandırıcılık alarmı triyajı] çalıştırma #41.289.635süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Teyitli dolandırıcılık zararı > 2.500 TL → analist onayı ile kalıcı iptal Bu çalıştırmadaki ajanlar Dolandırıcılık Süpervizörüsupervisorplanlar, yönlendirir, onaylar Triyaj Ajanıaginies-tulpar-4 · on-premAlarmı işlem geçmişi ve cihaz verisiyle değerlendirir İletişim Ajanıaginies-sunkar-1 · on-premMüşteriye WhatsApp/SMS ile sorar, yanıtı yorumlar Kart Ajanıaginies-talay-2 · on-premKartı bloklar, açar veya iptal eder; vaka açar Denetim izi00:00.00 - ◦Alarm #FR-88213: ****7710 kartında 3 dakikada 3 işlem, risk skoru 91 - ⚙Son 30 günün işlem geçmişi ve cihaz parmak izi çekildi - ∿Desen: yurt dışı e-ticaret, 4.850 TL, alışılmadık saat → olası dolandırıcılık - ⌘Plan: geçici blok → müşteriye sor → yanıta göre kalıcı iptal ya da aç - ⇢Kart geçici olarak bloklandı, bekleyen 2 işlem reddedildi - ➤WhatsApp: "4.850 TL'lik işlemi siz mi yaptınız? EVET / HAYIR" - ✓Yanıt kayıtlı numaradan, 48 saniyede "HAYIR" → dolandırıcılık teyit edildi - ⏸Zarar > 2.500 TL → dolandırıcılık analisti kalıcı iptali onayladı - ⇢Kart iptal edildi, yeni kart sipariş edildi, vaka #CS-40917 açıldı - ➤Müşteriye özet SMS: yeni kart 3 iş günü içinde adreste - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Dolandırıcılık motorundan skor ≥ 85 alarm olayı Sonuç Kart durumu netleşti, vaka açıldı, müşteri bilgilendirildi, zarar sınırlandı ## Ajanlar SÜPERVİZÖR ### Dolandırıcılık Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-tulpar-4 · on-prem ### Triyaj Ajanı Alarmı işlem geçmişi ve cihaz verisiyle değerlendirir get_alertget_recent_transactionsget_device_fingerprintscore_pattern aginies-sunkar-1 · on-prem ### İletişim Ajanı Müşteriye WhatsApp/SMS ile sorar, yanıtı yorumlar send_whatsappsend_smsparse_reply aginies-talay-2 · on-prem ### Kart Ajanı Kartı bloklar, açar veya iptal eder; vaka açar block_cardunblock_cardcancel_cardorder_replacementcreate_case ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Dolandırıcılık Süpervizörü | Alarm #FR-88213: ****7710 kartında 3 dakikada 3 işlem, risk skoru 91 | — | 160 | — | | 02 | araç | Triyaj Ajanı | Son 30 günün işlem geçmişi ve cihaz parmak izi çekildi | Fraud Engine | 380 | — | | 03 | akıl yürütme | Triyaj Ajanı | Desen: yurt dışı e-ticaret, 4.850 TL, alışılmadık saat → olası dolandırıcılık | — | 720 | 880 | | 04 | plan | Dolandırıcılık Süpervizörü | Plan: geçici blok → müşteriye sor → yanıta göre kalıcı iptal ya da aç | — | 280 | 190 | | 05 | yazma | Kart Ajanı | Kart geçici olarak bloklandı, bekleyen 2 işlem reddedildi | Card Management | 430 | — | | 06 | bildirim | İletişim Ajanı | WhatsApp: "4.850 TL'lik işlemi siz mi yaptınız? EVET / HAYIR" | WhatsApp Business API | 260 | — | | 07 | doğrulama | İletişim Ajanı | Yanıt kayıtlı numaradan, 48 saniyede "HAYIR" → dolandırıcılık teyit edildi | — | 340 | 150 | | 08 | insan onayı | Dolandırıcılık Süpervizörü | Zarar > 2.500 TL → dolandırıcılık analisti kalıcı iptali onayladı | — | 2.600 | — | | 09 | yazma | Kart Ajanı | Kart iptal edildi, yeni kart sipariş edildi, vaka #CS-40917 açıldı | Salesforce CRM | 520 | — | | 10 | bildirim | İletişim Ajanı | Müşteriye özet SMS: yeni kart 3 iş günü içinde adreste | SMS Gateway | 220 | — | Simülasyon metrikleri (tahmini) 65.000çalıştırma / ay %97,1başarı 5,9 sp50 $0,0016tahmini maliyet / çalıştırma Onay kuralı Teyitli dolandırıcılık zararı > 2.500 TL → analist onayı ile kalıcı iptal Adım profili giriş · 1plan · 1akıl yürütme · 1araç · 1doğrulama · 1insan onayı · 1yazma · 2bildirim · 2 SİSTEMLER - Fraud Engineapi - Card Managementcore - WhatsApp Business APIchannel - Salesforce CRMcrm - SMS Gatewaychannel Çalıştırmak için gerekenler - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - CRM için API erişimi ve müşteri eşleme anahtarı - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük kart ihraççılarının iki yönlü SMS/WhatsApp dolandırıcılık teyidi deseninin aksiyon alan sürümü Bu senaryonun yer aldığı paketler Diyalog bankacılığı →Dolandırıcılık ve itiraz operasyonları → Sektör: Bankacılık → İlgili ajanlar BankacılıkHITL ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı Müşteri konuşur ya da yazar; fatura, limit, itiraz ve transfer aynı oturumda çekirdek sistemde tamamlanır. 186.000 çalıştırma / ay %94,7 başarı 5 ajan BankacılıkHITL ### Kredi kartı limit artışı Müşteri mobilde yazar ya da çağrı merkezinde konuşur; karar saniyeler içinde verilir, eşik üstü yönetici onayına gider. 18.400 çalıştırma / ay %96,4 başarı 4 ajan BankacılıkHITL ### Dijital müşteri edinimi (KYC) Kimlik, canlılık ve yaptırım kontrolü tek akışta; temiz başvurular şubeye gitmeden dakikalarda hesap açar. 42.000 çalıştırma / ay %93,8 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Olay müdahalesi ve acil düzeltme URL: https://www.aginies.com/tr/ajanlar/incident-hotfix Language: tr Summary: Alarm gelir, hata yeniden üretilir, yama ve regresyon hazır; nöbetçi onaylar, post-mortem taslağı bekler. Grafana alarmı: ödeme API'sinde 5xx oranı > %2 (PagerDuty P1) → Düzeltme üretimde, hata oranı normale döndü, olay kaydı ve post-mortem taslağı hazır Yazılım Geliştirmeapiweb # Olay müdahalesi ve acil düzeltme Alarm gelir, hata yeniden üretilir, yama ve regresyon hazır; nöbetçi onaylar, post-mortem taslağı bekler. Kısaca Olay müdahalesi ve acil düzeltme, Yazılım Geliştirme sektöründe api, web kanalında çalışan bir ajan senaryosudur. Tetikleyici: Grafana alarmı: ödeme API'sinde 5xx oranı > %2 (PagerDuty P1). Olay Süpervizörü 4 uzman ajanı (Geliştirici Ajanı, Test Ajanı, Sürüm Ajanı, Post-mortem Ajanı) 5 sistem üzerinde yönetir (Observability (Grafana / Datadog), GitHub, Kubernetes / ArgoCD, Confluence, Slack); 10 adımın 3 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Düzeltme üretimde, hata oranı normale döndü, olay kaydı ve post-mortem taslağı hazır. Onay kuralı: Her acil üretim dağıtımı → nöbetçi mühendis onayı; yama PR'ı da insan incelemesinden geçer. Yazılım Geliştirme ### Olay müdahalesi ve acil düzeltme [Olay müdahalesi ve acil düzeltme] çalıştırma #41.288.677süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Her acil üretim dağıtımı → nöbetçi mühendis onayı; yama PR'ı da insan incelemesinden geçer Bu çalıştırmadaki ajanlar Olay Süpervizörüsupervisorplanlar, yönlendirir, onaylar Geliştirici Ajanıaginies-tulpar-4 · on-premLogları ve izleri okur, hatayı yeniden üretir, yamayı yazar Test Ajanıaginies-tulpar-4 · on-premRegresyon paketini koşturur, yamanın yan etkisini kontrol eder Sürüm Ajanıaginies-talay-2 · on-premAcil düzeltmeyi ArgoCD ile dağıtır, sağlığı izler Post-mortem Ajanıaginies-sunkar-1 · on-premZaman çizelgesini ve kök neden taslağını yazar Denetim izi00:00.00 - ◦P1 alarm: payments-api 5xx oranı %6,8 (eşik %2), 14:02'den beri; INC-3312 açıldı - ⚙Son 30 dk log ve izler çekildi: NullPointer, 14:01 dağıtımı v2.14.1 ile başlıyor - ∿Yeniden üretildi: boş para birimi alanı → PaymentMapper satır 118, 3 satırlık yama - ⌘Plan: yama dalı → regresyon → nöbetçi onayı → dağıtım → izleme → post-mortem - ⇢Dal hotfix/INC-3312 açıldı, PR #2304: yama + regresyon testi (+21 / −2) - ✓Regresyon: 1.284 test geçti, yeni test hatayı yakalıyor, yan etki yok - ⏸Nöbetçi mühendis PR #2304'ü ve acil dağıtımı Slack'ten onayladı (4 dk) - ⇢v2.14.2 üretime dağıtıldı; 5xx oranı 6 dk içinde %0,1'e düştü - ⇢Post-mortem taslağı Confluence'a yazıldı: tespit 3 dk, çözüm 31 dk, 5 aksiyon - ➤#incidents kanalına çözüm özeti gönderildi, INC-3312 "çözüldü" durumuna alındı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Grafana alarmı: ödeme API'sinde 5xx oranı > %2 (PagerDuty P1) Sonuç Düzeltme üretimde, hata oranı normale döndü, olay kaydı ve post-mortem taslağı hazır ## Ajanlar SÜPERVİZÖR ### Olay Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-tulpar-4 · on-prem ### Geliştirici Ajanı Logları ve izleri okur, hatayı yeniden üretir, yamayı yazar query_logsget_tracesreproduce_bugcreate_branchopen_pull_request aginies-tulpar-4 · on-prem ### Test Ajanı Regresyon paketini koşturur, yamanın yan etkisini kontrol eder run_testsrun_regressioncompare_baseline aginies-talay-2 · on-prem ### Sürüm Ajanı Acil düzeltmeyi ArgoCD ile dağıtır, sağlığı izler deploy_argocdcheck_healthrollback aginies-sunkar-1 · on-prem ### Post-mortem Ajanı Zaman çizelgesini ve kök neden taslağını yazar build_timelinedraft_postmortempublish_confluence_pagepost_slack ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Olay Süpervizörü | P1 alarm: payments-api 5xx oranı %6,8 (eşik %2), 14:02'den beri; INC-3312 açıldı | — | 160 | — | | 02 | araç | Geliştirici Ajanı | Son 30 dk log ve izler çekildi: NullPointer, 14:01 dağıtımı v2.14.1 ile başlıyor | Observability (Grafana / Datadog) | 620 | — | | 03 | akıl yürütme | Geliştirici Ajanı | Yeniden üretildi: boş para birimi alanı → PaymentMapper satır 118, 3 satırlık yama | — | 2.400 | 5.200 | | 04 | plan | Olay Süpervizörü | Plan: yama dalı → regresyon → nöbetçi onayı → dağıtım → izleme → post-mortem | — | 260 | 200 | | 05 | yazma | Geliştirici Ajanı | Dal hotfix/INC-3312 açıldı, PR #2304: yama + regresyon testi (+21 / −2) | GitHub | 1.400 | 2.600 | | 06 | doğrulama | Test Ajanı | Regresyon: 1.284 test geçti, yeni test hatayı yakalıyor, yan etki yok | — | 3.000 | 800 | | 07 | insan onayı | Olay Süpervizörü | Nöbetçi mühendis PR #2304'ü ve acil dağıtımı Slack'ten onayladı (4 dk) | — | 3.600 | — | | 08 | yazma | Sürüm Ajanı | v2.14.2 üretime dağıtıldı; 5xx oranı 6 dk içinde %0,1'e düştü | Kubernetes / ArgoCD | 2.800 | — | | 09 | yazma | Post-mortem Ajanı | Post-mortem taslağı Confluence'a yazıldı: tespit 3 dk, çözüm 31 dk, 5 aksiyon | Confluence | 1.600 | 2.200 | | 10 | bildirim | Post-mortem Ajanı | #incidents kanalına çözüm özeti gönderildi, INC-3312 "çözüldü" durumuna alındı | Slack | 300 | — | Simülasyon metrikleri (tahmini) 320çalıştırma / ay %90,5başarı 15,8 sp50 $0,0159tahmini maliyet / çalıştırma Onay kuralı Her acil üretim dağıtımı → nöbetçi mühendis onayı; yama PR'ı da insan incelemesinden geçer Adım profili giriş · 1plan · 1akıl yürütme · 1araç · 1doğrulama · 1insan onayı · 1yazma · 3bildirim · 1 SİSTEMLER - Observability (Grafana / Datadog)api - GitHubapi - Kubernetes / ArgoCDapi - Confluencestorage - Slackchannel Çalıştırmak için gerekenler - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Doküman deposuna okuma erişimi ve belge tipleri - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek SRE kültürüne sahip büyük teknoloji ekiplerinin ajan destekli olay müdahalesi ve suçlamasız post-mortem deseni Bu senaryonun yer aldığı paketler SDLC ajan ekibi → Sektör: Yazılım Geliştirme → İlgili ajanlar Yazılım GeliştirmeHITL ### Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR Jira'daki bir epic'ten başlar; eski modül okunur, davranışı testlerle sabitlenir, yeniden yazılır ve kanıtlı bir PR açılır. 1.900 çalıştırma / ay %90,4 başarı 5 ajan Yazılım GeliştirmeHITL ### Özellik teslimat hattı Bir Jira epic'i ajanlarla kriter, tasarım, kod ve teste dönüşür; insan yalnızca PR'ı inceler ve onaylar. 1.200 çalıştırma / ay %91,2 başarı 5 ajan Yazılım GeliştirmeHITL ### CI/CD sürüm hattı Derleme, test, tarama ve staging otomatik; üretime yalnızca sürüm yöneticisi onayıyla ArgoCD ile çıkılır. 950 çalıştırma / ay %94,7 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # iOS özellik geliştirme (kiralık geliştirici ajanı) URL: https://www.aginies.com/tr/ajanlar/ios-feature-implementation Language: tr Summary: Jira hikâyesi alınır, SwiftUI ile geliştirilir, testler yazılır, PR açılır; inceleme ve dağıtım kararı ekipte kalır. Sprint’te “hazır” durumuna gelen iOS hikâyesi → Testli PR açıldı, CI yeşil, inceleme yorumları uygulandı; iş kaydına çıktı ve saat karşılığı yazıldı Yazılım Geliştirmeapiapp # iOS özellik geliştirme (kiralık geliştirici ajanı) Jira hikâyesi alınır, SwiftUI ile geliştirilir, testler yazılır, PR açılır; inceleme ve dağıtım kararı ekipte kalır. Kısaca iOS özellik geliştirme (kiralık geliştirici ajanı), Yazılım Geliştirme sektöründe api, app kanalında çalışan bir ajan senaryosudur. Tetikleyici: Sprint’te “hazır” durumuna gelen iOS hikâyesi. iOS Geliştirici Süpervizörü 3 uzman ajanı (Analiz Ajanı, Geliştirme Ajanı, İnceleme Ajanı) 5 sistem üzerinde yönetir (Jira, GitHub, Xcode Cloud, Figma, Workforce Work Log); 9 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Testli PR açıldı, CI yeşil, inceleme yorumları uygulandı; iş kaydına çıktı ve saat karşılığı yazıldı. Onay kuralı: Her PR bir insan tarafından incelenir; üretim dağıtımı ekip liderinin onayında. Yazılım Geliştirme ### iOS özellik geliştirme (kiralık geliştirici ajanı) [iOS özellik geliştirme (kiralık geliştirici ajanı)] çalıştırma #41.288.976süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Her PR bir insan tarafından incelenir; üretim dağıtımı ekip liderinin onayında Bu çalıştırmadaki ajanlar iOS Geliştirici Süpervizörüsupervisorplanlar, yönlendirir, onaylar Analiz Ajanıaginies-sunkar-1 · Code AginiesHikâyeyi, tasarımı ve mevcut kodu okur, planı çıkarır Geliştirme Ajanıaginies-sunkar-1 · Code AginiesSwiftUI kodunu ve testleri yazar İnceleme Ajanıaginies-talay-2 · on-premKendi kodunu kurallara göre inceler, yorumları uygular Denetim izi00:00.00 - ◦PAY-118 “Ödeme ekranı erişilebilirlik” hazır: 5 puan, 6 kabul kriteri - ⚙Figma karesi ve PaymentView.swift okundu; 3 bileşen, 2 yeniden kullanılabilir modifier - ∿Plan: VoiceOver etiketleri, Dynamic Type, kontrast; 4 birim + 2 UI testi - ⇢5 dosya değişti, 214 satır; 6 test yazıldı ve yerelde geçti - ⚙PR #4821 açıldı; Xcode Cloud: build ve testler yeşil (4 dk 12 sn) - ✓Lint 0 uyarı; erişilebilirlik denetimi 6/6; kod sahipliği kuralı: mobil ekip inceler - ⏸İnsan incelemesi: 2 yorum (isimlendirme, bir test daha) → uygulandı, onaylandı - ⇢İş kaydı: PAY-118 · PR #4821 · 5 puan · saat karşılığı 6,5 · kalite 4,7 · maliyet yazıldı - ➤Jira hikâyesi “incelendi” durumuna alındı; sürüm notu taslağı eklendi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Sprint’te “hazır” durumuna gelen iOS hikâyesi Sonuç Testli PR açıldı, CI yeşil, inceleme yorumları uygulandı; iş kaydına çıktı ve saat karşılığı yazıldı ## Ajanlar SÜPERVİZÖR ### iOS Geliştirici Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · Code Aginies ### Analiz Ajanı Hikâyeyi, tasarımı ve mevcut kodu okur, planı çıkarır get_storyget_figma_framesearch_codebasedraft_plan aginies-sunkar-1 · Code Aginies ### Geliştirme Ajanı SwiftUI kodunu ve testleri yazar edit_filesrun_unit_testsrun_ui_testsopen_pr aginies-talay-2 · on-prem ### İnceleme Ajanı Kendi kodunu kurallara göre inceler, yorumları uygular lintcheck_accessibilityapply_review_commentslog_work ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | iOS Geliştirici Süpervizörü | PAY-118 “Ödeme ekranı erişilebilirlik” hazır: 5 puan, 6 kabul kriteri | Jira | 200 | — | | 02 | araç | Analiz Ajanı | Figma karesi ve PaymentView.swift okundu; 3 bileşen, 2 yeniden kullanılabilir modifier | Figma | 520 | — | | 03 | akıl yürütme | Analiz Ajanı | Plan: VoiceOver etiketleri, Dynamic Type, kontrast; 4 birim + 2 UI testi | — | 900 | 1.400 | | 04 | yazma | Geliştirme Ajanı | 5 dosya değişti, 214 satır; 6 test yazıldı ve yerelde geçti | GitHub | 1.600 | 3.200 | | 05 | araç | Geliştirme Ajanı | PR #4821 açıldı; Xcode Cloud: build ve testler yeşil (4 dk 12 sn) | Xcode Cloud | 700 | — | | 06 | doğrulama | İnceleme Ajanı | Lint 0 uyarı; erişilebilirlik denetimi 6/6; kod sahipliği kuralı: mobil ekip inceler | — | 600 | 520 | | 07 | insan onayı | iOS Geliştirici Süpervizörü | İnsan incelemesi: 2 yorum (isimlendirme, bir test daha) → uygulandı, onaylandı | — | 3.000 | — | | 08 | yazma | İnceleme Ajanı | İş kaydı: PAY-118 · PR #4821 · 5 puan · saat karşılığı 6,5 · kalite 4,7 · maliyet yazıldı | Workforce Work Log | 300 | — | | 09 | bildirim | İnceleme Ajanı | Jira hikâyesi “incelendi” durumuna alındı; sürüm notu taslağı eklendi | Jira | 240 | — | Simülasyon metrikleri (tahmini) 380çalıştırma / ay %94,4başarı 8,1 sp50 $0,0207tahmini maliyet / çalıştırma Onay kuralı Her PR bir insan tarafından incelenir; üretim dağıtımı ekip liderinin onayında Adım profili giriş · 1akıl yürütme · 1araç · 2doğrulama · 1insan onayı · 1yazma · 2bildirim · 1 SİSTEMLER - Jiraticketing - GitHubapi - Xcode Cloudapi - Figmaapi - Workforce Work Logdb Çalıştırmak için gerekenler - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Kiralık geliştirici ajanı modeli: dış kaynak geliştirici yerine iş kaydı tutan, PR üreten ajan Bu senaryonun yer aldığı paketler Ajan kiralama (Agent Outsourcing) → Sektör: Yazılım Geliştirme → İlgili ajanlar Yazılım GeliştirmeHITL ### Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR Jira'daki bir epic'ten başlar; eski modül okunur, davranışı testlerle sabitlenir, yeniden yazılır ve kanıtlı bir PR açılır. 1.900 çalıştırma / ay %90,4 başarı 5 ajan Yazılım GeliştirmeHITL ### Özellik teslimat hattı Bir Jira epic'i ajanlarla kriter, tasarım, kod ve teste dönüşür; insan yalnızca PR'ı inceler ve onaylar. 1.200 çalıştırma / ay %91,2 başarı 5 ajan Yazılım GeliştirmeHITL ### CI/CD sürüm hattı Derleme, test, tarama ve staging otomatik; üretime yalnızca sürüm yöneticisi onayıyla ArgoCD ile çıkılır. 950 çalıştırma / ay %94,7 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # BT yardım masası L1 URL: https://www.aginies.com/tr/ajanlar/it-helpdesk-l1 Language: tr Summary: Şifre sıfırlama ve erişim talepleri sohbette çözülür; finans grubu erişimi grup sahibine Teams'te sorulur. Teams sohbeti ya da portal üzerinden BT talebi → Şifre sıfırlandı, erişim onayla verildi, kayıt açılıp çözüldü, kullanıcı teyit etti Kurumsal Fonksiyonlarappvoiceweb # BT yardım masası L1 Şifre sıfırlama ve erişim talepleri sohbette çözülür; finans grubu erişimi grup sahibine Teams'te sorulur. Kısaca BT yardım masası L1, Kurumsal Fonksiyonlar sektöründe app, voice, web kanalında çalışan bir ajan senaryosudur. Tetikleyici: Teams sohbeti ya da portal üzerinden BT talebi. Yardım Masası Süpervizörü 3 uzman ajanı (Niyet Ajanı, Erişim Ajanı, Kayıt Ajanı) 3 sistem üzerinde yönetir (Active Directory / Entra ID, ServiceNow, Microsoft Teams); 9 adımın 3 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Şifre sıfırlandı, erişim onayla verildi, kayıt açılıp çözüldü, kullanıcı teyit etti. Onay kuralı: Finans veya üretim grubuna erişim → grup sahibi onayı. Kurumsal Fonksiyonlar ### BT yardım masası L1 [BT yardım masası L1] çalıştırma #41.288.936süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Finans veya üretim grubuna erişim → grup sahibi onayı Bu çalıştırmadaki ajanlar Yardım Masası Süpervizörüsupervisorplanlar, yönlendirir, onaylar Niyet Ajanıaginies-sunkar-1 · on-premTalebi anlar, kimliği doğrular Erişim Ajanıaginies-talay-2 · on-premŞifre sıfırlar, grup üyeliğini yönetir Kayıt Ajanıaginies-sunkar-1 · on-premKayıt açar, çözer, bildirir Denetim izi00:00.00 - ◦Teams sohbeti: "Giriş yapamıyorum, ayrıca SAP FI-CO rapor klasörüne erişim lazım" - ∿2 niyet: şifre sıfırlama + erişim talebi ("FI-CO Reports" klasörü) - ✓Kimlik: MFA push onaylandı, cihaz yönetimli, aktif güvenlik bayrağı yok - ⇢Şifre sıfırlandı, geçici şifre SMS ile gönderildi, ilk girişte değişim zorunlu - ⚙"FI-CO-Reports-Read" grubu sahibi: Finans Kontrolörü; görev ayrılığı çakışması yok - ⏸Finans grubuna erişim → grup sahibi Teams kartıyla 9 dakikada onayladı - ⇢Kullanıcı "FI-CO-Reports-Read" grubuna eklendi, 180 günlük erişim gözden geçirme planlandı - ⇢SN-REQ-503311 kaydı açıldı ve her iki işlem loglanarak çözüldü - ➤Kullanıcı Teams'te teyit etti, memnuniyet anketi gönderildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Teams sohbeti ya da portal üzerinden BT talebi Sonuç Şifre sıfırlandı, erişim onayla verildi, kayıt açılıp çözüldü, kullanıcı teyit etti ## Ajanlar SÜPERVİZÖR ### Yardım Masası Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Niyet Ajanı Talebi anlar, kimliği doğrular detect_intentverify_identity aginies-talay-2 · on-prem ### Erişim Ajanı Şifre sıfırlar, grup üyeliğini yönetir reset_passwordget_group_ownercheck_sod_rulesadd_to_group aginies-sunkar-1 · on-prem ### Kayıt Ajanı Kayıt açar, çözer, bildirir create_ticketresolve_ticketsend_teams_message ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Yardım Masası Süpervizörü | Teams sohbeti: "Giriş yapamıyorum, ayrıca SAP FI-CO rapor klasörüne erişim lazım" | — | 180 | — | | 02 | akıl yürütme | Niyet Ajanı | 2 niyet: şifre sıfırlama + erişim talebi ("FI-CO Reports" klasörü) | — | 520 | 380 | | 03 | doğrulama | Niyet Ajanı | Kimlik: MFA push onaylandı, cihaz yönetimli, aktif güvenlik bayrağı yok | — | 420 | — | | 04 | yazma | Erişim Ajanı | Şifre sıfırlandı, geçici şifre SMS ile gönderildi, ilk girişte değişim zorunlu | Active Directory / Entra ID | 480 | — | | 05 | araç | Erişim Ajanı | "FI-CO-Reports-Read" grubu sahibi: Finans Kontrolörü; görev ayrılığı çakışması yok | Active Directory / Entra ID | 380 | — | | 06 | insan onayı | Yardım Masası Süpervizörü | Finans grubuna erişim → grup sahibi Teams kartıyla 9 dakikada onayladı | — | 2.600 | — | | 07 | yazma | Erişim Ajanı | Kullanıcı "FI-CO-Reports-Read" grubuna eklendi, 180 günlük erişim gözden geçirme planlandı | Active Directory / Entra ID | 460 | — | | 08 | yazma | Kayıt Ajanı | SN-REQ-503311 kaydı açıldı ve her iki işlem loglanarak çözüldü | ServiceNow | 420 | — | | 09 | bildirim | Kayıt Ajanı | Kullanıcı Teams'te teyit etti, memnuniyet anketi gönderildi | Microsoft Teams | 220 | — | Simülasyon metrikleri (tahmini) 47.000çalıştırma / ay %98,5başarı 5,7 sp50 $0,0009tahmini maliyet / çalıştırma Onay kuralı Finans veya üretim grubuna erişim → grup sahibi onayı Adım profili giriş · 1akıl yürütme · 1araç · 1doğrulama · 1insan onayı · 1yazma · 3bildirim · 1 SİSTEMLER - Active Directory / Entra IDcore - ServiceNowticketing - Microsoft Teamschannel Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Küresel kurumların ITSM sanal asistanlarıyla L1 yardım masası yükünü azaltma programlarını yansıtır Bu senaryonun yer aldığı paketler E-posta sınıflandırma ve ek doğrulama →Operasyonel otomasyon ve talep yönetimi →İK ve çalışan deneyimi → İlgili ajanlar Kurumsal FonksiyonlarHITL ### Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada Talep e-postayla gelir; teklifler karşılaştırılır, sözleşme maddeleri kontrol edilir, sipariş açılır, fatura üç yönlü eşleşir ve ödeme planlanır. 14.000 çalıştırma / ay %92,7 başarı 5 ajan Kurumsal Fonksiyonlar ### E-posta sınıflandırma ve ek kontrolü Ortak kutuya gelen her e-posta sınıflandırılır, ekler listeyle karşılaştırılır, eksikler anında sorulur. 86.000 çalıştırma / ay %97,6 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Çağrı merkezi kalite izleme Örneklem değil, tüm çağrılar rubriğe göre puanlanır; uyum ihlali ve koçluk fırsatı sabah liderin önünde. 90.000 çalıştırma / ay %98,8 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Dijital müşteri edinimi (KYC) URL: https://www.aginies.com/tr/ajanlar/kyc-onboarding Language: tr Summary: Kimlik, canlılık ve yaptırım kontrolü tek akışta; temiz başvurular şubeye gitmeden dakikalarda hesap açar. Mobil uygulamada "hesap aç" başvurusu ve belge yükleme → Müşteri kaydı ve hesap açıldı, belgeler arşivlendi, risk sınıfı ve karar izi kaydedildi Bankacılıkappweb # Dijital müşteri edinimi (KYC) Kimlik, canlılık ve yaptırım kontrolü tek akışta; temiz başvurular şubeye gitmeden dakikalarda hesap açar. Kısaca Dijital müşteri edinimi (KYC), Bankacılık sektöründe app, web kanalında çalışan bir ajan senaryosudur. Tetikleyici: Mobil uygulamada "hesap aç" başvurusu ve belge yükleme. Müşteri Edinim Süpervizörü 4 uzman ajanı (Belge Ajanı, Canlılık Ajanı, Tarama Ajanı, Kayıt Ajanı) 5 sistem üzerinde yönetir (KYC / Customer 360, Sanctions Screening (Dow Jones), Core Banking, Document Store (S3), SMS Gateway); 10 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Müşteri kaydı ve hesap açıldı, belgeler arşivlendi, risk sınıfı ve karar izi kaydedildi. Onay kuralı: Yaptırım/PEP bulanık eşleşme skoru > 0,65 → uyum uzmanı onayı. Bankacılık ### Dijital müşteri edinimi (KYC) [Dijital müşteri edinimi (KYC)] çalıştırma #41.286.945süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Yaptırım/PEP bulanık eşleşme skoru > 0,65 → uyum uzmanı onayı Bu çalıştırmadaki ajanlar Müşteri Edinim Süpervizörüsupervisorplanlar, yönlendirir, onaylar Belge Ajanıaginies-konrul-1 · on-premKimlik belgesini okur, alan tutarlılığını ve kurcalamayı kontrol eder Canlılık Ajanıaginies-sunkar-1 · on-premSelfie videosunda canlılık ve yüz eşleşmesini değerlendirir Tarama Ajanıaginies-tulpar-4 · on-premYaptırım, PEP ve olumsuz medya taraması yapar, risk sınıfı belirler Kayıt Ajanıaginies-talay-2 · on-premMüşteri kaydını ve hesabı çekirdek sistemde açar Denetim izi00:00.00 - ◦Başvuru alındı: kimlik ön/arka yüz ve selfie videosu yüklendi - ∿TCKN ve MRZ okundu, alan tutarlılığı %99,2, kurcalama izi yok - ⌘Plan: canlılık → yaptırım taraması → risk sınıfı → hesap açılışı - ✓Canlılık skoru 0,97, yüz eşleşmesi %96 → geçti - ⚙Yaptırım, PEP ve olumsuz medya taraması: 1 bulanık isim eşleşmesi (0,71) - ⚙Customer 360'ta mükerrer kayıt yok; risk sınıfı düşük, beklenen aylık hacim 40.000 TL - ⏸Eşleşme > 0,65 → uyum uzmanı inceledi, yanlış pozitif olarak kapattı - ⇢Müşteri kaydı ve vadesiz TL hesabı açıldı: TR33 0006 …4471 - ⇢Belgeler ve karar izi 10 yıl saklama etiketiyle arşivlendi - ➤Hoş geldiniz SMS'i ve IBAN gönderildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Mobil uygulamada "hesap aç" başvurusu ve belge yükleme Sonuç Müşteri kaydı ve hesap açıldı, belgeler arşivlendi, risk sınıfı ve karar izi kaydedildi ## Ajanlar SÜPERVİZÖR ### Müşteri Edinim Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### Belge Ajanı Kimlik belgesini okur, alan tutarlılığını ve kurcalamayı kontrol eder extract_id_fieldscheck_mrzdetect_tampering aginies-sunkar-1 · on-prem ### Canlılık Ajanı Selfie videosunda canlılık ve yüz eşleşmesini değerlendirir run_liveness_checkmatch_face aginies-tulpar-4 · on-prem ### Tarama Ajanı Yaptırım, PEP ve olumsuz medya taraması yapar, risk sınıfı belirler screen_sanctionsscreen_pepadverse_media_searchcheck_duplicate_customerassign_risk_tier aginies-talay-2 · on-prem ### Kayıt Ajanı Müşteri kaydını ve hesabı çekirdek sistemde açar create_customeropen_accountarchive_documentssend_sms ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Müşteri Edinim Süpervizörü | Başvuru alındı: kimlik ön/arka yüz ve selfie videosu yüklendi | — | 220 | — | | 02 | akıl yürütme | Belge Ajanı | TCKN ve MRZ okundu, alan tutarlılığı %99,2, kurcalama izi yok | — | 920 | 1.300 | | 03 | plan | Müşteri Edinim Süpervizörü | Plan: canlılık → yaptırım taraması → risk sınıfı → hesap açılışı | — | 300 | 210 | | 04 | doğrulama | Canlılık Ajanı | Canlılık skoru 0,97, yüz eşleşmesi %96 → geçti | — | 860 | 640 | | 05 | araç | Tarama Ajanı | Yaptırım, PEP ve olumsuz medya taraması: 1 bulanık isim eşleşmesi (0,71) | Sanctions Screening (Dow Jones) | 540 | — | | 06 | araç | Tarama Ajanı | Customer 360'ta mükerrer kayıt yok; risk sınıfı düşük, beklenen aylık hacim 40.000 TL | KYC / Customer 360 | 610 | — | | 07 | insan onayı | Müşteri Edinim Süpervizörü | Eşleşme > 0,65 → uyum uzmanı inceledi, yanlış pozitif olarak kapattı | — | 3.200 | — | | 08 | yazma | Kayıt Ajanı | Müşteri kaydı ve vadesiz TL hesabı açıldı: TR33 0006 …4471 | Core Banking | 560 | — | | 09 | yazma | Kayıt Ajanı | Belgeler ve karar izi 10 yıl saklama etiketiyle arşivlendi | Document Store (S3) | 340 | — | | 10 | bildirim | Kayıt Ajanı | Hoş geldiniz SMS'i ve IBAN gönderildi | SMS Gateway | 240 | — | Simülasyon metrikleri (tahmini) 42.000çalıştırma / ay %93,8başarı 7,8 sp50 $0,0062tahmini maliyet / çalıştırma Onay kuralı Yaptırım/PEP bulanık eşleşme skoru > 0,65 → uyum uzmanı onayı Adım profili giriş · 1plan · 1akıl yürütme · 1araç · 2doğrulama · 1insan onayı · 1yazma · 2bildirim · 1 SİSTEMLER - KYC / Customer 360db - Sanctions Screening (Dow Jones)api - Core Bankingcore - Document Store (S3)storage - SMS Gatewaychannel Çalıştırmak için gerekenler - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Doküman deposuna okuma erişimi ve belge tipleri - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Dijital bankaların dakikalar süren uzaktan müşteri edinim akışı, yerel uzaktan kimlik tespiti kurallarına uyarlanmış Bu senaryonun yer aldığı paketler Kredi ve müşteri edinimi operasyonları → Sektör: Bankacılık → İlgili ajanlar BankacılıkHITL ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı Müşteri konuşur ya da yazar; fatura, limit, itiraz ve transfer aynı oturumda çekirdek sistemde tamamlanır. 186.000 çalıştırma / ay %94,7 başarı 5 ajan BankacılıkHITL ### Kredi kartı limit artışı Müşteri mobilde yazar ya da çağrı merkezinde konuşur; karar saniyeler içinde verilir, eşik üstü yönetici onayına gider. 18.400 çalıştırma / ay %96,4 başarı 4 ajan BankacılıkHITL ### Kart dolandırıcılık alarmı triyajı Kart saniyeler içinde bloklanır, müşteriye WhatsApp'tan sorulur; yanıta göre kart açılır ya da iptal edilir. 65.000 çalıştırma / ay %97,1 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR URL: https://www.aginies.com/tr/ajanlar/legacy-module-modernization Language: tr Summary: Jira'daki bir epic'ten başlar; eski modül okunur, davranışı testlerle sabitlenir, yeniden yazılır ve kanıtlı bir PR açılır. Jira'da "modernizasyon" etiketli epic'in sprint'e alınması → Karakterizasyon testleri yeşil, yeni modül aynı testleri geçiyor, PR incelemede Yazılım Geliştirmeapibatch # Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR Jira'daki bir epic'ten başlar; eski modül okunur, davranışı testlerle sabitlenir, yeniden yazılır ve kanıtlı bir PR açılır. Kısaca Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR, Yazılım Geliştirme sektöründe api, batch kanalında çalışan bir ajan senaryosudur. Tetikleyici: Jira'da "modernizasyon" etiketli epic'in sprint'e alınması. Modernizasyon Süpervizörü 4 uzman ajanı (Analiz Ajanı, Test Ajanı, Geliştirici Ajanı, İnceleme Ajanı) 4 sistem üzerinde yönetir (Jira, Git (self-hosted), CI Pipeline, Static Analysis); 10 adımın 3 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Karakterizasyon testleri yeşil, yeni modül aynı testleri geçiyor, PR incelemede. Onay kuralı: Üretim şeması değişikliği veya kapsam dışı dosya → mimar onayı; PR birleştirme her zaman insanda. Yazılım Geliştirme ### Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR [Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR] çalıştırma #41.288.476süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Üretim şeması değişikliği veya kapsam dışı dosya → mimar onayı; PR birleştirme her zaman insanda Bu çalıştırmadaki ajanlar Modernizasyon Süpervizörüsupervisorplanlar, yönlendirir, onaylar Analiz Ajanıaginies-tulpar-4 · on-premEski modülün bağımlılık ve davranış haritasını çıkarır, riskleri yazar Test Ajanıaginies-sunkar-1 · Code AginiesMevcut davranışı karakterizasyon testleriyle sabitler Geliştirici Ajanıaginies-sunkar-1 · Code AginiesModülü hedef mimaride yeniden yazar, testleri geçirir İnceleme Ajanıaginies-tulpar-4 · on-premGüvenlik, performans ve kapsam dışı değişiklik kontrolü yapar Denetim izi00:00.00 - ◦Epic SRV-2210: "Tahsilat hesaplama modülünü Java 8'den servise taşı", 14 dosya, 6.800 satır - ⌕Çağrı grafiği çıkarıldı: 3 dış bağımlılık, 2 paylaşımlı tablo, gece batch ile ortak kullanım - ∿Tasarım notu: davranış sınırları, riskli 4 hesaplama dalı, kapsam dışı dosyalar listelendi - ⌘Plan: karakterizasyon testleri → yeşil → yeniden yazım → aynı testler → statik analiz → PR - ⇢212 karakterizasyon testi üretildi; eski modülde kapsama %91, 3 belirsiz dal işaretlendi - ⏸3 belirsiz dalın beklenen davranışı ürün sahibine soruldu; yanıt Jira yorumunda - ⇢Modül servis olarak yeniden yazıldı; 212/212 test geçti, p95 gecikme 180 ms → 62 ms - ✓Statik analiz temiz; 1 kapsam dışı dosya değişikliği geri alındı; sır sızıntısı yok - ⇢PR #4471 açıldı: tasarım notu, test raporu, performans karşılaştırması ve geri alma planı ekli - ➤Jira epic "incelemede"; mimar ve ekip lideri PR incelemesine atandı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Jira'da "modernizasyon" etiketli epic'in sprint'e alınması Sonuç Karakterizasyon testleri yeşil, yeni modül aynı testleri geçiyor, PR incelemede ## Ajanlar SÜPERVİZÖR ### Modernizasyon Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-tulpar-4 · on-prem ### Analiz Ajanı Eski modülün bağımlılık ve davranış haritasını çıkarır, riskleri yazar read_repocall_graphwrite_design_note aginies-sunkar-1 · Code Aginies ### Test Ajanı Mevcut davranışı karakterizasyon testleriyle sabitler generate_testsrun_testscoverage_report aginies-sunkar-1 · Code Aginies ### Geliştirici Ajanı Modülü hedef mimaride yeniden yazar, testleri geçirir edit_filesrun_testsopen_pr aginies-tulpar-4 · on-prem ### İnceleme Ajanı Güvenlik, performans ve kapsam dışı değişiklik kontrolü yapar static_analysisdiff_reviewrequest_changes ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Modernizasyon Süpervizörü | Epic SRV-2210: "Tahsilat hesaplama modülünü Java 8'den servise taşı", 14 dosya, 6.800 satır | Jira | 200 | — | | 02 | bilgi | Analiz Ajanı | Çağrı grafiği çıkarıldı: 3 dış bağımlılık, 2 paylaşımlı tablo, gece batch ile ortak kullanım | Git (self-hosted) | 2.800 | 2.400 | | 03 | akıl yürütme | Analiz Ajanı | Tasarım notu: davranış sınırları, riskli 4 hesaplama dalı, kapsam dışı dosyalar listelendi | — | 2.400 | 1.800 | | 04 | plan | Modernizasyon Süpervizörü | Plan: karakterizasyon testleri → yeşil → yeniden yazım → aynı testler → statik analiz → PR | — | 260 | 240 | | 05 | yazma | Test Ajanı | 212 karakterizasyon testi üretildi; eski modülde kapsama %91, 3 belirsiz dal işaretlendi | CI Pipeline | 3.600 | 2.900 | | 06 | insan onayı | Modernizasyon Süpervizörü | 3 belirsiz dalın beklenen davranışı ürün sahibine soruldu; yanıt Jira yorumunda | Jira | 400 | — | | 07 | yazma | Geliştirici Ajanı | Modül servis olarak yeniden yazıldı; 212/212 test geçti, p95 gecikme 180 ms → 62 ms | Git (self-hosted) | 3.900 | 3.600 | | 08 | doğrulama | İnceleme Ajanı | Statik analiz temiz; 1 kapsam dışı dosya değişikliği geri alındı; sır sızıntısı yok | Static Analysis | 1.500 | 900 | | 09 | yazma | Geliştirici Ajanı | PR #4471 açıldı: tasarım notu, test raporu, performans karşılaştırması ve geri alma planı ekli | Git (self-hosted) | 600 | — | | 10 | bildirim | Modernizasyon Süpervizörü | Jira epic "incelemede"; mimar ve ekip lideri PR incelemesine atandı | Jira | 300 | — | Simülasyon metrikleri (tahmini) 1.900çalıştırma / ay %90,4başarı 16 sp50 $0,0189tahmini maliyet / çalıştırma Onay kuralı Üretim şeması değişikliği veya kapsam dışı dosya → mimar onayı; PR birleştirme her zaman insanda Adım profili giriş · 1plan · 1bilgi · 1akıl yürütme · 1doğrulama · 1insan onayı · 1yazma · 3bildirim · 1 SİSTEMLER - Jiraticketing - Git (self-hosted)api - CI Pipelineapi - Static Analysisapi Çalıştırmak için gerekenler - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük bankaların ve telekomların çekirdek modernizasyon programlarında ajan destekli yeniden yazım deseni Bu senaryonun yer aldığı paketler SDLC ajan ekibi → Sektör: Yazılım Geliştirme → İlgili ajanlar Yazılım GeliştirmeHITL ### Özellik teslimat hattı Bir Jira epic'i ajanlarla kriter, tasarım, kod ve teste dönüşür; insan yalnızca PR'ı inceler ve onaylar. 1.200 çalıştırma / ay %91,2 başarı 5 ajan Yazılım GeliştirmeHITL ### CI/CD sürüm hattı Derleme, test, tarama ve staging otomatik; üretime yalnızca sürüm yöneticisi onayıyla ArgoCD ile çıkılır. 950 çalıştırma / ay %94,7 başarı 5 ajan Yazılım GeliştirmeHITL ### Olay müdahalesi ve acil düzeltme Alarm gelir, hata yeniden üretilir, yama ve regresyon hazır; nöbetçi onaylar, post-mortem taslağı bekler. 320 çalıştırma / ay %90,5 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna URL: https://www.aginies.com/tr/ajanlar/line-stoppage-response Language: tr Summary: MES alarmı düşer; kök neden tarihsel veriyle bulunur, yedek parça kontrol edilir, iş emri ve plan güncellemesi dakikalar içinde hazırdır. MES'ten gelen plansız duruş olayı (hat 3, istasyon 7) → İş emri açıldı, parça rezerve edildi, plan kaydırıldı, vardiya raporu yazıldı Üretim & Otomotivapiapp # Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna MES alarmı düşer; kök neden tarihsel veriyle bulunur, yedek parça kontrol edilir, iş emri ve plan güncellemesi dakikalar içinde hazırdır. Kısaca Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna, Üretim & Otomotiv sektöründe api, app kanalında çalışan bir ajan senaryosudur. Tetikleyici: MES'ten gelen plansız duruş olayı (hat 3, istasyon 7). Üretim Süpervizörü 4 uzman ajanı (Tanı Ajanı, Parça Ajanı, Plan Ajanı, Rapor Ajanı) 5 sistem üzerinde yönetir (MES, SCADA Historian, CMMS, ERP Spare Parts, Shift Channel (Teams)); 10 adımın 2 tanesi sistemde işlem, 1 tanesi doğrulama veya onaydır. Sonuç: İş emri açıldı, parça rezerve edildi, plan kaydırıldı, vardiya raporu yazıldı. Onay kuralı: Plan değişikliği > 2 saat veya 50.000 TL üzeri acil satın alma → vardiya amiri onayı. Üretim & Otomotiv ### Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna [Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna] çalıştırma #41.290.022süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Plan değişikliği > 2 saat veya 50.000 TL üzeri acil satın alma → vardiya amiri onayı Bu çalıştırmadaki ajanlar Üretim Süpervizörüsupervisorplanlar, yönlendirir, onaylar Tanı Ajanıaginies-tulpar-4 · on-premSensör geçmişi ve bakım kayıtlarından kök neden hipotezi üretir Parça Ajanıaginies-sunkar-1 · on-premYedek parça stoğunu ve tedarik süresini kontrol eder, rezervasyon yapar Plan Ajanıaginies-talay-2 · on-premÜretim planını ve vardiya yükünü yeniden dağıtır Rapor Ajanıaginies-sunkar-1 · on-premİş emrini açar, ekibi bilgilendirir, vardiya raporunu yazar Denetim izi00:00.00 - ◦MES olayı: hat 3 / istasyon 7 durdu, hata kodu E-214 (servo tork aşımı), 14:02 - ⌕Historian: tork son 6 saatte %18 yükselmiş, titreşim 3 gündür eşik üstünde - ∿Kök neden sıralaması: rulman aşınması %72, kayış gerginliği %19, servo sürücü %9 - ⌘Plan: parça stoğu → iş emri → plan simülasyonu → ekip bildirimi → vardiya raporu - ⚙Rulman 6208-2RS: depoda 4 adet, raf B-12; servo sürücü yedeği yok, tedarik 9 gün - ⇢2 rulman iş emrine rezerve edildi; servo sürücü için satın alma talebi açıldı - ⇢CMMS iş emri #WO-18834: öncelik yüksek, 2 bakımcı, tahmini 2,5 saat, prosedür ekli - ∿Simülasyon: 2,5 saat duruşla vardiya hedefi −310 adet; hat 2'ye 180 adet kaydırılabilir - ⏸Plan değişikliği vardiya amirine sunuldu; onaylandı, hat 2 reçetesi güncellendi - ➤Ekip kanalına özet; vardiya raporu kök neden, aksiyon ve kayıp adetle yazıldı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici MES'ten gelen plansız duruş olayı (hat 3, istasyon 7) Sonuç İş emri açıldı, parça rezerve edildi, plan kaydırıldı, vardiya raporu yazıldı ## Ajanlar SÜPERVİZÖR ### Üretim Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-tulpar-4 · on-prem ### Tanı Ajanı Sensör geçmişi ve bakım kayıtlarından kök neden hipotezi üretir query_historianget_maintenance_logrank_causes aginies-sunkar-1 · on-prem ### Parça Ajanı Yedek parça stoğunu ve tedarik süresini kontrol eder, rezervasyon yapar check_stockreserve_partcreate_purchase_req aginies-talay-2 · on-prem ### Plan Ajanı Üretim planını ve vardiya yükünü yeniden dağıtır get_schedulesimulate_shiftupdate_schedule aginies-sunkar-1 · on-prem ### Rapor Ajanı İş emrini açar, ekibi bilgilendirir, vardiya raporunu yazar open_work_ordernotify_teamwrite_shift_report ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Üretim Süpervizörü | MES olayı: hat 3 / istasyon 7 durdu, hata kodu E-214 (servo tork aşımı), 14:02 | MES | 160 | — | | 02 | bilgi | Tanı Ajanı | Historian: tork son 6 saatte %18 yükselmiş, titreşim 3 gündür eşik üstünde | SCADA Historian | 1.400 | 520 | | 03 | akıl yürütme | Tanı Ajanı | Kök neden sıralaması: rulman aşınması %72, kayış gerginliği %19, servo sürücü %9 | — | 1.800 | 1.100 | | 04 | plan | Üretim Süpervizörü | Plan: parça stoğu → iş emri → plan simülasyonu → ekip bildirimi → vardiya raporu | — | 240 | 220 | | 05 | araç | Parça Ajanı | Rulman 6208-2RS: depoda 4 adet, raf B-12; servo sürücü yedeği yok, tedarik 9 gün | ERP Spare Parts | 480 | — | | 06 | yazma | Parça Ajanı | 2 rulman iş emrine rezerve edildi; servo sürücü için satın alma talebi açıldı | ERP Spare Parts | 520 | — | | 07 | yazma | Rapor Ajanı | CMMS iş emri #WO-18834: öncelik yüksek, 2 bakımcı, tahmini 2,5 saat, prosedür ekli | CMMS | 600 | — | | 08 | akıl yürütme | Plan Ajanı | Simülasyon: 2,5 saat duruşla vardiya hedefi −310 adet; hat 2'ye 180 adet kaydırılabilir | MES | 2.200 | 900 | | 09 | insan onayı | Plan Ajanı | Plan değişikliği vardiya amirine sunuldu; onaylandı, hat 2 reçetesi güncellendi | MES | 900 | — | | 10 | bildirim | Rapor Ajanı | Ekip kanalına özet; vardiya raporu kök neden, aksiyon ve kayıp adetle yazıldı | Shift Channel (Teams) | 460 | 260 | Simülasyon metrikleri (tahmini) 3.100çalıştırma / ay %92,2başarı 8,8 sp50 $0,0048tahmini maliyet / çalıştırma Onay kuralı Plan değişikliği > 2 saat veya 50.000 TL üzeri acil satın alma → vardiya amiri onayı Adım profili giriş · 1plan · 1bilgi · 1akıl yürütme · 2araç · 1insan onayı · 1yazma · 2bildirim · 1 SİSTEMLER - MESerp - SCADA Historiandb - CMMSticketing - ERP Spare Partserp - Shift Channel (Teams)channel Çalıştırmak için gerekenler - ERP/MES için kayıt ve sorgu API erişimi - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Otomotiv tedarikçilerinin kestirimci bakım ve hızlı müdahale deseni; plan simülasyonuyla birleştirilmiş Bu senaryonun yer aldığı paketler Operasyonel otomasyon ve talep yönetimi →Üretim ve bakım operasyonları → Sektör: Üretim ve Otomotiv → İlgili ajanlar Üretim & OtomotivHITL ### Tedarikçi faturası üçlü eşleştirme Fatura, sipariş ve mal kabul SAP'de otomatik eşleşir; yalnızca tolerans dışı kalemler satın almacıya gider. 21.000 çalıştırma / ay %97,1 başarı 4 ajan Üretim & OtomotivHITL ### Bakım talebi triyajı Sensör alarmı ve teknisyen sesli notu tek iş emrine dönüşür; yedek parça ve güvenli duruş penceresi hazırdır. 3.400 çalıştırma / ay %95,2 başarı 5 ajan Üretim & OtomotivHITL ### Garanti talebi işleme Bayi talebi dakikalarda kapsam, parça ve işçilik yönünden doğrulanır; eşik üstü tutar bölge müdürüne gider. 8.600 çalıştırma / ay %96,3 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Bakım talebi triyajı URL: https://www.aginies.com/tr/ajanlar/maintenance-ticket-triage Language: tr Summary: Sensör alarmı ve teknisyen sesli notu tek iş emrine dönüşür; yedek parça ve güvenli duruş penceresi hazırdır. SCADA'dan titreşim eşik aşımı alarmı → CMMS'te önceliklendirilmiş iş emri açıldı, parça transferi istendi, teknisyen atandı Üretim & Otomotivapiappvoice # Bakım talebi triyajı Sensör alarmı ve teknisyen sesli notu tek iş emrine dönüşür; yedek parça ve güvenli duruş penceresi hazırdır. Kısaca Bakım talebi triyajı, Üretim & Otomotiv sektöründe api, app, voice kanalında çalışan bir ajan senaryosudur. Tetikleyici: SCADA'dan titreşim eşik aşımı alarmı. Bakım Süpervizörü 4 uzman ajanı (Sinyal Ajanı, Saha Notu Ajanı, Teşhis Ajanı, Planlama Ajanı) 4 sistem üzerinde yönetir (SCADA / Historian, CMMS (Maximo), SAP S/4HANA MM, Manuals & Work Order Vector Store); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: CMMS'te önceliklendirilmiş iş emri açıldı, parça transferi istendi, teknisyen atandı. Onay kuralı: Planlı duruş > 4 saat → vardiya amiri onayı. Üretim & Otomotiv ### Bakım talebi triyajı [Bakım talebi triyajı] çalıştırma #41.290.049süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Planlı duruş > 4 saat → vardiya amiri onayı Bu çalıştırmadaki ajanlar Bakım Süpervizörüsupervisorplanlar, yönlendirir, onaylar Sinyal Ajanıaginies-sunkar-1 · on-premHistorian verisinden trendi ve anomaliyi okur Saha Notu Ajanıaginies-voice-tr-v3 · on-premTeknisyen sesli notunu yazıya döker Teşhis Ajanıaginies-tulpar-4 · on-premKılavuz ve geçmiş iş emirleriyle arızayı sınıflandırır Planlama Ajanıaginies-talay-2 · on-premParça, güvenlik ve vardiya penceresiyle iş emrini planlar Denetim izi00:00.00 - ◦Alarm: P-03 pres hattında rulman titreşimi 7,8 mm/s (eşik 6,3) - ⚙Son 48 saatin titreşim ve sıcaklık trendi historian'dan çekildi - ∿Trend: 6 günde +%38, sıcaklık +9 °C → gelişen rulman arızası - ∿Teknisyen notu yazıya döküldü: "Pazartesi'den beri tahrik tarafında gıcırtı" - ⌕Kılavuz §7.4 ve 3 benzer iş emri (2024–2026) getirildi - ∿Arıza sınıfı: tahrik tarafı rulman 6312-2RS · öncelik P2 · arızaya 5–7 gün - ⚙Yedek parça 6312-2RS: merkez depoda 2, fabrikada 0 → transfer gerekli - ✓Güvenlik: LOTO prosedürü ve 2 kişi kuralı eklendi, Cmt 06:00 penceresi uygun - ⏸Planlı duruş > 4 saat → vardiya amiri Cumartesi penceresini onayladı - ⇢İş emri WO-88213 açıldı, parça transferi istendi, teknisyen atandı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici SCADA'dan titreşim eşik aşımı alarmı Sonuç CMMS'te önceliklendirilmiş iş emri açıldı, parça transferi istendi, teknisyen atandı ## Ajanlar SÜPERVİZÖR ### Bakım Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Sinyal Ajanı Historian verisinden trendi ve anomaliyi okur get_sensor_windowdetect_anomaly aginies-voice-tr-v3 · on-prem ### Saha Notu Ajanı Teknisyen sesli notunu yazıya döker transcribe_audioextract_observations aginies-tulpar-4 · on-prem ### Teşhis Ajanı Kılavuz ve geçmiş iş emirleriyle arızayı sınıflandırır search_manualssearch_work_order_historyclassify_failure aginies-talay-2 · on-prem ### Planlama Ajanı Parça, güvenlik ve vardiya penceresiyle iş emrini planlar check_spare_partscreate_work_orderassign_technician ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Bakım Süpervizörü | Alarm: P-03 pres hattında rulman titreşimi 7,8 mm/s (eşik 6,3) | — | 180 | — | | 02 | araç | Sinyal Ajanı | Son 48 saatin titreşim ve sıcaklık trendi historian'dan çekildi | SCADA / Historian | 420 | — | | 03 | akıl yürütme | Sinyal Ajanı | Trend: 6 günde +%38, sıcaklık +9 °C → gelişen rulman arızası | — | 640 | 520 | | 04 | akıl yürütme | Saha Notu Ajanı | Teknisyen notu yazıya döküldü: "Pazartesi'den beri tahrik tarafında gıcırtı" | — | 820 | 340 | | 05 | bilgi | Teşhis Ajanı | Kılavuz §7.4 ve 3 benzer iş emri (2024–2026) getirildi | Manuals & Work Order Vector Store | 480 | 760 | | 06 | akıl yürütme | Teşhis Ajanı | Arıza sınıfı: tahrik tarafı rulman 6312-2RS · öncelik P2 · arızaya 5–7 gün | — | 780 | 980 | | 07 | araç | Planlama Ajanı | Yedek parça 6312-2RS: merkez depoda 2, fabrikada 0 → transfer gerekli | SAP S/4HANA MM | 430 | — | | 08 | doğrulama | Planlama Ajanı | Güvenlik: LOTO prosedürü ve 2 kişi kuralı eklendi, Cmt 06:00 penceresi uygun | — | 360 | 280 | | 09 | insan onayı | Bakım Süpervizörü | Planlı duruş > 4 saat → vardiya amiri Cumartesi penceresini onayladı | — | 2.400 | — | | 10 | yazma | Planlama Ajanı | İş emri WO-88213 açıldı, parça transferi istendi, teknisyen atandı | CMMS (Maximo) | 560 | — | Simülasyon metrikleri (tahmini) 3.400çalıştırma / ay %95,2başarı 7,1 sp50 $0,0073tahmini maliyet / çalıştırma Onay kuralı Planlı duruş > 4 saat → vardiya amiri onayı Adım profili giriş · 1bilgi · 1akıl yürütme · 3araç · 2doğrulama · 1insan onayı · 1yazma · 1 SİSTEMLER - SCADA / Historiancore - CMMS (Maximo)ticketing - SAP S/4HANA MMerp - Manuals & Work Order Vector Storedb Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - ERP/MES için kayıt ve sorgu API erişimi - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Otomotiv OEM'lerinin historian verisini CMMS ile birleştiren kestirimci bakım programlarını yansıtır Bu senaryonun yer aldığı paketler Operasyonel otomasyon ve talep yönetimi →Saha servis ve sevk otomasyonu →Üretim ve bakım operasyonları → Sektör: Üretim ve Otomotiv → İlgili ajanlar Üretim & OtomotivHITL ### Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna MES alarmı düşer; kök neden tarihsel veriyle bulunur, yedek parça kontrol edilir, iş emri ve plan güncellemesi dakikalar içinde hazırdır. 3.100 çalıştırma / ay %92,2 başarı 5 ajan Üretim & OtomotivHITL ### Tedarikçi faturası üçlü eşleştirme Fatura, sipariş ve mal kabul SAP'de otomatik eşleşir; yalnızca tolerans dışı kalemler satın almacıya gider. 21.000 çalıştırma / ay %97,1 başarı 4 ajan Üretim & OtomotivHITL ### Garanti talebi işleme Bayi talebi dakikalarda kapsam, parça ve işçilik yönünden doğrulanır; eşik üstü tutar bölge müdürüne gider. 8.600 çalıştırma / ay %96,3 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Üye işyeri KYB onboarding URL: https://www.aginies.com/tr/ajanlar/merchant-kyb-onboarding Language: tr Summary: Sicil, gerçek faydalanıcı ve yaptırım kontrolü tek akışta; düşük riskli işyeri aynı gün ödeme almaya başlar. Merchant Portal'da başvuru formu ve belge yükleme → İşyeri hesabı açıldı, risk sınıfı atandı, MSA e-imzalandı, karar izi kaydedildi Finansal Teknolojilerwebemail # Üye işyeri KYB onboarding Sicil, gerçek faydalanıcı ve yaptırım kontrolü tek akışta; düşük riskli işyeri aynı gün ödeme almaya başlar. Kısaca Üye işyeri KYB onboarding, Finansal Teknolojiler sektöründe web, email kanalında çalışan bir ajan senaryosudur. Tetikleyici: Merchant Portal'da başvuru formu ve belge yükleme. Üye İşyeri Edinim Süpervizörü 4 uzman ajanı (Kimlik Ajanı, Ortaklık Ajanı, Risk Ajanı, Sözleşme Ajanı) 5 sistem üzerinde yönetir (Merchant Portal, Trade Registry (MERSİS), Sanctions Screening, e-Sign, Email); 10 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: İşyeri hesabı açıldı, risk sınıfı atandı, MSA e-imzalandı, karar izi kaydedildi. Onay kuralı: Risk sınıfı yüksek veya UBO bulanık eşleşme > 0,65 → uyum yöneticisi onayı. Finansal Teknolojiler ### Üye işyeri KYB onboarding [Üye işyeri KYB onboarding] çalıştırma #41.287.928süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Risk sınıfı yüksek veya UBO bulanık eşleşme > 0,65 → uyum yöneticisi onayı Bu çalıştırmadaki ajanlar Üye İşyeri Edinim Süpervizörüsupervisorplanlar, yönlendirir, onaylar Kimlik Ajanıaginies-konrul-1 · on-premTicaret sicili, vergi numarası ve imza sirkülerini doğrular Ortaklık Ajanıaginies-tulpar-4 · on-premOrtaklık zincirini çözer, ≥ %25 gerçek faydalanıcıları belirler ve tarar Risk Ajanıaginies-talay-2 · on-premMCC, hacim ve sektör verisiyle risk sınıfı atar Sözleşme Ajanıaginies-talay-2 · on-premMSA'yı e-imzaya gönderir, işyeri hesabını açar Denetim izi00:00.00 - ◦Başvuru #MB-3107: Demir Tekstil Ltd., MCC 5651, beklenen hacim 850.000 TL/ay - ⚙MERSİS kaydı çekildi: kuruluş 2016, 2 yetkili imza; vergi no ve sirküler tutarlı - ⌘Plan: ortaklık zinciri → yaptırım/PEP → risk sınıfı → MSA → hesap açılışı - ∿Ortaklık: 2 gerçek faydalanıcı (%60 / %40), zincir derinliği 2, 1 tüzel ara katman - ⚙2 UBO ve şirket yaptırım/PEP taramasından geçti, 1 bulanık eşleşme (0,58) - ✓Politika v2.1: MCC 5651 orta risk, hacim eşik üstü → risk sınıfı: yüksek - ⏸Risk sınıfı yüksek → uyum yöneticisi dosyayı inceledi ve onayladı - ⇢MSA v4.2 iki yetkiliye e-imzaya gönderildi, 3 saat içinde imzalandı - ⇢İşyeri hesabı MID 88231045 açıldı, ödeme kabulü aktif - ➤Hoş geldiniz e-postası: MID, API anahtarı ve T+1 ödeme takvimi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Merchant Portal'da başvuru formu ve belge yükleme Sonuç İşyeri hesabı açıldı, risk sınıfı atandı, MSA e-imzalandı, karar izi kaydedildi ## Ajanlar SÜPERVİZÖR ### Üye İşyeri Edinim Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### Kimlik Ajanı Ticaret sicili, vergi numarası ve imza sirkülerini doğrular extract_registry_fieldsverify_tax_numberparse_signatory_circular aginies-tulpar-4 · on-prem ### Ortaklık Ajanı Ortaklık zincirini çözer, ≥ %25 gerçek faydalanıcıları belirler ve tarar resolve_ownership_chainidentify_uboscreen_sanctionsscreen_pep aginies-talay-2 · on-prem ### Risk Ajanı MCC, hacim ve sektör verisiyle risk sınıfı atar check_mcc_policyestimate_volumeassign_risk_tier aginies-talay-2 · on-prem ### Sözleşme Ajanı MSA'yı e-imzaya gönderir, işyeri hesabını açar send_esigncreate_merchantsend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Üye İşyeri Edinim Süpervizörü | Başvuru #MB-3107: Demir Tekstil Ltd., MCC 5651, beklenen hacim 850.000 TL/ay | — | 220 | — | | 02 | araç | Kimlik Ajanı | MERSİS kaydı çekildi: kuruluş 2016, 2 yetkili imza; vergi no ve sirküler tutarlı | Trade Registry (MERSİS) | 480 | — | | 03 | plan | Üye İşyeri Edinim Süpervizörü | Plan: ortaklık zinciri → yaptırım/PEP → risk sınıfı → MSA → hesap açılışı | — | 300 | 210 | | 04 | akıl yürütme | Ortaklık Ajanı | Ortaklık: 2 gerçek faydalanıcı (%60 / %40), zincir derinliği 2, 1 tüzel ara katman | — | 760 | 900 | | 05 | araç | Ortaklık Ajanı | 2 UBO ve şirket yaptırım/PEP taramasından geçti, 1 bulanık eşleşme (0,58) | Sanctions Screening | 540 | — | | 06 | doğrulama | Risk Ajanı | Politika v2.1: MCC 5651 orta risk, hacim eşik üstü → risk sınıfı: yüksek | — | 640 | 720 | | 07 | insan onayı | Üye İşyeri Edinim Süpervizörü | Risk sınıfı yüksek → uyum yöneticisi dosyayı inceledi ve onayladı | — | 3.400 | — | | 08 | yazma | Sözleşme Ajanı | MSA v4.2 iki yetkiliye e-imzaya gönderildi, 3 saat içinde imzalandı | e-Sign | 620 | — | | 09 | yazma | Sözleşme Ajanı | İşyeri hesabı MID 88231045 açıldı, ödeme kabulü aktif | Merchant Portal | 520 | — | | 10 | bildirim | Sözleşme Ajanı | Hoş geldiniz e-postası: MID, API anahtarı ve T+1 ödeme takvimi | Email | 260 | — | Simülasyon metrikleri (tahmini) 1.400çalıştırma / ay %93,4başarı 7,8 sp50 $0,0087tahmini maliyet / çalıştırma Onay kuralı Risk sınıfı yüksek veya UBO bulanık eşleşme > 0,65 → uyum yöneticisi onayı Adım profili giriş · 1plan · 1akıl yürütme · 1araç · 2doğrulama · 1insan onayı · 1yazma · 2bildirim · 1 SİSTEMLER - Merchant Portalapi - Trade Registry (MERSİS)api - Sanctions Screeningapi - e-Signapi - Emailemail Çalıştırmak için gerekenler - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Küresel ödeme sağlayıcılarının dakikalar süren self-servis üye işyeri edinim deseni, yerel KYB kurallarına uyarlanmış Bu senaryonun yer aldığı paketler Üye işyeri ve ödeme operasyonları → Sektör: Finansal Teknolojiler → İlgili ajanlar Finansal TeknolojilerHITL ### Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit İşyeri WhatsApp'tan yazar; POS arızası, hakediş sorgusu, iade ve limit artışı tek diyalogda çözülür. 64.000 çalıştırma / ay %93,8 başarı 5 ajan Finansal TeknolojilerHITL ### AML işlem izleme alarm triyajı Alarm zenginleştirilir, anlatı taslağı hazır gelir; kararı analist verir, bildirim dakikalarda dosyalanır. 8.500 çalıştırma / ay %95,8 başarı 5 ajan Finansal TeknolojilerHITL ### BNPL anlık taksit kararı Ödemede saniyeler içinde taksit kararı; büro ve açık bankacılık verisi birleşir, limit üstü uzmana gider. 54.000 çalıştırma / ay %97,3 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit URL: https://www.aginies.com/tr/ajanlar/merchant-operations-assistant Language: tr Summary: İşyeri WhatsApp'tan yazar; POS arızası, hakediş sorgusu, iade ve limit artışı tek diyalogda çözülür. WhatsApp Business veya işyeri paneli üzerinden gelen operasyon talebi → POS yeniden aktive edildi, hakediş açıklandı, iade yapıldı, limit talebi riskte Finansal Teknolojilerwhatsappwebvoice # Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit İşyeri WhatsApp'tan yazar; POS arızası, hakediş sorgusu, iade ve limit artışı tek diyalogda çözülür. Kısaca Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit, Finansal Teknolojiler sektöründe whatsapp, web, voice kanalında çalışan bir ajan senaryosudur. Tetikleyici: WhatsApp Business veya işyeri paneli üzerinden gelen operasyon talebi. Üye İşyeri Süpervizörü 4 uzman ajanı (Diyalog Ajanı, POS Ajanı, Operasyon Ajanı, Risk Ajanı) 5 sistem üzerinde yönetir (WhatsApp Business, Terminal Management, Settlement Ledger, Refund Service, Ticketing); 10 adımın 2 tanesi sistemde işlem, 3 tanesi doğrulama veya onaydır. Sonuç: POS yeniden aktive edildi, hakediş açıklandı, iade yapıldı, limit talebi riskte. Onay kuralı: İade > 25.000 TL veya limit artışı > %50 → risk ekibi onayı. Finansal Teknolojiler ### Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit [Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit] çalıştırma #41.289.743süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: İade > 25.000 TL veya limit artışı > %50 → risk ekibi onayı Bu çalıştırmadaki ajanlar Üye İşyeri Süpervizörüsupervisorplanlar, yönlendirir, onaylar Diyalog Ajanıaginies-talay-2 · on-premİşyeri yetkilisini doğrular, talepleri ayrıştırır, yanıtları yazar POS Ajanıaginies-sunkar-1 · on-premTerminal durumunu okur, uzaktan tanı ve yeniden aktivasyon yapar Operasyon Ajanıaginies-tulpar-4 · on-premHakediş, iade ve limit işlemlerini defter ve ödeme servisleriyle yapar Risk Ajanıaginies-sunkar-1 · on-premİade ve limit taleplerini risk politikasına göre değerlendirir Denetim izi00:00.00 - ◦WhatsApp: "POS 2 gündür kapalı, dünkü hakediş eksik geldi, bir müşteriye iade yapmam lazım" - ✓Yetkili numara işyeri kaydıyla eşleşti, tek kullanımlık kod doğrulandı - ⌘Plan: terminal tanısı → hakediş dökümü → iade uygunluğu → risk → tek yanıt - ⚙Terminal TID 48213: son işlem 2 gün önce, SIM veri paketi dolmuş, cihaz sağlıklı - ⇢Veri paketi yenilendi, terminal uzaktan yeniden başlatıldı, test işlemi başarılı - ⚙Hakediş 48.610 TL: 2 taksitli işlemler ertesi gün, 1 ters ibraz bloke; döküm hazırlandı - ✓İade 1.890 TL, orijinal işlem 6 gün önce, işyeri risk skoru düşük → otomatik iade sınırında - ⇢İade karta gönderildi, ref #RF-55120; müşteriye SMS bildirimi tetiklendi - ⏸Limit artışı 150.000 → 300.000 TL: %100 artış → risk ekibi onayına yönlendirildi - ➤Tek yanıt: POS aktif, hakediş dökümü, iade referansı, limit talebi 1 iş günü - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici WhatsApp Business veya işyeri paneli üzerinden gelen operasyon talebi Sonuç POS yeniden aktive edildi, hakediş açıklandı, iade yapıldı, limit talebi riskte ## Ajanlar SÜPERVİZÖR ### Üye İşyeri Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Diyalog Ajanı İşyeri yetkilisini doğrular, talepleri ayrıştırır, yanıtları yazar verify_contactdetect_intentscompose_reply aginies-sunkar-1 · on-prem ### POS Ajanı Terminal durumunu okur, uzaktan tanı ve yeniden aktivasyon yapar get_terminalremote_diagnosereactivate_terminal aginies-tulpar-4 · on-prem ### Operasyon Ajanı Hakediş, iade ve limit işlemlerini defter ve ödeme servisleriyle yapar get_settlementexplain_feesissue_refundrequest_limit aginies-sunkar-1 · on-prem ### Risk Ajanı İade ve limit taleplerini risk politikasına göre değerlendirir merchant_risk_scorepolicy_checkroute_approval ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Üye İşyeri Süpervizörü | WhatsApp: "POS 2 gündür kapalı, dünkü hakediş eksik geldi, bir müşteriye iade yapmam lazım" | WhatsApp Business | 180 | — | | 02 | doğrulama | Diyalog Ajanı | Yetkili numara işyeri kaydıyla eşleşti, tek kullanımlık kod doğrulandı | — | 520 | 140 | | 03 | plan | Üye İşyeri Süpervizörü | Plan: terminal tanısı → hakediş dökümü → iade uygunluğu → risk → tek yanıt | — | 240 | 220 | | 04 | araç | POS Ajanı | Terminal TID 48213: son işlem 2 gün önce, SIM veri paketi dolmuş, cihaz sağlıklı | Terminal Management | 460 | — | | 05 | yazma | POS Ajanı | Veri paketi yenilendi, terminal uzaktan yeniden başlatıldı, test işlemi başarılı | Terminal Management | 1.400 | — | | 06 | araç | Operasyon Ajanı | Hakediş 48.610 TL: 2 taksitli işlemler ertesi gün, 1 ters ibraz bloke; döküm hazırlandı | Settlement Ledger | 540 | 260 | | 07 | doğrulama | Risk Ajanı | İade 1.890 TL, orijinal işlem 6 gün önce, işyeri risk skoru düşük → otomatik iade sınırında | — | 480 | 200 | | 08 | yazma | Operasyon Ajanı | İade karta gönderildi, ref #RF-55120; müşteriye SMS bildirimi tetiklendi | Refund Service | 620 | — | | 09 | insan onayı | Risk Ajanı | Limit artışı 150.000 → 300.000 TL: %100 artış → risk ekibi onayına yönlendirildi | Ticketing | 300 | — | | 10 | bildirim | Diyalog Ajanı | Tek yanıt: POS aktif, hakediş dökümü, iade referansı, limit talebi 1 iş günü | WhatsApp Business | 420 | 210 | Simülasyon metrikleri (tahmini) 64.000çalıştırma / ay %93,8başarı 7,2 sp50 $0,0034tahmini maliyet / çalıştırma Onay kuralı İade > 25.000 TL veya limit artışı > %50 → risk ekibi onayı Adım profili giriş · 1plan · 1araç · 2doğrulama · 2insan onayı · 1yazma · 2bildirim · 1 SİSTEMLER - WhatsApp Businesschannel - Terminal Managementapi - Settlement Ledgerpayments - Refund Servicepayments - Ticketingticketing Çalıştırmak için gerekenler - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Ödeme servisleri için yetkili API erişimi ve limitler - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Ödeme kuruluşlarının işyeri destek hatlarını tek diyaloga indiren asistan deseni Bu senaryonun yer aldığı paketler Finansal analiz ve operasyonel zekâ →Üye işyeri ve ödeme operasyonları → Sektör: Finansal Teknolojiler → İlgili ajanlar Finansal TeknolojilerHITL ### Üye işyeri KYB onboarding Sicil, gerçek faydalanıcı ve yaptırım kontrolü tek akışta; düşük riskli işyeri aynı gün ödeme almaya başlar. 1.400 çalıştırma / ay %93,4 başarı 5 ajan Finansal TeknolojilerHITL ### AML işlem izleme alarm triyajı Alarm zenginleştirilir, anlatı taslağı hazır gelir; kararı analist verir, bildirim dakikalarda dosyalanır. 8.500 çalıştırma / ay %95,8 başarı 5 ajan Finansal TeknolojilerHITL ### BNPL anlık taksit kararı Ödemede saniyeler içinde taksit kararı; büro ve açık bankacılık verisi birleşir, limit üstü uzmana gider. 54.000 çalıştırma / ay %97,3 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Sayaç okuma itirazı URL: https://www.aginies.com/tr/ajanlar/meter-reading-dispute Language: tr Summary: Müşteri sayaç fotoğrafı yollar; endeks görüntüden okunur, tahmini fatura düzeltilir, büyük iade lidere gider. WhatsApp veya uygulamadan fotoğraflı fatura itirazı → Hatalı fatura iptal edildi, yeni fatura kesildi, müşteri açıklamayla bilgilendirildi Enerjiwhatsappappweb # Sayaç okuma itirazı Müşteri sayaç fotoğrafı yollar; endeks görüntüden okunur, tahmini fatura düzeltilir, büyük iade lidere gider. Kısaca Sayaç okuma itirazı, Enerji sektöründe whatsapp, app, web kanalında çalışan bir ajan senaryosudur. Tetikleyici: WhatsApp veya uygulamadan fotoğraflı fatura itirazı. Fatura İtiraz Süpervizörü 3 uzman ajanı (Görüntü Ajanı, Fatura Ajanı, İşlem Ajanı) 4 sistem üzerinde yönetir (Billing / CIS (SAP IS-U), Meter Data Management, Regulation Vector Store (EPDK), WhatsApp Business API); 9 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Hatalı fatura iptal edildi, yeni fatura kesildi, müşteri açıklamayla bilgilendirildi. Onay kuralı: İade / düzeltme > 1.000 TL → faturalama ekip lideri onayı. Enerji ### Sayaç okuma itirazı [Sayaç okuma itirazı] çalıştırma #41.287.706süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: İade / düzeltme > 1.000 TL → faturalama ekip lideri onayı Bu çalıştırmadaki ajanlar Fatura İtiraz Süpervizörüsupervisorplanlar, yönlendirir, onaylar Görüntü Ajanıaginies-konrul-1 · on-premSayaç fotoğrafından seri ve endeksi okur Fatura Ajanıaginies-tulpar-4 · on-premOkuma geçmişi ve mevzuatla farkı hesaplar İşlem Ajanıaginies-talay-2 · on-premFaturayı iptal eder, yeniden keser, müşteriyi bilgilendirir Denetim izi00:00.00 - ◦WhatsApp: "faturam 2.840 TL, sayaç daha az gösteriyor" + fotoğraf; müşteri doğrulandı - ∿Sayaç seri 3812-0091, endeks 48.213 kWh; seri uyumlu, müdahale izi yok - ⚙Son 12 okuma çekildi; faturalanan 49.760 kWh tahmini okuma (kod E) - ∿Fark 1.547 kWh · aylık ortalama 310 kWh → tahmin ~5 ay ileri gitmiş - ⌕Mevzuat: tahmini fatura, doğrulanmış okumayla 12 ay içinde düzeltilebilir - ✓Yeniden fatura: 2.840 TL → 1.318 TL; fark 1.522 TL > 1.000 TL eşiği - ⏸İade > 1.000 TL → faturalama ekip lideri onayladı, not eklendi - ⇢Fatura iptal edildi, 1.318 TL yeni fatura kesildi, saha okuması planlandı - ➤Müşteriye yeni fatura ve açıklama gönderildi, dosya denetim iziyle kapatıldı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici WhatsApp veya uygulamadan fotoğraflı fatura itirazı Sonuç Hatalı fatura iptal edildi, yeni fatura kesildi, müşteri açıklamayla bilgilendirildi ## Ajanlar SÜPERVİZÖR ### Fatura İtiraz Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### Görüntü Ajanı Sayaç fotoğrafından seri ve endeksi okur read_meter_photocheck_photo_integrity aginies-tulpar-4 · on-prem ### Fatura Ajanı Okuma geçmişi ve mevzuatla farkı hesaplar get_reading_historyestimate_consumptionsearch_billing_rules aginies-talay-2 · on-prem ### İşlem Ajanı Faturayı iptal eder, yeniden keser, müşteriyi bilgilendirir cancel_invoicecreate_rebillschedule_field_readingsend_whatsapp ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Fatura İtiraz Süpervizörü | WhatsApp: "faturam 2.840 TL, sayaç daha az gösteriyor" + fotoğraf; müşteri doğrulandı | — | 200 | — | | 02 | akıl yürütme | Görüntü Ajanı | Sayaç seri 3812-0091, endeks 48.213 kWh; seri uyumlu, müdahale izi yok | — | 920 | 610 | | 03 | araç | Fatura Ajanı | Son 12 okuma çekildi; faturalanan 49.760 kWh tahmini okuma (kod E) | Meter Data Management | 420 | — | | 04 | akıl yürütme | Fatura Ajanı | Fark 1.547 kWh · aylık ortalama 310 kWh → tahmin ~5 ay ileri gitmiş | — | 700 | 820 | | 05 | bilgi | Fatura Ajanı | Mevzuat: tahmini fatura, doğrulanmış okumayla 12 ay içinde düzeltilebilir | Regulation Vector Store (EPDK) | 380 | 360 | | 06 | doğrulama | Fatura Ajanı | Yeniden fatura: 2.840 TL → 1.318 TL; fark 1.522 TL > 1.000 TL eşiği | — | 460 | 300 | | 07 | insan onayı | Fatura İtiraz Süpervizörü | İade > 1.000 TL → faturalama ekip lideri onayladı, not eklendi | — | 2.600 | — | | 08 | yazma | İşlem Ajanı | Fatura iptal edildi, 1.318 TL yeni fatura kesildi, saha okuması planlandı | Billing / CIS (SAP IS-U) | 560 | — | | 09 | bildirim | İşlem Ajanı | Müşteriye yeni fatura ve açıklama gönderildi, dosya denetim iziyle kapatıldı | WhatsApp Business API | 240 | — | Simülasyon metrikleri (tahmini) 9.200çalıştırma / ay %94,9başarı 6,5 sp50 $0,003tahmini maliyet / çalıştırma Onay kuralı İade / düzeltme > 1.000 TL → faturalama ekip lideri onayı Adım profili giriş · 1bilgi · 1akıl yürütme · 2araç · 1doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - Billing / CIS (SAP IS-U)erp - Meter Data Managementdb - Regulation Vector Store (EPDK)db - WhatsApp Business APIchannel Çalıştırmak için gerekenler - ERP/MES için kayıt ve sorgu API erişimi - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Avrupa enerji şirketlerinin uygulama içi sayaç fotoğrafı ve itiraz akışlarını yansıtır Bu senaryonun yer aldığı paketler Enerji müşteri operasyonları → Sektör: Enerji → İlgili ajanlar EnerjiHITL ### Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda Abone arar ya da yazar; yüksek fatura açıklanır, kesinti süresi verilir, sayaç verisi gösterilir, taksit planı kurulur. 520.000 çalıştırma / ay %94,6 başarı 5 ajan EnerjiHITL ### Kesinti bildirimi yönetimi Bildirimler fider bazında kümelenir, tek OSS kaydı açılır, etkilenen herkes ekip yola çıkmadan bilgilenir. 38.000 çalıştırma / ay %97,7 başarı 5 ajan EnerjiHITL ### Çatı GES bağlantı başvurusu Başvuru belgeleri okunur, trafo kapasitesi ve mevzuat kontrol edilir, eksikler aynı gün bildirilir; bağlantı görüşü mühendisin. 5.400 çalıştırma / ay %95,3 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Ay sonu kapanış ve mutabakat (kiralık muhasebe ajanı) URL: https://www.aginies.com/tr/ajanlar/month-end-close-bookkeeping Language: tr Summary: Faturalar işlenir, banka ve tedarikçi mutabakatı yapılır, kapanış listesi yürütülür; kayıtlar mali müşavir onayıyla deftere işlenir. Ay sonu kapanış takviminin açılması → Kapanış görevleri tamamlandı, farklar açıklandı, onaylı kayıtlar deftere işlendi; iş kaydı ve fatura kalemi hazır Kurumsal Fonksiyonlarbatchemailapi # Ay sonu kapanış ve mutabakat (kiralık muhasebe ajanı) Faturalar işlenir, banka ve tedarikçi mutabakatı yapılır, kapanış listesi yürütülür; kayıtlar mali müşavir onayıyla deftere işlenir. Kısaca Ay sonu kapanış ve mutabakat (kiralık muhasebe ajanı), Kurumsal Fonksiyonlar sektöründe batch, email, api kanalında çalışan bir ajan senaryosudur. Tetikleyici: Ay sonu kapanış takviminin açılması. Muhasebe Süpervizörü 3 uzman ajanı (Belge Ajanı, Mutabakat Ajanı, Kapanış Ajanı) 4 sistem üzerinde yönetir (ERP (SAP / Logo), Bank API, e-Invoice Portal, Workforce Work Log); 9 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Kapanış görevleri tamamlandı, farklar açıklandı, onaylı kayıtlar deftere işlendi; iş kaydı ve fatura kalemi hazır. Onay kuralı: Deftere işlenen her kayıt ve beyanname mali müşavir onayıyla. Kurumsal Fonksiyonlar ### Ay sonu kapanış ve mutabakat (kiralık muhasebe ajanı) [Ay sonu kapanış ve mutabakat (kiralık muhasebe ajanı)] çalıştırma #41.289.910süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Deftere işlenen her kayıt ve beyanname mali müşavir onayıyla Bu çalıştırmadaki ajanlar Muhasebe Süpervizörüsupervisorplanlar, yönlendirir, onaylar Belge Ajanıaginies-konrul-1 · on-premFatura ve ekstreleri okur, kayıt önerir Mutabakat Ajanıaginies-tulpar-4 · on-premBanka, tedarikçi ve müşteri hesaplarını eşleştirir Kapanış Ajanıaginies-talay-2 · on-premKontrol listesini yürütür, kayıtları işler, raporlar Denetim izi00:00.00 - ◦Eylül kapanışı açıldı: 41 görev, 238 bekleyen fatura, 3 banka hesabı - ⚙238 e-fatura ve 3 ekstre alındı; alanlar çıkarıldı, 2 mükerrer adayı işaretlendi - ∿236 kayıt önerisi: hesap, maliyet merkezi, KDV oranı; 4 tanesi düşük güvenle işaretli - ⚙Banka hareketleri eşleştirildi: 1.184 / 1.191; 7 açık kalem - ∿7 fark açıklandı: 5 zamanlama, 1 banka masrafı, 1 tedarikçi çift ödeme adayı - ✓Kontrol listesi 38/41; kalan 3 görev onay bekliyor (tahakkuk, karşılık, çift ödeme) - ⏸Mali müşavir: 236 kaydı ve 2 düzeltmeyi onayladı, çift ödemeyi tedarikçiye iade olarak işaretledi - ⇢Kayıtlar deftere işlendi; KDV özeti ve kapanış raporu ERP’ye eklendi - ⇢İş kaydı: 236 kayıt · 3 mutabakat · 41 görev · saat karşılığı 27 · kalite 4,9 - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Ay sonu kapanış takviminin açılması Sonuç Kapanış görevleri tamamlandı, farklar açıklandı, onaylı kayıtlar deftere işlendi; iş kaydı ve fatura kalemi hazır ## Ajanlar SÜPERVİZÖR ### Muhasebe Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### Belge Ajanı Fatura ve ekstreleri okur, kayıt önerir extract_invoicefetch_bank_statementpropose_journal_entry aginies-tulpar-4 · on-prem ### Mutabakat Ajanı Banka, tedarikçi ve müşteri hesaplarını eşleştirir match_transactionsexplain_breaksget_vendor_ledger aginies-talay-2 · on-prem ### Kapanış Ajanı Kontrol listesini yürütür, kayıtları işler, raporlar run_checklistpost_entriesprepare_vat_summarylog_work ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Muhasebe Süpervizörü | Eylül kapanışı açıldı: 41 görev, 238 bekleyen fatura, 3 banka hesabı | ERP (SAP / Logo) | 200 | — | | 02 | araç | Belge Ajanı | 238 e-fatura ve 3 ekstre alındı; alanlar çıkarıldı, 2 mükerrer adayı işaretlendi | e-Invoice Portal | 1.400 | — | | 03 | akıl yürütme | Belge Ajanı | 236 kayıt önerisi: hesap, maliyet merkezi, KDV oranı; 4 tanesi düşük güvenle işaretli | — | 1.200 | 2.400 | | 04 | araç | Mutabakat Ajanı | Banka hareketleri eşleştirildi: 1.184 / 1.191; 7 açık kalem | Bank API | 900 | — | | 05 | akıl yürütme | Mutabakat Ajanı | 7 fark açıklandı: 5 zamanlama, 1 banka masrafı, 1 tedarikçi çift ödeme adayı | — | 800 | 900 | | 06 | doğrulama | Kapanış Ajanı | Kontrol listesi 38/41; kalan 3 görev onay bekliyor (tahakkuk, karşılık, çift ödeme) | — | 520 | 380 | | 07 | insan onayı | Muhasebe Süpervizörü | Mali müşavir: 236 kaydı ve 2 düzeltmeyi onayladı, çift ödemeyi tedarikçiye iade olarak işaretledi | — | 3.200 | — | | 08 | yazma | Kapanış Ajanı | Kayıtlar deftere işlendi; KDV özeti ve kapanış raporu ERP’ye eklendi | ERP (SAP / Logo) | 700 | — | | 09 | yazma | Kapanış Ajanı | İş kaydı: 236 kayıt · 3 mutabakat · 41 görev · saat karşılığı 27 · kalite 4,9 | Workforce Work Log | 300 | — | Simülasyon metrikleri (tahmini) 60çalıştırma / ay %96,2başarı 9,2 sp50 $0,0213tahmini maliyet / çalıştırma Onay kuralı Deftere işlenen her kayıt ve beyanname mali müşavir onayıyla Adım profili giriş · 1akıl yürütme · 2araç · 2doğrulama · 1insan onayı · 1yazma · 2 SİSTEMLER - ERP (SAP / Logo)erp - Bank APIpayments - e-Invoice Portalapi - Workforce Work Logdb Çalıştırmak için gerekenler - ERP/MES için kayıt ve sorgu API erişimi - Ödeme servisleri için yetkili API erişimi ve limitler - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Kiralık muhasebe ajanı modeli: dış kaynak muhasebeci yerine kapanışı yürüten, kayıt öneren ajan Bu senaryonun yer aldığı paketler Ajan kiralama (Agent Outsourcing) → İlgili ajanlar Kurumsal FonksiyonlarHITL ### Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada Talep e-postayla gelir; teklifler karşılaştırılır, sözleşme maddeleri kontrol edilir, sipariş açılır, fatura üç yönlü eşleşir ve ödeme planlanır. 14.000 çalıştırma / ay %92,7 başarı 5 ajan Kurumsal Fonksiyonlar ### E-posta sınıflandırma ve ek kontrolü Ortak kutuya gelen her e-posta sınıflandırılır, ekler listeyle karşılaştırılır, eksikler anında sorulur. 86.000 çalıştırma / ay %97,6 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Çağrı merkezi kalite izleme Örneklem değil, tüm çağrılar rubriğe göre puanlanır; uyum ihlali ve koçluk fırsatı sabah liderin önünde. 90.000 çalıştırma / ay %98,8 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Konut kredisi belge incelemesi URL: https://www.aginies.com/tr/ajanlar/mortgage-document-review Language: tr Summary: Bordro, SGK dökümü ve ekspertiz otomatik okunur; gelir, LTV ve eksik belge kontrolü danışman beklemeden biter. Kredi dosyasına belge yüklendi (web, e-posta veya şube) → Dosya ön onaylı, eksik belge listesi müşteride, gelir ve LTV kanıtları dosyada Bankacılıkwebemailbranch # Konut kredisi belge incelemesi Bordro, SGK dökümü ve ekspertiz otomatik okunur; gelir, LTV ve eksik belge kontrolü danışman beklemeden biter. Kısaca Konut kredisi belge incelemesi, Bankacılık sektöründe web, email, branch kanalında çalışan bir ajan senaryosudur. Tetikleyici: Kredi dosyasına belge yüklendi (web, e-posta veya şube). Kredi Dosya Süpervizörü 3 uzman ajanı (Çıkarım Ajanı, Gelir Ajanı, Değerleme Ajanı) 4 sistem üzerinde yönetir (Loan Origination System, Document Store (S3), Credit Bureau (KKB), Email (Exchange)); 10 adımın 1 tanesi sistemde işlem, 3 tanesi doğrulama veya onaydır. Sonuç: Dosya ön onaylı, eksik belge listesi müşteride, gelir ve LTV kanıtları dosyada. Onay kuralı: Kredi tutarı > 3.000.000 TL veya DTI > %45 → kredi tahsis uzmanı onayı. Bankacılık ### Konut kredisi belge incelemesi [Konut kredisi belge incelemesi] çalıştırma #41.289.484süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Kredi tutarı > 3.000.000 TL veya DTI > %45 → kredi tahsis uzmanı onayı Bu çalıştırmadaki ajanlar Kredi Dosya Süpervizörüsupervisorplanlar, yönlendirir, onaylar Çıkarım Ajanıaginies-konrul-1 · on-premBelgeleri sınıflar, alanları ve tabloları çıkarır Gelir Ajanıaginies-talay-2 · on-premNet geliri hesaplar, SGK ve KKB verisiyle çapraz kontrol eder Değerleme Ajanıaginies-tulpar-4 · on-premEkspertizi okur, LTV'yi ve eksik belgeleri belirler Denetim izi00:00.00 - ◦Dosya #ML-7731: 9 belge yüklendi (bordro, SGK dökümü, ekspertiz, tapu) - ⌕Belgeler S3'ten alındı, sınıflandı; 3 bordro ve SGK dökümü alanları çıkarıldı - ⌘Plan: gelir doğrulama → değerleme/LTV → eksik belge → dosya güncelle - ⚙KKB raporu: skor 1.612, mevcut taksit yükü 9.800 TL/ay - ✓Net gelir 68.500 TL, SGK ile %2 içinde tutarlı, DTI %41 → uygun - ∿Ekspertiz 6.400.000 TL, talep 4.800.000 TL → LTV %75, sınır %80 - ✓Eksik: DASK poliçesi ve son 3 ay hesap özeti → 2 belge - ⏸Kredi > 3.000.000 TL → kredi tahsis uzmanı dosyayı ön onayladı - ⇢LOS'ta dosya "ön onay – belge bekleniyor" durumuna alındı, kontrol listesi eklendi - ➤Müşteriye ve şube danışmanına eksik belge e-postası gönderildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Kredi dosyasına belge yüklendi (web, e-posta veya şube) Sonuç Dosya ön onaylı, eksik belge listesi müşteride, gelir ve LTV kanıtları dosyada ## Ajanlar SÜPERVİZÖR ### Kredi Dosya Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### Çıkarım Ajanı Belgeleri sınıflar, alanları ve tabloları çıkarır classify_documentextract_fieldsocr_page aginies-talay-2 · on-prem ### Gelir Ajanı Net geliri hesaplar, SGK ve KKB verisiyle çapraz kontrol eder compute_net_incomecross_check_sgkget_credit_bureaucompute_dti aginies-tulpar-4 · on-prem ### Değerleme Ajanı Ekspertizi okur, LTV'yi ve eksik belgeleri belirler parse_appraisalcheck_ltvlist_missing_docsupdate_loan_filesend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Kredi Dosya Süpervizörü | Dosya #ML-7731: 9 belge yüklendi (bordro, SGK dökümü, ekspertiz, tapu) | — | 240 | — | | 02 | bilgi | Çıkarım Ajanı | Belgeler S3'ten alındı, sınıflandı; 3 bordro ve SGK dökümü alanları çıkarıldı | Document Store (S3) | 1.400 | 2.600 | | 03 | plan | Kredi Dosya Süpervizörü | Plan: gelir doğrulama → değerleme/LTV → eksik belge → dosya güncelle | — | 300 | 200 | | 04 | araç | Gelir Ajanı | KKB raporu: skor 1.612, mevcut taksit yükü 9.800 TL/ay | Credit Bureau (KKB) | 480 | — | | 05 | doğrulama | Gelir Ajanı | Net gelir 68.500 TL, SGK ile %2 içinde tutarlı, DTI %41 → uygun | — | 760 | 820 | | 06 | akıl yürütme | Değerleme Ajanı | Ekspertiz 6.400.000 TL, talep 4.800.000 TL → LTV %75, sınır %80 | — | 720 | 690 | | 07 | doğrulama | Değerleme Ajanı | Eksik: DASK poliçesi ve son 3 ay hesap özeti → 2 belge | — | 520 | 430 | | 08 | insan onayı | Kredi Dosya Süpervizörü | Kredi > 3.000.000 TL → kredi tahsis uzmanı dosyayı ön onayladı | — | 3.100 | — | | 09 | yazma | Değerleme Ajanı | LOS'ta dosya "ön onay – belge bekleniyor" durumuna alındı, kontrol listesi eklendi | Loan Origination System | 460 | — | | 10 | bildirim | Değerleme Ajanı | Müşteriye ve şube danışmanına eksik belge e-postası gönderildi | Email (Exchange) | 260 | — | Simülasyon metrikleri (tahmini) 6.200çalıştırma / ay %94,7başarı 8,3 sp50 $0,0111tahmini maliyet / çalıştırma Onay kuralı Kredi tutarı > 3.000.000 TL veya DTI > %45 → kredi tahsis uzmanı onayı Adım profili giriş · 1plan · 1bilgi · 1akıl yürütme · 1araç · 1doğrulama · 2insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - Loan Origination Systemcore - Document Store (S3)storage - Credit Bureau (KKB)api - Email (Exchange)email Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Doküman deposuna okuma erişimi ve belge tipleri - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek ABD/İngiltere ipotek kredisi süreçlerindeki belge sınıflandırma ve gelir doğrulama otomasyonu deseni Bu senaryonun yer aldığı paketler KVKK kişisel veri koruma →Kredi ve müşteri edinimi operasyonları → Sektör: Bankacılık → İlgili ajanlar BankacılıkHITL ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı Müşteri konuşur ya da yazar; fatura, limit, itiraz ve transfer aynı oturumda çekirdek sistemde tamamlanır. 186.000 çalıştırma / ay %94,7 başarı 5 ajan BankacılıkHITL ### Kredi kartı limit artışı Müşteri mobilde yazar ya da çağrı merkezinde konuşur; karar saniyeler içinde verilir, eşik üstü yönetici onayına gider. 18.400 çalıştırma / ay %96,4 başarı 4 ajan BankacılıkHITL ### Dijital müşteri edinimi (KYC) Kimlik, canlılık ve yaptırım kontrolü tek akışta; temiz başvurular şubeye gitmeden dakikalarda hesap açar. 42.000 çalıştırma / ay %93,8 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Uçtan uca kasko hasarı: ihbardan ödemeye URL: https://www.aginies.com/tr/ajanlar/motor-claim-end-to-end Language: tr Summary: Sesli ihbar, fotoğraf ekspertizi, anlaşmalı servis ataması ve ödeme talimatı tek akışta; eksper yalnızca eşik üstünde devreye girer. Çağrı merkezine gelen hasar ihbarı veya uygulamadan fotoğraflı bildirim → Dosya açıldı, hasar tutarı belirlendi, servis atandı, ödeme talimatı verildi Sigortavoiceapp # Uçtan uca kasko hasarı: ihbardan ödemeye Sesli ihbar, fotoğraf ekspertizi, anlaşmalı servis ataması ve ödeme talimatı tek akışta; eksper yalnızca eşik üstünde devreye girer. Kısaca Uçtan uca kasko hasarı: ihbardan ödemeye, Sigorta sektöründe voice, app kanalında çalışan bir ajan senaryosudur. Tetikleyici: Çağrı merkezine gelen hasar ihbarı veya uygulamadan fotoğraflı bildirim. Hasar Süpervizörü 4 uzman ajanı (İhbar Ajanı, Foto Ekspertiz Ajanı, Dosya Ajanı, İletişim Ajanı) 5 sistem üzerinde yönetir (Call Centre SIP, Policy Admin, Claims System, Repair Network, Payment Hub); 10 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Dosya açıldı, hasar tutarı belirlendi, servis atandı, ödeme talimatı verildi. Onay kuralı: Tahmin > 150.000 TL, kusur oranı belirsiz veya önceki hasarla çakışma → eksper onayı. Sigorta ### Uçtan uca kasko hasarı: ihbardan ödemeye [Uçtan uca kasko hasarı: ihbardan ödemeye] çalıştırma #41.290.297süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Tahmin > 150.000 TL, kusur oranı belirsiz veya önceki hasarla çakışma → eksper onayı Bu çalıştırmadaki ajanlar Hasar Süpervizörüsupervisorplanlar, yönlendirir, onaylar İhbar Ajanıaginies-voice-tr-v3 · on-premSesli ihbarı yazıya döker, olay bilgilerini toplar, poliçeyi eşler Foto Ekspertiz Ajanıaginies-konrul-1 · on-premHasar fotoğraflarından parça ve işçilik tahmini üretir Dosya Ajanıaginies-tulpar-4 · on-premTeminat, muafiyet ve kusur oranını değerlendirir, servisi atar, ödemeyi hazırlar İletişim Ajanıaginies-sunkar-1 · on-premMüşteriyi ve servisi her adımda bilgilendirir Denetim izi00:00.00 - ◦Çağrı: "Otoparkta arkadan çarptılar, karşı taraf kabul etti, fotoğrafları uygulamaya yükledim" - ∿Transkript ve olay formu: tarih, yer, karşı taraf plakası, yaralı yok; poliçe 34 ABC 123 aktif - ⌘Plan: foto ekspertiz → teminat ve muafiyet → kusur → servis → ödeme → rücu - ∿6 fotoğraf: arka tampon değişim, bagaj kapağı onarım, park sensörü 2 adet; tahmin 38.400 TL - ✓Önceki hasar kaydıyla karşılaştırıldı: çakışan bölge yok, fotoğraf meta verisi tutarlı - ✓Teminat kasko tam, muafiyet 0, karşı taraf %100 kusurlu (tutanak) → eksik belge yok - ⚙Anlaşmalı servis: 4 km mesafede, kapasite 2 gün, parça stokta; randevu önerildi - ⇢Dosya #CL-230517: tutar 38.400 TL onaylandı, servis atandı, rücu dosyası açıldı - ⇢Ödeme talimatı servise işçilik tamamlanınca; parça avansı 14.200 TL bugün - ➤Müşteriye sesli özet ve WhatsApp: randevu, ikame araç, dosya takip bağlantısı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Çağrı merkezine gelen hasar ihbarı veya uygulamadan fotoğraflı bildirim Sonuç Dosya açıldı, hasar tutarı belirlendi, servis atandı, ödeme talimatı verildi ## Ajanlar SÜPERVİZÖR ### Hasar Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-voice-tr-v3 · on-prem ### İhbar Ajanı Sesli ihbarı yazıya döker, olay bilgilerini toplar, poliçeyi eşler transcribe_streamcollect_fnolmatch_policy aginies-konrul-1 · on-prem ### Foto Ekspertiz Ajanı Hasar fotoğraflarından parça ve işçilik tahmini üretir detect_damageestimate_repaircheck_prior_damage aginies-tulpar-4 · on-prem ### Dosya Ajanı Teminat, muafiyet ve kusur oranını değerlendirir, servisi atar, ödemeyi hazırlar check_coverageassign_repair_shopprepare_paymentopen_subrogation aginies-sunkar-1 · on-prem ### İletişim Ajanı Müşteriyi ve servisi her adımda bilgilendirir send_smssend_whatsappnotify_shop ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Hasar Süpervizörü | Çağrı: "Otoparkta arkadan çarptılar, karşı taraf kabul etti, fotoğrafları uygulamaya yükledim" | Call Centre SIP | 220 | — | | 02 | akıl yürütme | İhbar Ajanı | Transkript ve olay formu: tarih, yer, karşı taraf plakası, yaralı yok; poliçe 34 ABC 123 aktif | Policy Admin | 1.400 | 520 | | 03 | plan | Hasar Süpervizörü | Plan: foto ekspertiz → teminat ve muafiyet → kusur → servis → ödeme → rücu | — | 260 | 240 | | 04 | akıl yürütme | Foto Ekspertiz Ajanı | 6 fotoğraf: arka tampon değişim, bagaj kapağı onarım, park sensörü 2 adet; tahmin 38.400 TL | — | 2.600 | 1.900 | | 05 | doğrulama | Foto Ekspertiz Ajanı | Önceki hasar kaydıyla karşılaştırıldı: çakışan bölge yok, fotoğraf meta verisi tutarlı | Claims System | 700 | 380 | | 06 | doğrulama | Dosya Ajanı | Teminat kasko tam, muafiyet 0, karşı taraf %100 kusurlu (tutanak) → eksik belge yok | Policy Admin | 820 | 460 | | 07 | araç | Dosya Ajanı | Anlaşmalı servis: 4 km mesafede, kapasite 2 gün, parça stokta; randevu önerildi | Repair Network | 540 | — | | 08 | yazma | Dosya Ajanı | Dosya #CL-230517: tutar 38.400 TL onaylandı, servis atandı, rücu dosyası açıldı | Claims System | 620 | — | | 09 | yazma | Dosya Ajanı | Ödeme talimatı servise işçilik tamamlanınca; parça avansı 14.200 TL bugün | Payment Hub | 480 | — | | 10 | bildirim | İletişim Ajanı | Müşteriye sesli özet ve WhatsApp: randevu, ikame araç, dosya takip bağlantısı | Call Centre SIP | 440 | 200 | Simülasyon metrikleri (tahmini) 21.000çalıştırma / ay %91,2başarı 9,6 sp50 $0,0062tahmini maliyet / çalıştırma Onay kuralı Tahmin > 150.000 TL, kusur oranı belirsiz veya önceki hasarla çakışma → eksper onayı Adım profili giriş · 1plan · 1akıl yürütme · 2araç · 1doğrulama · 2yazma · 2bildirim · 1 SİSTEMLER - Call Centre SIPvoice - Policy Admincore - Claims Systemcore - Repair Networkapi - Payment Hubpayments Çalıştırmak için gerekenler - Santral ile SIP entegrasyonu ve kayıt erişimi - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Ödeme servisleri için yetkili API erişimi ve limitler - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Dijital sigortacıların "dakikalar içinde hasar" deseni; fotoğraf ekspertizi ve servis ağıyla birleştirilmiş Bu senaryonun yer aldığı paketler Sesli AI ile hasar ve talep kaydı →Hasar operasyonları uçtan uca → Sektör: Sigorta → İlgili ajanlar SigortaHITL ### Sesli hasar ihbarı (FNOL) Müşteri arar ve anlatır; poliçe plakadan bulunur, teminat kontrol edilir, dosya açılır ve eksper atanır. 34.000 çalıştırma / ay %95,8 başarı 5 ajan SigortaHITL ### Poliçe yenileme teklifi Poliçeler yeniden fiyatlanır, sert artışlar onaya gider; müşteri tek dokunuşla onaylar, poliçe anında kesilir. 22.000 çalıştırma / ay %94,4 başarı 4 ajan SigortaHITL ### Underwriting belge çıkarımı Sağlık raporu ve tahliller okunur, kılavuza göre değerlendirilir; underwriter yalnızca eşik üstünü görür. 4.800 çalıştırma / ay %94,9 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Fotoğrafla kasko hasar tespiti URL: https://www.aginies.com/tr/ajanlar/motor-claim-photo-assessment Language: tr Summary: Fotoğraflardan hasar tespit edilir, parça/işçilikle tahmin çıkar, eksper onaylar; servis randevusu alınır. Açık hasar dosyasına WhatsApp veya uygulamadan fotoğraf yüklenmesi → Onaylı tahmin ve rezerv dosyada, servis randevusu alındı, müşteri bilgilendirildi Sigortaappwhatsapp # Fotoğrafla kasko hasar tespiti Fotoğraflardan hasar tespit edilir, parça/işçilikle tahmin çıkar, eksper onaylar; servis randevusu alınır. Kısaca Fotoğrafla kasko hasar tespiti, Sigorta sektöründe app, whatsapp kanalında çalışan bir ajan senaryosudur. Tetikleyici: Açık hasar dosyasına WhatsApp veya uygulamadan fotoğraf yüklenmesi. Hasar Süpervizörü 3 uzman ajanı (Görüntü Ajanı, Ekspertiz Ajanı, Servis Ajanı) 4 sistem üzerinde yönetir (Guidewire ClaimCenter, Parts & Labour Catalogue (Audatex), Repair Network Portal, WhatsApp Business API); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Onaylı tahmin ve rezerv dosyada, servis randevusu alındı, müşteri bilgilendirildi. Onay kuralı: Tahmin > 15.000 TL veya model güveni < %80 → eksper onayı. Sigorta ### Fotoğrafla kasko hasar tespiti [Fotoğrafla kasko hasar tespiti] çalıştırma #41.287.160süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Tahmin > 15.000 TL veya model güveni < %80 → eksper onayı Bu çalıştırmadaki ajanlar Hasar Süpervizörüsupervisorplanlar, yönlendirir, onaylar Görüntü Ajanıaginies-konrul-1 · on-premFotoğraf kalitesini ve hasarı değerlendirir, sahteciliği kontrol eder Ekspertiz Ajanıaginies-tulpar-4 · on-premParça ve işçilik fiyatlarıyla tahmini oluşturur Servis Ajanıaginies-talay-2 · on-premAnlaşmalı servisi bulur, randevu alır, dosyayı günceller Denetim izi00:00.00 - ◦Dosya #CL-2026-118342 için WhatsApp'tan 6 fotoğraf geldi - ✓Görüntü kalitesi yeterli, EXIF tarihi olayla uyumlu, tekrar kullanım yok - ∿Hasar: sağ ön çamurluk + tampon, derinlik orta, far sağlam - ⌘Plan: parça/işçilik → tahmin → eşik kontrolü → servis yönlendirme - ⚙Fiyatlar: çamurluk 4.200 TL, tampon 6.800 TL, boya 3,5 saat - ∿Tahmini hasar 19.600 TL (± %12), pert değil, muafiyet 2.000 TL - ⏸Tahmin > 15.000 TL → eksper 3 dakikada onayladı, 1 kalem düzeltildi - ⚙5 km içinde 3 anlaşmalı servis, en erken slot yarın 14:00 - ⇢Dosyaya tahmin ve servis randevusu yazıldı, rezerv 19.600 TL - ➤Müşteriye WhatsApp: tahmin özeti, servis adresi ve randevu - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Açık hasar dosyasına WhatsApp veya uygulamadan fotoğraf yüklenmesi Sonuç Onaylı tahmin ve rezerv dosyada, servis randevusu alındı, müşteri bilgilendirildi ## Ajanlar SÜPERVİZÖR ### Hasar Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### Görüntü Ajanı Fotoğraf kalitesini ve hasarı değerlendirir, sahteciliği kontrol eder check_image_qualitydetect_damageestimate_severitydetect_reuse aginies-tulpar-4 · on-prem ### Ekspertiz Ajanı Parça ve işçilik fiyatlarıyla tahmini oluşturur get_parts_pricesget_labour_ratesbuild_estimate aginies-talay-2 · on-prem ### Servis Ajanı Anlaşmalı servisi bulur, randevu alır, dosyayı günceller find_repair_shopsbook_slotupdate_claimsend_whatsapp ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Hasar Süpervizörü | Dosya #CL-2026-118342 için WhatsApp'tan 6 fotoğraf geldi | — | 240 | — | | 02 | doğrulama | Görüntü Ajanı | Görüntü kalitesi yeterli, EXIF tarihi olayla uyumlu, tekrar kullanım yok | — | 620 | 540 | | 03 | akıl yürütme | Görüntü Ajanı | Hasar: sağ ön çamurluk + tampon, derinlik orta, far sağlam | — | 1.200 | 1.650 | | 04 | plan | Hasar Süpervizörü | Plan: parça/işçilik → tahmin → eşik kontrolü → servis yönlendirme | — | 280 | 190 | | 05 | araç | Ekspertiz Ajanı | Fiyatlar: çamurluk 4.200 TL, tampon 6.800 TL, boya 3,5 saat | Parts & Labour Catalogue (Audatex) | 460 | — | | 06 | akıl yürütme | Ekspertiz Ajanı | Tahmini hasar 19.600 TL (± %12), pert değil, muafiyet 2.000 TL | — | 780 | 720 | | 07 | insan onayı | Hasar Süpervizörü | Tahmin > 15.000 TL → eksper 3 dakikada onayladı, 1 kalem düzeltildi | — | 2.900 | — | | 08 | araç | Servis Ajanı | 5 km içinde 3 anlaşmalı servis, en erken slot yarın 14:00 | Repair Network Portal | 380 | — | | 09 | yazma | Servis Ajanı | Dosyaya tahmin ve servis randevusu yazıldı, rezerv 19.600 TL | Guidewire ClaimCenter | 520 | — | | 10 | bildirim | Servis Ajanı | Müşteriye WhatsApp: tahmin özeti, servis adresi ve randevu | WhatsApp Business API | 260 | — | Simülasyon metrikleri (tahmini) 15.500çalıştırma / ay %93,3başarı 7,6 sp50 $0,0053tahmini maliyet / çalıştırma Onay kuralı Tahmin > 15.000 TL veya model güveni < %80 → eksper onayı Adım profili giriş · 1plan · 1akıl yürütme · 2araç · 2doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - Guidewire ClaimCentercore - Parts & Labour Catalogue (Audatex)api - Repair Network Portalapi - WhatsApp Business APIchannel Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek ABD/AB kasko sigortacılarının yapay görü tabanlı "anında tahmin" deseni Bu senaryonun yer aldığı paketler Hasar operasyonları uçtan uca → Sektör: Sigorta → İlgili ajanlar SigortaHITL ### Uçtan uca kasko hasarı: ihbardan ödemeye Sesli ihbar, fotoğraf ekspertizi, anlaşmalı servis ataması ve ödeme talimatı tek akışta; eksper yalnızca eşik üstünde devreye girer. 21.000 çalıştırma / ay %91,2 başarı 5 ajan SigortaHITL ### Sesli hasar ihbarı (FNOL) Müşteri arar ve anlatır; poliçe plakadan bulunur, teminat kontrol edilir, dosya açılır ve eksper atanır. 34.000 çalıştırma / ay %95,8 başarı 5 ajan SigortaHITL ### Poliçe yenileme teklifi Poliçeler yeniden fiyatlanır, sert artışlar onaya gider; müşteri tek dokunuşla onaylar, poliçe anında kesilir. 22.000 çalıştırma / ay %94,4 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Şebeke kesintisi iletişimi URL: https://www.aginies.com/tr/ajanlar/network-outage-communications Language: tr Summary: Kesinti kaydı açılınca etkilenen aboneler bulunur, mesaj yazılıp onaylanır; müşteri aramadan bilgilendirilir. ServiceNow'da P1/P2 şebeke kesintisi kaydı → Etkilenen aboneler bilgilendirildi, kayıt güncellendi, "düzeldi" mesajı planlandı Telekomsmswhatsappapibatch # Şebeke kesintisi iletişimi Kesinti kaydı açılınca etkilenen aboneler bulunur, mesaj yazılıp onaylanır; müşteri aramadan bilgilendirilir. Kısaca Şebeke kesintisi iletişimi, Telekom sektöründe sms, whatsapp, api, batch kanalında çalışan bir ajan senaryosudur. Tetikleyici: ServiceNow'da P1/P2 şebeke kesintisi kaydı. Kesinti İletişim Süpervizörü 3 uzman ajanı (Tespit Ajanı, Etki Ajanı, İletişim Ajanı) 4 sistem üzerinde yönetir (ServiceNow, Network OSS, Salesforce CRM, SMS Gateway); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Etkilenen aboneler bilgilendirildi, kayıt güncellendi, "düzeldi" mesajı planlandı. Onay kuralı: Etkilenen abone > 10.000 veya P1 kesinti → şebeke operasyon yöneticisi onayı. Telekom ### Şebeke kesintisi iletişimi [Şebeke kesintisi iletişimi] çalıştırma #41.287.584süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Etkilenen abone > 10.000 veya P1 kesinti → şebeke operasyon yöneticisi onayı Bu çalıştırmadaki ajanlar Kesinti İletişim Süpervizörüsupervisorplanlar, yönlendirir, onaylar Tespit Ajanıaginies-talay-2 · on-premKesintinin nedenini, hücrelerini ve onarım süresini alır Etki Ajanıaginies-sunkar-1 · on-premHücreleri abonelere eşler, segment ve kanal tercihi çıkarır İletişim Ajanıaginies-tulpar-4 · on-premMesajı yazar, guardrail kontrolü yapar, toplu gönderir Denetim izi00:00.00 - ◦ServiceNow INC-204188: Bursa Nilüfer'de 14 baz istasyonu erişilemez, P1 - ⚙OSS: fiber kesintisi, 14 hücre, tahmini onarım 3 saat 20 dakika - ⌘Plan: etkilenen aboneler → segment → mesaj → onay → gönderim → güncelleme - ⚙38.400 etkilenen abone; 2.150 kurumsal, 310 SLA'lı hat - ∿Segment: kurumsal → hesap yöneticisi + SMS; bireysel → tercihe göre SMS/WhatsApp - ∿Mesaj taslağı: neden, etkilenen bölge, tahmini süre, özür; 148 karakter - ✓Guardrail: kişisel veri yok, SLA taahhüdü verilmiyor, marka tonuna uygun - ⏸> 10.000 abone → şebeke operasyon yöneticisi 4 dakikada onayladı - ⇢INC-204188 "müşteriler bilgilendirildi" olarak güncellendi, durum sayfası yayında - ➤36.250 SMS + 2.150 WhatsApp gönderildi; onarım sonrası "düzeldi" mesajı planlandı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici ServiceNow'da P1/P2 şebeke kesintisi kaydı Sonuç Etkilenen aboneler bilgilendirildi, kayıt güncellendi, "düzeldi" mesajı planlandı ## Ajanlar SÜPERVİZÖR ### Kesinti İletişim Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Tespit Ajanı Kesintinin nedenini, hücrelerini ve onarım süresini alır get_incidentget_affected_cellsestimate_restoration aginies-sunkar-1 · on-prem ### Etki Ajanı Hücreleri abonelere eşler, segment ve kanal tercihi çıkarır map_cells_to_subscriberssegment_customersget_contact_preferences aginies-tulpar-4 · on-prem ### İletişim Ajanı Mesajı yazar, guardrail kontrolü yapar, toplu gönderir draft_messagecheck_guardrailssend_sms_batchsend_whatsapp_batchupdate_incident ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Kesinti İletişim Süpervizörü | ServiceNow INC-204188: Bursa Nilüfer'de 14 baz istasyonu erişilemez, P1 | — | 220 | — | | 02 | araç | Tespit Ajanı | OSS: fiber kesintisi, 14 hücre, tahmini onarım 3 saat 20 dakika | Network OSS | 460 | — | | 03 | plan | Kesinti İletişim Süpervizörü | Plan: etkilenen aboneler → segment → mesaj → onay → gönderim → güncelleme | — | 260 | 200 | | 04 | araç | Etki Ajanı | 38.400 etkilenen abone; 2.150 kurumsal, 310 SLA'lı hat | Salesforce CRM | 720 | — | | 05 | akıl yürütme | Etki Ajanı | Segment: kurumsal → hesap yöneticisi + SMS; bireysel → tercihe göre SMS/WhatsApp | — | 540 | 460 | | 06 | akıl yürütme | İletişim Ajanı | Mesaj taslağı: neden, etkilenen bölge, tahmini süre, özür; 148 karakter | — | 620 | 540 | | 07 | doğrulama | İletişim Ajanı | Guardrail: kişisel veri yok, SLA taahhüdü verilmiyor, marka tonuna uygun | — | 380 | 300 | | 08 | insan onayı | Kesinti İletişim Süpervizörü | > 10.000 abone → şebeke operasyon yöneticisi 4 dakikada onayladı | — | 2.800 | — | | 09 | yazma | İletişim Ajanı | INC-204188 "müşteriler bilgilendirildi" olarak güncellendi, durum sayfası yayında | ServiceNow | 420 | — | | 10 | bildirim | İletişim Ajanı | 36.250 SMS + 2.150 WhatsApp gönderildi; onarım sonrası "düzeldi" mesajı planlandı | SMS Gateway | 900 | — | Simülasyon metrikleri (tahmini) 1.400çalıştırma / ay %98,5başarı 7,3 sp50 $0,0075tahmini maliyet / çalıştırma Onay kuralı Etkilenen abone > 10.000 veya P1 kesinti → şebeke operasyon yöneticisi onayı Adım profili giriş · 1plan · 1akıl yürütme · 2araç · 2doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - ServiceNowticketing - Network OSSapi - Salesforce CRMcrm - SMS Gatewaychannel Çalıştırmak için gerekenler - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - CRM için API erişimi ve müşteri eşleme anahtarı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük operatörlerin proaktif kesinti bildirimiyle gelen çağrı hacmini yaklaşık %30 azaltan deseni Bu senaryonun yer aldığı paketler Abone hizmetleri ve şebeke desteği → Sektör: Telekom → İlgili ajanlar TelekomHITL ### Abone hizmet asistanı: fatura, paket, arıza ve taahhüt Abone arar ya da yazar; fatura itirazı, paket değişikliği ve hat arızası aynı oturumda BSS ve şebeke sistemlerinde çözülür. 410.000 çalıştırma / ay %94,3 başarı 5 ajan TelekomHITL ### Fatura itirazı çözümü Fatura kalemleri ve kullanım kayıtları saniyede açıklanır; iyi niyet kredisi politikaya göre uygulanır. 88.000 çalıştırma / ay %96,3 başarı 4 ajan TelekomHITL ### SIM değişimi dolandırıcılık kontrolü Her SIM değişimi provizyondan önce risk skorlanır; şüpheli talepler bekletilir ve bankalara sinyal iletilir. 41.000 çalıştırma / ay %97,7 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Siparişim nerede? (WISMO) URL: https://www.aginies.com/tr/ajanlar/order-status-wismo Language: tr Summary: Müşteri sorar, sipariş ve kargo olayları anında birleşir; gecikmede yeni tarih verilir, kargo araştırılır. WhatsApp, web veya uygulamada "siparişim nerede?" mesajı → Müşteri güncel durumu ve yeni tarihi biliyor, kargoda araştırma açıldı, takip planlandı Perakende & E-ticaretwhatsappwebappvoice # Siparişim nerede? (WISMO) Müşteri sorar, sipariş ve kargo olayları anında birleşir; gecikmede yeni tarih verilir, kargo araştırılır. Kısaca Siparişim nerede? (WISMO), Perakende & E-ticaret sektöründe whatsapp, web, app, voice kanalında çalışan bir ajan senaryosudur. Tetikleyici: WhatsApp, web veya uygulamada "siparişim nerede?" mesajı. Sipariş Süpervizörü 3 uzman ajanı (Niyet Ajanı, Takip Ajanı, Yanıt Ajanı) 4 sistem üzerinde yönetir (Shopify / OMS, Carrier API (Yurtiçi, DHL), WhatsApp Business API, Zendesk); 9 adımın 1 tanesi sistemde işlem, 1 tanesi doğrulama veya onaydır. Sonuç: Müşteri güncel durumu ve yeni tarihi biliyor, kargoda araştırma açıldı, takip planlandı. Perakende & E-ticaret ### Siparişim nerede? (WISMO) [Siparişim nerede? (WISMO)] çalıştırma #41.287.774süre 0 mstoken 0tahmini maliyet $0,000durum — Bu çalıştırmadaki ajanlar Sipariş Süpervizörüsupervisorplanlar, yönlendirir, onaylar Niyet Ajanıaginies-sunkar-1 · on-premTalebi anlar, sipariş numarasını ve müşteriyi doğrular Takip Ajanıaginies-talay-2 · on-premSipariş ve kargo olaylarını birleştirir, yeni teslimat tarihi tahmin eder Yanıt Ajanıaginies-tulpar-4 · on-premPolitikayı uygular, yanıtı yazar, gerekirse talep açar Denetim izi00:00.00 - ◦WhatsApp: "siparişim 5 gündür gelmedi, numarası 100482" - ∿Niyet: WISMO · sipariş #100482 · telefon eşleşti → doğrulandı - ⚙Sipariş 3 kalem, 02.09 kargoya verildi, takip no YK-7731 8820 - ⚙Kargo olayları: 04.09'dan beri transfer merkezinde, 2 gündür hareket yok - ⌘Plan: gecikme sınıfı → yeni tarih → politika → kargo talebi → yanıt - ∿Gecikme taşıyıcı kaynaklı; tahmini yeni teslimat 10.09, güven %82 - ✓Politika: gecikme > 7 gün → 100 TL kupon; bu vaka 5 gün → yalnızca bilgi - ⇢Zendesk'te kargo araştırma talebi #ZD-51937 açıldı, taşıyıcıya iletildi - ➤WhatsApp: durum, yeni tarih ve canlı takip linki; 10.09'da otomatik kontrol - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici WhatsApp, web veya uygulamada "siparişim nerede?" mesajı Sonuç Müşteri güncel durumu ve yeni tarihi biliyor, kargoda araştırma açıldı, takip planlandı ## Ajanlar SÜPERVİZÖR ### Sipariş Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Niyet Ajanı Talebi anlar, sipariş numarasını ve müşteriyi doğrular detect_intentextract_order_refverify_customer aginies-talay-2 · on-prem ### Takip Ajanı Sipariş ve kargo olaylarını birleştirir, yeni teslimat tarihi tahmin eder get_orderget_shipmentget_carrier_eventsestimate_eta aginies-tulpar-4 · on-prem ### Yanıt Ajanı Politikayı uygular, yanıtı yazar, gerekirse talep açar check_delay_policycompose_replysend_whatsappcreate_ticket ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Sipariş Süpervizörü | WhatsApp: "siparişim 5 gündür gelmedi, numarası 100482" | — | 180 | — | | 02 | akıl yürütme | Niyet Ajanı | Niyet: WISMO · sipariş #100482 · telefon eşleşti → doğrulandı | — | 420 | 260 | | 03 | araç | Takip Ajanı | Sipariş 3 kalem, 02.09 kargoya verildi, takip no YK-7731 8820 | Shopify / OMS | 340 | — | | 04 | araç | Takip Ajanı | Kargo olayları: 04.09'dan beri transfer merkezinde, 2 gündür hareket yok | Carrier API (Yurtiçi, DHL) | 460 | — | | 05 | plan | Sipariş Süpervizörü | Plan: gecikme sınıfı → yeni tarih → politika → kargo talebi → yanıt | — | 240 | 170 | | 06 | akıl yürütme | Takip Ajanı | Gecikme taşıyıcı kaynaklı; tahmini yeni teslimat 10.09, güven %82 | — | 520 | 430 | | 07 | doğrulama | Yanıt Ajanı | Politika: gecikme > 7 gün → 100 TL kupon; bu vaka 5 gün → yalnızca bilgi | — | 380 | 290 | | 08 | yazma | Yanıt Ajanı | Zendesk'te kargo araştırma talebi #ZD-51937 açıldı, taşıyıcıya iletildi | Zendesk | 400 | — | | 09 | bildirim | Yanıt Ajanı | WhatsApp: durum, yeni tarih ve canlı takip linki; 10.09'da otomatik kontrol | WhatsApp Business API | 300 | 240 | Simülasyon metrikleri (tahmini) 110.000çalıştırma / ay %98,1başarı 3,3 sp50 $0,0006tahmini maliyet / çalıştırma Adım profili giriş · 1plan · 1akıl yürütme · 2araç · 2doğrulama · 1yazma · 1bildirim · 1 SİSTEMLER - Shopify / OMSerp - Carrier API (Yurtiçi, DHL)api - WhatsApp Business APIchannel - Zendeskticketing Çalıştırmak için gerekenler - ERP/MES için kayıt ve sorgu API erişimi - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük e-ticaret oyuncularının WISMO taleplerinin %60'ından fazlasını self-servise çeviren proaktif bildirim deseni Bu senaryonun yer aldığı paketler E-ticaret müşteri desteği ve aksiyon → Sektör: Perakende & E-ticaret → İlgili ajanlar Perakende & E-ticaretHITL ### Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin Müşteri bir fotoğraf ve bir cümle gönderir; ajan stokta olan, bedenine uyan ve bütçeye sığan üç kombin kurar, seçileni sepete ekler. 142.000 çalıştırma / ay %96,1 başarı 5 ajan Perakende & E-ticaretHITL ### İade ve para iadesi Politika ve müşteri geçmişi saniyede kontrol edilir, etiket üretilir; iade tutarı kargo taramasıyla ödenir. 48.000 çalıştırma / ay %96,7 başarı 4 ajan Perakende & E-ticaret ### Ürün soru-cevap asistanı Beden, uyumluluk ve özellik soruları katalog ve iade verisiyle yanıtlanır; stok ve fiyat doğrulanır. 120.000 çalıştırma / ay %98,7 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Kademeli OTA güncelleme dağıtımı URL: https://www.aginies.com/tr/ajanlar/ota-update-staged-rollout Language: tr Summary: Firmware dağıtımı %1 → %10 → %100 kademelerinde ilerler; her kademede sağlık kapısı, kritik filolarda insan onayı. CI/CD’den imzalı firmware sürümü yayınlandı olayı → Sürüm sağlık kapılarını geçerek filoya yayıldı ya da otomatik geri alındı; her adım kayıtlı IoT ve Bağlı Cihazlarapibatch # Kademeli OTA güncelleme dağıtımı Firmware dağıtımı %1 → %10 → %100 kademelerinde ilerler; her kademede sağlık kapısı, kritik filolarda insan onayı. Kısaca Kademeli OTA güncelleme dağıtımı, IoT ve Bağlı Cihazlar sektöründe api, batch kanalında çalışan bir ajan senaryosudur. Tetikleyici: CI/CD’den imzalı firmware sürümü yayınlandı olayı. Dağıtım Süpervizörü 3 uzman ajanı (Plan Ajanı, Sağlık Ajanı, Dağıtım Ajanı) 4 sistem üzerinde yönetir (CI/CD (signed artefacts), Device Management / OTA, Observability (metrics), Slack); 9 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Sürüm sağlık kapılarını geçerek filoya yayıldı ya da otomatik geri alındı; her adım kayıtlı. Onay kuralı: Kritik filo segmentine dağıtım → ürün sahibi onayı; sağlık kapısı düşerse otomatik geri alma. IoT ve Bağlı Cihazlar ### Kademeli OTA güncelleme dağıtımı [Kademeli OTA güncelleme dağıtımı] çalıştırma #41.287.820süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Kritik filo segmentine dağıtım → ürün sahibi onayı; sağlık kapısı düşerse otomatik geri alma Bu çalıştırmadaki ajanlar Dağıtım Süpervizörüsupervisorplanlar, yönlendirir, onaylar Plan Ajanıaginies-talay-2 · on-premKademeleri ve kanarya grubunu seçer Sağlık Ajanıaginies-tulpar-4 · on-premKademe sonrası metrikleri değerlendirir Dağıtım Ajanıaginies-sunkar-1 · on-premKademeyi başlatır, gerekirse geri alır Denetim izi00:00.00 - ◦Sürüm 5.0.3 imzalandı: 46.200 cihaz hedefte, 3 filo segmenti - ⚙Segmentler çekildi: depo sensörleri, araç telematiği, kritik soğuk zincir - ∿Plan: %1 kanarya (462 cihaz, düşük risk) → %10 → %100; kritik segment son ve onaylı - ⇢Kademe 1 başlatıldı: 462 cihaz, 30 dakikalık gözlem penceresi - ⚙Kademe metrikleri: bağlantı %99,6, çökme 0, pil tüketimi +%1,2 (eşik %5) - ✓Sağlık kapısı geçildi; taban çizgiyle fark istatistiksel olarak anlamsız - ⇢Kademe 2 başlatıldı: 4.620 cihaz; kapı geçildi, kademe 3 hazır - ⏸Kritik soğuk zincir segmenti → ürün sahibi %100 dağıtımı onayladı - ➤Ekibe rapor: 3 kademe, 0 geri alma, 46.200 cihaz 5.0.3; kayıt sürümle saklandı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici CI/CD’den imzalı firmware sürümü yayınlandı olayı Sonuç Sürüm sağlık kapılarını geçerek filoya yayıldı ya da otomatik geri alındı; her adım kayıtlı ## Ajanlar SÜPERVİZÖR ### Dağıtım Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Plan Ajanı Kademeleri ve kanarya grubunu seçer get_fleet_segmentsselect_canarycreate_rollout_plan aginies-tulpar-4 · on-prem ### Sağlık Ajanı Kademe sonrası metrikleri değerlendirir get_stage_metricscompare_baselinedecide_gate aginies-sunkar-1 · on-prem ### Dağıtım Ajanı Kademeyi başlatır, gerekirse geri alır start_stagepause_rolloutrollback_stagenotify_team ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Dağıtım Süpervizörü | Sürüm 5.0.3 imzalandı: 46.200 cihaz hedefte, 3 filo segmenti | CI/CD (signed artefacts) | 200 | — | | 02 | araç | Plan Ajanı | Segmentler çekildi: depo sensörleri, araç telematiği, kritik soğuk zincir | Device Management / OTA | 420 | — | | 03 | akıl yürütme | Plan Ajanı | Plan: %1 kanarya (462 cihaz, düşük risk) → %10 → %100; kritik segment son ve onaylı | — | 700 | 820 | | 04 | yazma | Dağıtım Ajanı | Kademe 1 başlatıldı: 462 cihaz, 30 dakikalık gözlem penceresi | Device Management / OTA | 480 | — | | 05 | araç | Sağlık Ajanı | Kademe metrikleri: bağlantı %99,6, çökme 0, pil tüketimi +%1,2 (eşik %5) | Observability (metrics) | 520 | — | | 06 | doğrulama | Sağlık Ajanı | Sağlık kapısı geçildi; taban çizgiyle fark istatistiksel olarak anlamsız | — | 560 | 520 | | 07 | yazma | Dağıtım Ajanı | Kademe 2 başlatıldı: 4.620 cihaz; kapı geçildi, kademe 3 hazır | Device Management / OTA | 520 | — | | 08 | insan onayı | Dağıtım Süpervizörü | Kritik soğuk zincir segmenti → ürün sahibi %100 dağıtımı onayladı | — | 2.800 | — | | 09 | bildirim | Dağıtım Ajanı | Ekibe rapor: 3 kademe, 0 geri alma, 46.200 cihaz 5.0.3; kayıt sürümle saklandı | Slack | 240 | — | Simülasyon metrikleri (tahmini) 320çalıştırma / ay %98,1başarı 6,9 sp50 $0,0083tahmini maliyet / çalıştırma Onay kuralı Kritik filo segmentine dağıtım → ürün sahibi onayı; sağlık kapısı düşerse otomatik geri alma Adım profili giriş · 1akıl yürütme · 1araç · 2doğrulama · 1insan onayı · 1yazma · 2bildirim · 1 SİSTEMLER - CI/CD (signed artefacts)api - Device Management / OTAcore - Observability (metrics)db - Slackchannel Çalıştırmak için gerekenler - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük IoT operatörlerinin kanarya ve sağlık kapılı OTA dağıtım pratiklerini yansıtır Bu senaryonun yer aldığı paketler IoT cihaz filosu operasyonu → Sektör: IoT ve Bağlı Cihazlar → İlgili ajanlar IoT ve Bağlı CihazlarHITL ### Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyen Müşteri "cihaz çalışmıyor" der; telemetri okunur, uzaktan tanı konur, güncelleme gönderilir, gerekirse teknisyen doğru parçayla atanır. 37.000 çalıştırma / ay %94,1 başarı 5 ajan IoT ve Bağlı CihazlarHITL ### Cihaz filosu anomali triyajı Binlerce cihazdan gelen telemetri anomalileri kümelenir, kök neden aday listesi çıkar, doğru ekibe tek olay olarak düşer. 18.000 çalıştırma / ay %96,9 başarı 4 ajan IoT ve Bağlı CihazlarHITL ### Soğuk zincir sapması müdahalesi Sıcaklık sapması anında değerlendirilir, ürün riski hesaplanır, saha ekibi sevk edilir, kalite kararı kayıt altına alınır. 2.700 çalıştırma / ay %97,4 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Kesinti bildirimi yönetimi URL: https://www.aginies.com/tr/ajanlar/outage-report-handling Language: tr Summary: Bildirimler fider bazında kümelenir, tek OSS kaydı açılır, etkilenen herkes ekip yola çıkmadan bilgilenir. 186 hattına sesli arama ya da uygulamadan kesinti bildirimi → Tek OSS kaydı açıldı, ekip sevk edildi, etkilenen müşteriler ETA ve kapanış mesajı aldı Enerjivoiceappwhatsappweb # Kesinti bildirimi yönetimi Bildirimler fider bazında kümelenir, tek OSS kaydı açılır, etkilenen herkes ekip yola çıkmadan bilgilenir. Kısaca Kesinti bildirimi yönetimi, Enerji sektöründe voice, app, whatsapp, web kanalında çalışan bir ajan senaryosudur. Tetikleyici: 186 hattına sesli arama ya da uygulamadan kesinti bildirimi. Kesinti Süpervizörü 4 uzman ajanı (Ses Ajanı, Bildirim Ajanı, Kümeleme Ajanı, İletişim Ajanı) 5 sistem üzerinde yönetir (Customer CRM / Billing, SCADA / Historian, Network OSS, WhatsApp Business API, SMS Gateway); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Tek OSS kaydı açıldı, ekip sevk edildi, etkilenen müşteriler ETA ve kapanış mesajı aldı. Onay kuralı: Toplu bilgilendirme > 1.000 müşteri → kontrol odası dispeçeri onayı. Enerji ### Kesinti bildirimi yönetimi [Kesinti bildirimi yönetimi] çalıştırma #41.286.869süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Toplu bilgilendirme > 1.000 müşteri → kontrol odası dispeçeri onayı Bu çalıştırmadaki ajanlar Kesinti Süpervizörüsupervisorplanlar, yönlendirir, onaylar Ses Ajanıaginies-voice-tr-v3 · on-premSesli aramayı yazıya döker Bildirim Ajanıaginies-sunkar-1 · on-premAdres, sayaç ve niyeti çıkarır Kümeleme Ajanıaginies-tulpar-4 · on-premBildirimleri SCADA ve fider topolojisiyle ilişkilendirir İletişim Ajanıaginies-talay-2 · on-premOSS kaydı açar ve müşterileri bilgilendirir Denetim izi00:00.00 - ◦186 hattına arama: "20:15'ten beri elektrik yok", arayan numarayla doğrulandı - ∿Yazıya döküldü (38 sn); niyet: kesinti bildirimi; adres: Kadıköy, Moda Cad. - ⚙Sayaç 4020-77812, fider KD-17; aynı fiderden 12 dakikada 3 bildirim daha - ⚙SCADA: KD-17 kesicisi 20:14'te açtı, 1.840 müşteri etkilendi - ∿Bildirimler tek olayda kümelendi; olası neden fider arızası, tekil abone değil - ✓Tekilleştirme: KD-17 için açık OSS kaydı yok → yeni kayıt gerekli - ⇢OSS-2026-51873 kaydı açıldı, öncelik P1, ekip sevki istendi - ⏸Toplu mesaj > 1.000 müşteri → kontrol odası dispeçeri onayladı - ➤Arayan ve izinli 1.640 müşteriye: ekip yolda, tahmini 22:30 - ➤Enerji 22:12'de verildi → kapanış SMS'i gönderildi, 2 tekil vaka yükseltildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici 186 hattına sesli arama ya da uygulamadan kesinti bildirimi Sonuç Tek OSS kaydı açıldı, ekip sevk edildi, etkilenen müşteriler ETA ve kapanış mesajı aldı ## Ajanlar SÜPERVİZÖR ### Kesinti Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-voice-tr-v3 · on-prem ### Ses Ajanı Sesli aramayı yazıya döker transcribe_audio aginies-sunkar-1 · on-prem ### Bildirim Ajanı Adres, sayaç ve niyeti çıkarır detect_intentextract_locationget_customer_meter aginies-tulpar-4 · on-prem ### Kümeleme Ajanı Bildirimleri SCADA ve fider topolojisiyle ilişkilendirir get_feeder_topologyget_scada_eventscluster_reportsfind_open_ticket aginies-talay-2 · on-prem ### İletişim Ajanı OSS kaydı açar ve müşterileri bilgilendirir create_oss_ticketupdate_oss_ticketsend_whatsappsend_sms ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Kesinti Süpervizörü | 186 hattına arama: "20:15'ten beri elektrik yok", arayan numarayla doğrulandı | — | 200 | — | | 02 | akıl yürütme | Ses Ajanı | Yazıya döküldü (38 sn); niyet: kesinti bildirimi; adres: Kadıköy, Moda Cad. | — | 780 | 340 | | 03 | araç | Bildirim Ajanı | Sayaç 4020-77812, fider KD-17; aynı fiderden 12 dakikada 3 bildirim daha | Customer CRM / Billing | 420 | — | | 04 | araç | Kümeleme Ajanı | SCADA: KD-17 kesicisi 20:14'te açtı, 1.840 müşteri etkilendi | SCADA / Historian | 380 | — | | 05 | akıl yürütme | Kümeleme Ajanı | Bildirimler tek olayda kümelendi; olası neden fider arızası, tekil abone değil | — | 640 | 720 | | 06 | doğrulama | Kümeleme Ajanı | Tekilleştirme: KD-17 için açık OSS kaydı yok → yeni kayıt gerekli | — | 300 | 180 | | 07 | yazma | İletişim Ajanı | OSS-2026-51873 kaydı açıldı, öncelik P1, ekip sevki istendi | Network OSS | 520 | — | | 08 | insan onayı | Kesinti Süpervizörü | Toplu mesaj > 1.000 müşteri → kontrol odası dispeçeri onayladı | — | 2.200 | — | | 09 | bildirim | İletişim Ajanı | Arayan ve izinli 1.640 müşteriye: ekip yolda, tahmini 22:30 | WhatsApp Business API | 300 | — | | 10 | bildirim | İletişim Ajanı | Enerji 22:12'de verildi → kapanış SMS'i gönderildi, 2 tekil vaka yükseltildi | SMS Gateway | 260 | — | Simülasyon metrikleri (tahmini) 38.000çalıştırma / ay %97,7başarı 6 sp50 $0,0015tahmini maliyet / çalıştırma Onay kuralı Toplu bilgilendirme > 1.000 müşteri → kontrol odası dispeçeri onayı Adım profili giriş · 1akıl yürütme · 2araç · 2doğrulama · 1insan onayı · 1yazma · 1bildirim · 2 SİSTEMLER - Customer CRM / Billingcrm - SCADA / Historiancore - Network OSSticketing - WhatsApp Business APIchannel - SMS Gatewaychannel Çalıştırmak için gerekenler - CRM için API erişimi ve müşteri eşleme anahtarı - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Avrupa dağıtım şirketlerinin proaktif müşteri bilgilendirmeli kesinti iletişimini yansıtır Bu senaryonun yer aldığı paketler Enerji müşteri operasyonları → Sektör: Enerji → İlgili ajanlar EnerjiHITL ### Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda Abone arar ya da yazar; yüksek fatura açıklanır, kesinti süresi verilir, sayaç verisi gösterilir, taksit planı kurulur. 520.000 çalıştırma / ay %94,6 başarı 5 ajan EnerjiHITL ### Sayaç okuma itirazı Müşteri sayaç fotoğrafı yollar; endeks görüntüden okunur, tahmini fatura düzeltilir, büyük iade lidere gider. 9.200 çalıştırma / ay %94,9 başarı 4 ajan EnerjiHITL ### Çatı GES bağlantı başvurusu Başvuru belgeleri okunur, trafo kapasitesi ve mevzuat kontrol edilir, eksikler aynı gün bildirilir; bağlantı görüşü mühendisin. 5.400 çalıştırma / ay %95,3 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin URL: https://www.aginies.com/tr/ajanlar/outfit-stylist-agent Language: tr Summary: Müşteri bir fotoğraf ve bir cümle gönderir; ajan stokta olan, bedenine uyan ve bütçeye sığan üç kombin kurar, seçileni sepete ekler. Uygulama, web veya WhatsApp üzerinden "bununla ne giyerim" tipi talep → Üç kombin sunuldu, biri sepette; beden profili ve tercihler kaydedildi Perakende & E-ticaretappwebwhatsapp # Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin Müşteri bir fotoğraf ve bir cümle gönderir; ajan stokta olan, bedenine uyan ve bütçeye sığan üç kombin kurar, seçileni sepete ekler. Kısaca Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin, Perakende & E-ticaret sektöründe app, web, whatsapp kanalında çalışan bir ajan senaryosudur. Tetikleyici: Uygulama, web veya WhatsApp üzerinden "bununla ne giyerim" tipi talep. Stil Süpervizörü 4 uzman ajanı (Görsel Ajanı, Stilist Ajanı, Katalog Ajanı, Sepet Ajanı) 5 sistem üzerinde yönetir (WhatsApp Business, Product Catalogue (PIM), Inventory, Customer Profile, Cart & Checkout API); 10 adımın 2 tanesi sistemde işlem, 1 tanesi doğrulama veya onaydır. Sonuç: Üç kombin sunuldu, biri sepette; beden profili ve tercihler kaydedildi. Onay kuralı: Kampanya kuralı dışı indirim veya kişiye özel fiyat talebi → müşteri hizmetleri. Perakende & E-ticaret ### Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin [Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin] çalıştırma #41.289.960süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Kampanya kuralı dışı indirim veya kişiye özel fiyat talebi → müşteri hizmetleri Bu çalıştırmadaki ajanlar Stil Süpervizörüsupervisorplanlar, yönlendirir, onaylar Görsel Ajanıaginies-konrul-1 · on-premGönderilen parçanın tür, renk, kalıp ve kumaşını çıkarır Stilist Ajanıaginies-tulpar-4 · on-premStil kuralları, beden profili ve bütçeyle kombinleri kurar ve gerekçelendirir Katalog Ajanıaginies-sunkar-1 · on-premStok, beden ve fiyat filtresiyle aday ürünleri getirir Sepet Ajanıaginies-sunkar-1 · on-premSeçimi sepete ekler, teslimat ve iade koşullarını bildirir Denetim izi00:00.00 - ◦WhatsApp: ceket fotoğrafı + "Cuma akşam yemeği, smart casual, 3.000 TL'ye kadar" - ∿Parça: lacivert yün karışımlı blazer, slim kalıp; katalogda eşleşen ürün bulundu - ⌕Beden profili: 50 / 32-32 / 43; geçmiş iadelerde dar kalıp pantolon sapması - ⌘Plan: adayları getir → iade riskini ele → 3 kombin kur → stok teyidi → sepet - ⚙Gömlek, pantolon, ayakkabı, aksesuar: beden ve bütçe filtresiyle 38 aday, hepsi stokta - ✓İade oranı > %35 olan 5 ürün elendi; beden tablosu sapması olan 2 pantolon çıkarıldı - ∿3 kombin: 2.740 / 2.890 / 2.980 TL; renk ve kalıp gerekçeleriyle, görsel kolaj hazırlandı - ➤Kombinler WhatsApp'ta gönderildi; müşteri 2. kombini seçti, ayakkabıyı 44 istedi - ⇢4 ürün sepete eklendi, ayakkabı 44 stok teyitli; mağazadan 2 saatte teslim seçeneği - ⇢Beden profili ve stil tercihleri güncellendi; bir sonraki öneri için not düşüldü - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Uygulama, web veya WhatsApp üzerinden "bununla ne giyerim" tipi talep Sonuç Üç kombin sunuldu, biri sepette; beden profili ve tercihler kaydedildi ## Ajanlar SÜPERVİZÖR ### Stil Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### Görsel Ajanı Gönderilen parçanın tür, renk, kalıp ve kumaşını çıkarır describe_garmentmatch_catalogue_item aginies-tulpar-4 · on-prem ### Stilist Ajanı Stil kuralları, beden profili ve bütçeyle kombinleri kurar ve gerekçelendirir get_style_rulesget_size_profilecompose_outfits aginies-sunkar-1 · on-prem ### Katalog Ajanı Stok, beden ve fiyat filtresiyle aday ürünleri getirir search_cataloguecheck_stockget_return_rate aginies-sunkar-1 · on-prem ### Sepet Ajanı Seçimi sepete ekler, teslimat ve iade koşullarını bildirir add_to_cartget_delivery_optionssave_preferences ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Stil Süpervizörü | WhatsApp: ceket fotoğrafı + "Cuma akşam yemeği, smart casual, 3.000 TL'ye kadar" | WhatsApp Business | 180 | — | | 02 | akıl yürütme | Görsel Ajanı | Parça: lacivert yün karışımlı blazer, slim kalıp; katalogda eşleşen ürün bulundu | Product Catalogue (PIM) | 1.300 | 760 | | 03 | bilgi | Stilist Ajanı | Beden profili: 50 / 32-32 / 43; geçmiş iadelerde dar kalıp pantolon sapması | Customer Profile | 420 | 180 | | 04 | plan | Stil Süpervizörü | Plan: adayları getir → iade riskini ele → 3 kombin kur → stok teyidi → sepet | — | 240 | 220 | | 05 | araç | Katalog Ajanı | Gömlek, pantolon, ayakkabı, aksesuar: beden ve bütçe filtresiyle 38 aday, hepsi stokta | Inventory | 560 | — | | 06 | doğrulama | Katalog Ajanı | İade oranı > %35 olan 5 ürün elendi; beden tablosu sapması olan 2 pantolon çıkarıldı | — | 380 | 150 | | 07 | akıl yürütme | Stilist Ajanı | 3 kombin: 2.740 / 2.890 / 2.980 TL; renk ve kalıp gerekçeleriyle, görsel kolaj hazırlandı | — | 2.200 | 1.500 | | 08 | bildirim | Sepet Ajanı | Kombinler WhatsApp'ta gönderildi; müşteri 2. kombini seçti, ayakkabıyı 44 istedi | WhatsApp Business | 1.800 | 240 | | 09 | yazma | Sepet Ajanı | 4 ürün sepete eklendi, ayakkabı 44 stok teyitli; mağazadan 2 saatte teslim seçeneği | Cart & Checkout API | 520 | — | | 10 | yazma | Sepet Ajanı | Beden profili ve stil tercihleri güncellendi; bir sonraki öneri için not düşüldü | Customer Profile | 300 | — | Simülasyon metrikleri (tahmini) 142.000çalıştırma / ay %96,1başarı 6,1 sp50 $0,0016tahmini maliyet / çalıştırma Onay kuralı Kampanya kuralı dışı indirim veya kişiye özel fiyat talebi → müşteri hizmetleri Adım profili giriş · 1plan · 1bilgi · 1akıl yürütme · 2araç · 1doğrulama · 1yazma · 2bildirim · 1 SİSTEMLER - WhatsApp Businesschannel - Product Catalogue (PIM)db - Inventoryerp - Customer Profilecrm - Cart & Checkout APIapi Çalıştırmak için gerekenler - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - ERP/MES için kayıt ve sorgu API erişimi - CRM için API erişimi ve müşteri eşleme anahtarı - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük moda perakendecilerinin stil asistanı deseni; stok, beden ve iade verisiyle birleştirilmiş Bu senaryonun yer aldığı paketler E-ticaret müşteri desteği ve aksiyon → Sektör: Perakende & E-ticaret → İlgili ajanlar Perakende & E-ticaret ### Siparişim nerede? (WISMO) Müşteri sorar, sipariş ve kargo olayları anında birleşir; gecikmede yeni tarih verilir, kargo araştırılır. 110.000 çalıştırma / ay %98,1 başarı 4 ajan Perakende & E-ticaretHITL ### İade ve para iadesi Politika ve müşteri geçmişi saniyede kontrol edilir, etiket üretilir; iade tutarı kargo taramasıyla ödenir. 48.000 çalıştırma / ay %96,7 başarı 4 ajan Perakende & E-ticaret ### Ürün soru-cevap asistanı Beden, uyumluluk ve özellik soruları katalog ve iade verisiyle yanıtlanır; stok ve fiyat doğrulanır. 120.000 çalıştırma / ay %98,7 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Hasta ön kabul ve triyaj URL: https://www.aginies.com/tr/ajanlar/patient-intake-triage Language: tr Summary: Hasta yazar ya da anlatır; anamnez ve aciliyet skoru saniyeler sürer, son kararı her zaman hemşire verir. Uygulamada veya WhatsApp'ta sesli/yazılı şikâyet bildirimi → Aciliyet hemşire tarafından onaylandı, uygun klinikte randevu oluşturuldu, hasta yönlendirildi Sağlıkappvoicewhatsapp # Hasta ön kabul ve triyaj Hasta yazar ya da anlatır; anamnez ve aciliyet skoru saniyeler sürer, son kararı her zaman hemşire verir. Kısaca Hasta ön kabul ve triyaj, Sağlık sektöründe app, voice, whatsapp kanalında çalışan bir ajan senaryosudur. Tetikleyici: Uygulamada veya WhatsApp'ta sesli/yazılı şikâyet bildirimi. Triyaj Süpervizörü 4 uzman ajanı (Ses Ajanı, Semptom Ajanı, Aciliyet Ajanı, Yönlendirme Ajanı) 4 sistem üzerinde yönetir (EHR (HL7/FHIR), HIS Scheduling, Triage Protocol Vector Store, WhatsApp Business API); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Aciliyet hemşire tarafından onaylandı, uygun klinikte randevu oluşturuldu, hasta yönlendirildi. Onay kuralı: Her vakada aciliyet seviyesini triyaj hemşiresi onaylar; "acil durum" vakaları doğrudan 112'ye yönlendirilir. Sağlık ### Hasta ön kabul ve triyaj [Hasta ön kabul ve triyaj] çalıştırma #41.290.240süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Her vakada aciliyet seviyesini triyaj hemşiresi onaylar; "acil durum" vakaları doğrudan 112'ye yönlendirilir Bu çalıştırmadaki ajanlar Triyaj Süpervizörüsupervisorplanlar, yönlendirir, onaylar Ses Ajanıaginies-voice-tr-v3 · on-premSesli mesajı yazıya döker, dili tespit eder Semptom Ajanıaginies-sunkar-1 · on-premEk sorular sorar, şikâyetleri yapılandırır Aciliyet Ajanıaginies-tulpar-4 · on-premProtokole göre aciliyet skorlar; asla nihai karar vermez Yönlendirme Ajanıaginies-talay-2 · on-premUygun kliniği ve saati bulur, randevuyu oluşturur Denetim izi00:00.00 - ◦Uygulamadan sesli mesaj alındı, hasta kimliği OTP ile doğrulandı - ∿Yazıya döküldü (Türkçe, 42 sn): "iki gündür göğsümde sıkışma, merdivende artıyor" - ⚙Geçmiş çekildi: 58 yaş, hipertansiyon, önceki kardiyak olay yok - ∿3 ek soru soruldu: yayılan ağrı yok, istirahatte nefes darlığı yok - ⌕Göğüs ağrısı için Manchester Triyaj ayırıcıları getirildi - ∿Aciliyet skoru 71/100 → "acil" (aynı gün kardiyoloji), kırmızı bayrak 1/6 - ✓Koruma: hastaya tanı verilmedi, acil durum ifadesi kontrol edildi - ⏸Triyaj hemşiresi panelde aciliyet seviyesini onayladı (her vakada zorunlu) - ⇢Kardiyoloji randevusu bugün 15:40, Dr. Kaya, oda C-204 olarak oluşturuldu - ➤Hastaya WhatsApp ile saat, yol tarifi ve kötüleşirse 112 uyarısı gönderildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Uygulamada veya WhatsApp'ta sesli/yazılı şikâyet bildirimi Sonuç Aciliyet hemşire tarafından onaylandı, uygun klinikte randevu oluşturuldu, hasta yönlendirildi ## Ajanlar SÜPERVİZÖR ### Triyaj Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-voice-tr-v3 · on-prem ### Ses Ajanı Sesli mesajı yazıya döker, dili tespit eder transcribe_audiodetect_language aginies-sunkar-1 · on-prem ### Semptom Ajanı Ek sorular sorar, şikâyetleri yapılandırır ask_followupextract_symptomsget_patient_history aginies-tulpar-4 · on-prem ### Aciliyet Ajanı Protokole göre aciliyet skorlar; asla nihai karar vermez search_triage_protocolscore_urgencycheck_red_flags aginies-talay-2 · on-prem ### Yönlendirme Ajanı Uygun kliniği ve saati bulur, randevuyu oluşturur find_clinic_slotbook_appointmentsend_whatsapp ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Triyaj Süpervizörü | Uygulamadan sesli mesaj alındı, hasta kimliği OTP ile doğrulandı | — | 220 | — | | 02 | akıl yürütme | Ses Ajanı | Yazıya döküldü (Türkçe, 42 sn): "iki gündür göğsümde sıkışma, merdivende artıyor" | — | 900 | 380 | | 03 | araç | Semptom Ajanı | Geçmiş çekildi: 58 yaş, hipertansiyon, önceki kardiyak olay yok | EHR (HL7/FHIR) | 430 | — | | 04 | akıl yürütme | Semptom Ajanı | 3 ek soru soruldu: yayılan ağrı yok, istirahatte nefes darlığı yok | — | 1.100 | 720 | | 05 | bilgi | Aciliyet Ajanı | Göğüs ağrısı için Manchester Triyaj ayırıcıları getirildi | Triage Protocol Vector Store | 380 | 520 | | 06 | akıl yürütme | Aciliyet Ajanı | Aciliyet skoru 71/100 → "acil" (aynı gün kardiyoloji), kırmızı bayrak 1/6 | — | 760 | 940 | | 07 | doğrulama | Aciliyet Ajanı | Koruma: hastaya tanı verilmedi, acil durum ifadesi kontrol edildi | — | 340 | 260 | | 08 | insan onayı | Triyaj Süpervizörü | Triyaj hemşiresi panelde aciliyet seviyesini onayladı (her vakada zorunlu) | — | 2.800 | — | | 09 | yazma | Yönlendirme Ajanı | Kardiyoloji randevusu bugün 15:40, Dr. Kaya, oda C-204 olarak oluşturuldu | HIS Scheduling | 480 | — | | 10 | bildirim | Yönlendirme Ajanı | Hastaya WhatsApp ile saat, yol tarifi ve kötüleşirse 112 uyarısı gönderildi | WhatsApp Business API | 260 | — | Simülasyon metrikleri (tahmini) 24.500çalıştırma / ay %95,8başarı 7,7 sp50 $0,0034tahmini maliyet / çalıştırma Onay kuralı Her vakada aciliyet seviyesini triyaj hemşiresi onaylar; "acil durum" vakaları doğrudan 112'ye yönlendirilir Adım profili giriş · 1bilgi · 1akıl yürütme · 3araç · 1doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - EHR (HL7/FHIR)core - HIS Schedulingcore - Triage Protocol Vector Storedb - WhatsApp Business APIchannel Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek NHS 111 ve büyük hastane gruplarının dijital ön kapı ve semptom değerlendirme yaklaşımını yansıtır Bu senaryonun yer aldığı paketler Sağlık operasyonları ve klinik yönlendirme → Sektör: Sağlık → İlgili ajanlar SağlıkHITL ### Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuç Hasta yazar ya da arar; randevu, ön hazırlık talimatı, sigorta ön onayı ve sonuç bildirimi tek koordinatörde birleşir. 58.000 çalıştırma / ay %95,1 başarı 5 ajan SağlıkHITL ### Ön yetkilendirme (prior authorization) İstem düşer düşmez ödeyici kuralları taranır, klinik kanıt toplanır, dosya dakikalar içinde gönderime hazır. 6.200 çalıştırma / ay %94,1 başarı 4 ajan Sağlık ### Randevu planlama ve no-show azaltma Gelmeme riski yüksek hastalar WhatsApp'tan önceden aranır, boşalan slotlar bekleme listesinden anında dolar. 62.000 çalıştırma / ay %98,2 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuç URL: https://www.aginies.com/tr/ajanlar/patient-journey-coordinator Language: tr Summary: Hasta yazar ya da arar; randevu, ön hazırlık talimatı, sigorta ön onayı ve sonuç bildirimi tek koordinatörde birleşir. Hekim istemi sonrası hastanın WhatsApp veya çağrı merkezi üzerinden randevu talebi → Randevu alındı, hazırlık talimatı iletildi, ön onay alındı, sonuç hatırlatması planlandı Sağlıkwhatsappvoiceapp # Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuç Hasta yazar ya da arar; randevu, ön hazırlık talimatı, sigorta ön onayı ve sonuç bildirimi tek koordinatörde birleşir. Kısaca Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuç, Sağlık sektöründe whatsapp, voice, app kanalında çalışan bir ajan senaryosudur. Tetikleyici: Hekim istemi sonrası hastanın WhatsApp veya çağrı merkezi üzerinden randevu talebi. Hasta Yolculuğu Süpervizörü 4 uzman ajanı (Diyalog Ajanı, Randevu Ajanı, Klinik Hazırlık Ajanı, Sigorta Ajanı) 5 sistem üzerinde yönetir (WhatsApp Business, Hospital Information System, Scheduling, Insurer Portal, Lab / Imaging); 10 adımın 3 tanesi sistemde işlem, 3 tanesi doğrulama veya onaydır. Sonuç: Randevu alındı, hazırlık talimatı iletildi, ön onay alındı, sonuç hatırlatması planlandı. Onay kuralı: İlaç etkileşimi, kontrendikasyon şüphesi veya ek tetkik ihtiyacı → hekim onayı. Sağlık ### Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuç [Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuç] çalıştırma #41.289.309süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: İlaç etkileşimi, kontrendikasyon şüphesi veya ek tetkik ihtiyacı → hekim onayı Bu çalıştırmadaki ajanlar Hasta Yolculuğu Süpervizörüsupervisorplanlar, yönlendirir, onaylar Diyalog Ajanıaginies-talay-2 · on-premHastayla konuşur, kimliği doğrular, tercihleri alır Randevu Ajanıaginies-sunkar-1 · on-premCihaz, hekim ve hasta uygunluğunu eşler, randevuyu oluşturur Klinik Hazırlık Ajanıaginies-tulpar-4 · on-premProtokole göre hazırlık talimatı üretir, ilaç ve kontrendikasyon kontrolü yapar Sigorta Ajanıaginies-sunkar-1 · on-premÖn onay başvurusunu hazırlar ve sonucu izler Denetim izi00:00.00 - ◦WhatsApp: "Doktor MR istedi, en kısa zamanda, sabah olsun, özel sigortam var" - ✓Hasta TC ve doğum tarihiyle doğrulandı; HIS'te kontrastlı lomber MR istemi bulundu - ⌘Plan: klinik kontrol → ön onay → uygun slot → randevu → hazırlık → hatırlatma - ✓İlaç listesi: metformin → kontrast protokolü gereği 48 saat ara; böbrek değeri 3 ay içinde yok - ⏸Kreatinin istemi için hekim onayı istendi; hekim mobil onayla ekledi - ⇢Ön onay başvurusu gönderildi: ICD ve işlem kodu, hekim raporu ekli - ⚙Sigorta yanıtı: onaylandı, katılım payı %10, geçerlilik 30 gün - ⇢Perşembe 08:40 MR, Çarşamba 08:00 kan alma; iki randevu HIS'te oluşturuldu - ➤Hazırlık talimatı: metformin arası, açlık, metal eşya; katılım payı ve adres bilgisi iletildi - ⇢Hatırlatmalar planlandı: 24 saat önce, sonuç yüklendiğinde hekim randevusu önerisi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Hekim istemi sonrası hastanın WhatsApp veya çağrı merkezi üzerinden randevu talebi Sonuç Randevu alındı, hazırlık talimatı iletildi, ön onay alındı, sonuç hatırlatması planlandı ## Ajanlar SÜPERVİZÖR ### Hasta Yolculuğu Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Diyalog Ajanı Hastayla konuşur, kimliği doğrular, tercihleri alır verify_patientdetect_intentscompose_reply aginies-sunkar-1 · on-prem ### Randevu Ajanı Cihaz, hekim ve hasta uygunluğunu eşler, randevuyu oluşturur find_slotsbook_slotreschedule aginies-tulpar-4 · on-prem ### Klinik Hazırlık Ajanı Protokole göre hazırlık talimatı üretir, ilaç ve kontrendikasyon kontrolü yapar get_protocolcheck_medicationsflag_physician aginies-sunkar-1 · on-prem ### Sigorta Ajanı Ön onay başvurusunu hazırlar ve sonucu izler prepare_preauthsubmit_preauthpoll_status ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Hasta Yolculuğu Süpervizörü | WhatsApp: "Doktor MR istedi, en kısa zamanda, sabah olsun, özel sigortam var" | WhatsApp Business | 180 | — | | 02 | doğrulama | Diyalog Ajanı | Hasta TC ve doğum tarihiyle doğrulandı; HIS'te kontrastlı lomber MR istemi bulundu | Hospital Information System | 520 | 180 | | 03 | plan | Hasta Yolculuğu Süpervizörü | Plan: klinik kontrol → ön onay → uygun slot → randevu → hazırlık → hatırlatma | — | 240 | 220 | | 04 | doğrulama | Klinik Hazırlık Ajanı | İlaç listesi: metformin → kontrast protokolü gereği 48 saat ara; böbrek değeri 3 ay içinde yok | Hospital Information System | 1.100 | 620 | | 05 | insan onayı | Klinik Hazırlık Ajanı | Kreatinin istemi için hekim onayı istendi; hekim mobil onayla ekledi | Hospital Information System | 900 | — | | 06 | yazma | Sigorta Ajanı | Ön onay başvurusu gönderildi: ICD ve işlem kodu, hekim raporu ekli | Insurer Portal | 700 | — | | 07 | araç | Sigorta Ajanı | Sigorta yanıtı: onaylandı, katılım payı %10, geçerlilik 30 gün | Insurer Portal | 2.200 | — | | 08 | yazma | Randevu Ajanı | Perşembe 08:40 MR, Çarşamba 08:00 kan alma; iki randevu HIS'te oluşturuldu | Scheduling | 640 | — | | 09 | bildirim | Diyalog Ajanı | Hazırlık talimatı: metformin arası, açlık, metal eşya; katılım payı ve adres bilgisi iletildi | WhatsApp Business | 520 | 320 | | 10 | yazma | Randevu Ajanı | Hatırlatmalar planlandı: 24 saat önce, sonuç yüklendiğinde hekim randevusu önerisi | Lab / Imaging | 300 | — | Simülasyon metrikleri (tahmini) 58.000çalıştırma / ay %95,1başarı 7,3 sp50 $0,0026tahmini maliyet / çalıştırma Onay kuralı İlaç etkileşimi, kontrendikasyon şüphesi veya ek tetkik ihtiyacı → hekim onayı Adım profili giriş · 1plan · 1araç · 1doğrulama · 2insan onayı · 1yazma · 3bildirim · 1 SİSTEMLER - WhatsApp Businesschannel - Hospital Information Systemcore - Schedulingapi - Insurer Portalapi - Lab / Imagingapi Çalıştırmak için gerekenler - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük hastane gruplarının hasta erişim merkezi deseni; klinik protokol kontrolüyle birleştirilmiş Bu senaryonun yer aldığı paketler Sağlık operasyonları ve klinik yönlendirme → Sektör: Sağlık → İlgili ajanlar SağlıkHITL ### Ön yetkilendirme (prior authorization) İstem düşer düşmez ödeyici kuralları taranır, klinik kanıt toplanır, dosya dakikalar içinde gönderime hazır. 6.200 çalıştırma / ay %94,1 başarı 4 ajan SağlıkHITL ### Hasta ön kabul ve triyaj Hasta yazar ya da anlatır; anamnez ve aciliyet skoru saniyeler sürer, son kararı her zaman hemşire verir. 24.500 çalıştırma / ay %95,8 başarı 5 ajan Sağlık ### Randevu planlama ve no-show azaltma Gelmeme riski yüksek hastalar WhatsApp'tan önceden aranır, boşalan slotlar bekleme listesinden anında dolar. 62.000 çalıştırma / ay %98,2 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Ödeme mutabakatı istisnaları URL: https://www.aginies.com/tr/ajanlar/payment-reconciliation-exceptions Language: tr Summary: PSP mutabakat dosyası ana defterle eşlenir; küçük farklar otomatik düzeltilir, büyükleri finans onayına gider. Gece toplu işi: PSP/acquirer T+1 mutabakat dosyası alındı → Tüm istisnalar eşlendi ya da düzeltildi, yevmiye kayıtları atıldı, açık kalemler finansta Finansal Teknolojilerbatchapiemail # Ödeme mutabakatı istisnaları PSP mutabakat dosyası ana defterle eşlenir; küçük farklar otomatik düzeltilir, büyükleri finans onayına gider. Kısaca Ödeme mutabakatı istisnaları, Finansal Teknolojiler sektöründe batch, api, email kanalında çalışan bir ajan senaryosudur. Tetikleyici: Gece toplu işi: PSP/acquirer T+1 mutabakat dosyası alındı. Mutabakat Süpervizörü 3 uzman ajanı (Eşleştirme Ajanı, İnceleme Ajanı, Düzeltme Ajanı) 5 sistem üzerinde yönetir (PSP / Acquirer API, Card Network, Core Ledger, Case Management, Email); 10 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Tüm istisnalar eşlendi ya da düzeltildi, yevmiye kayıtları atıldı, açık kalemler finansta. Onay kuralı: Tek fark > 10.000 TL veya toplam > 50.000 TL → finans müdürü onayı ile ayarlama kaydı. Finansal Teknolojiler ### Ödeme mutabakatı istisnaları [Ödeme mutabakatı istisnaları] çalıştırma #41.288.595süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Tek fark > 10.000 TL veya toplam > 50.000 TL → finans müdürü onayı ile ayarlama kaydı Bu çalıştırmadaki ajanlar Mutabakat Süpervizörüsupervisorplanlar, yönlendirir, onaylar Eşleştirme Ajanıaginies-talay-2 · on-premMutabakat kalemlerini ana defter işlemleriyle eşler, istisnaları ayırır İnceleme Ajanıaginies-tulpar-4 · on-premFark nedenini belirler: ücret, kur, iade, zamanlama Düzeltme Ajanıaginies-sunkar-1 · on-premTolerans içi düzeltme yevmiyesini atar, kalanı finansa yönlendirir Denetim izi00:00.00 - ◦Mutabakat dosyası alındı: 184.320 işlem, 41.870.500 TL, T+1 - ⚙PSP detayı çekildi; %99,86 eşleşti, 258 fark: 143 ücret, 61 iade, 39 kur, 15 eksik - ⌘Plan: neden sınıflandırma → tolerans kontrolü → otomatik düzeltme → finans onayı - ⚙Kart ağı ücret tarifesi çekildi: 143 kalem şema ücreti değişikliğiyle açıklandı - ⌕Ana defter kayıtları alındı: 61 iade T+2'de kaydedilmiş → zamanlama farkı - ✓Tolerans: 243 istisna ≤ 500 TL → otomatik; 15 eksik işlem 186.400 TL → finans - ⇢243 düzeltme yevmiyesi atıldı (ücret ve zamanlama), toplam 21.340 TL - ⏸Fark > 10.000 TL → finans müdürü 15 eksik işlem için ayarlama kaydını onayladı - ⇢Vaka #REC-1188 açıldı, PSP'ye eksik işlem sorgusu iletildi - ➤Finans ekibine özet e-postası: 243 otomatik düzeltildi, 15 kalem vakada - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Gece toplu işi: PSP/acquirer T+1 mutabakat dosyası alındı Sonuç Tüm istisnalar eşlendi ya da düzeltildi, yevmiye kayıtları atıldı, açık kalemler finansta ## Ajanlar SÜPERVİZÖR ### Mutabakat Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Eşleştirme Ajanı Mutabakat kalemlerini ana defter işlemleriyle eşler, istisnaları ayırır parse_settlement_filematch_transactionslist_exceptions aginies-tulpar-4 · on-prem ### İnceleme Ajanı Fark nedenini belirler: ücret, kur, iade, zamanlama get_psp_transactionget_ledger_entryget_card_network_feesclassify_break aginies-sunkar-1 · on-prem ### Düzeltme Ajanı Tolerans içi düzeltme yevmiyesini atar, kalanı finansa yönlendirir post_journalcreate_exception_casesend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Mutabakat Süpervizörü | Mutabakat dosyası alındı: 184.320 işlem, 41.870.500 TL, T+1 | — | 300 | — | | 02 | araç | Eşleştirme Ajanı | PSP detayı çekildi; %99,86 eşleşti, 258 fark: 143 ücret, 61 iade, 39 kur, 15 eksik | PSP / Acquirer API | 1.400 | — | | 03 | plan | Mutabakat Süpervizörü | Plan: neden sınıflandırma → tolerans kontrolü → otomatik düzeltme → finans onayı | — | 280 | 200 | | 04 | araç | İnceleme Ajanı | Kart ağı ücret tarifesi çekildi: 143 kalem şema ücreti değişikliğiyle açıklandı | Card Network | 540 | — | | 05 | bilgi | İnceleme Ajanı | Ana defter kayıtları alındı: 61 iade T+2'de kaydedilmiş → zamanlama farkı | Core Ledger | 480 | — | | 06 | doğrulama | İnceleme Ajanı | Tolerans: 243 istisna ≤ 500 TL → otomatik; 15 eksik işlem 186.400 TL → finans | — | 760 | 820 | | 07 | yazma | Düzeltme Ajanı | 243 düzeltme yevmiyesi atıldı (ücret ve zamanlama), toplam 21.340 TL | Core Ledger | 680 | — | | 08 | insan onayı | Mutabakat Süpervizörü | Fark > 10.000 TL → finans müdürü 15 eksik işlem için ayarlama kaydını onayladı | — | 3.200 | — | | 09 | yazma | Düzeltme Ajanı | Vaka #REC-1188 açıldı, PSP'ye eksik işlem sorgusu iletildi | Case Management | 440 | — | | 10 | bildirim | Düzeltme Ajanı | Finans ekibine özet e-postası: 243 otomatik düzeltildi, 15 kalem vakada | Email | 260 | — | Simülasyon metrikleri (tahmini) 620çalıştırma / ay %95,1başarı 8,1 sp50 $0,0133tahmini maliyet / çalıştırma Onay kuralı Tek fark > 10.000 TL veya toplam > 50.000 TL → finans müdürü onayı ile ayarlama kaydı Adım profili giriş · 1plan · 1bilgi · 1araç · 2doğrulama · 1insan onayı · 1yazma · 2bildirim · 1 SİSTEMLER - PSP / Acquirer APIpayments - Card Networkapi - Core Ledgercore - Case Managementticketing - Emailemail Çalıştırmak için gerekenler - Ödeme servisleri için yetkili API erişimi ve limitler - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük ödeme kuruluşlarının çok PSP'li otomatik mutabakat ve istisna yönetimi deseni Bu senaryonun yer aldığı paketler Finansal analiz ve operasyonel zekâ →Üye işyeri ve ödeme operasyonları → Sektör: Finansal Teknolojiler → İlgili ajanlar Finansal TeknolojilerHITL ### Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit İşyeri WhatsApp'tan yazar; POS arızası, hakediş sorgusu, iade ve limit artışı tek diyalogda çözülür. 64.000 çalıştırma / ay %93,8 başarı 5 ajan Finansal TeknolojilerHITL ### Üye işyeri KYB onboarding Sicil, gerçek faydalanıcı ve yaptırım kontrolü tek akışta; düşük riskli işyeri aynı gün ödeme almaya başlar. 1.400 çalıştırma / ay %93,4 başarı 5 ajan Finansal TeknolojilerHITL ### AML işlem izleme alarm triyajı Alarm zenginleştirilir, anlatı taslağı hazır gelir; kararı analist verir, bildirim dakikalarda dosyalanır. 8.500 çalıştırma / ay %95,8 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Ruhsat başvurusu ön inceleme URL: https://www.aginies.com/tr/ajanlar/permit-application-check Language: tr Summary: Eksik belge ve mevzuat uyumu dakikalar içinde raporlanır; kararı her zaman ruhsat memuru verir. e-Devlet üzerinden işyeri açma ruhsatı başvurusu → Memur ön inceleme özetiyle karar verdi, başvurana eksiklik listesi ve süre bildirildi Kamu ve Hükümetwebemail # Ruhsat başvurusu ön inceleme Eksik belge ve mevzuat uyumu dakikalar içinde raporlanır; kararı her zaman ruhsat memuru verir. Kısaca Ruhsat başvurusu ön inceleme, Kamu ve Hükümet sektöründe web, email kanalında çalışan bir ajan senaryosudur. Tetikleyici: e-Devlet üzerinden işyeri açma ruhsatı başvurusu. Ruhsat Süpervizörü 3 uzman ajanı (Belge Ajanı, Mevzuat Ajanı, Dosya Ajanı) 5 sistem üzerinde yönetir (e-Devlet Gateway, Case Management, GIS / Zoning Plan, Regulation Vector Store, Exchange / M365 Mail); 9 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Memur ön inceleme özetiyle karar verdi, başvurana eksiklik listesi ve süre bildirildi. Onay kuralı: Her ruhsat kararı ruhsat memuru tarafından verilir; sistem yalnızca ön inceleme yapar. Kamu ve Hükümet ### Ruhsat başvurusu ön inceleme [Ruhsat başvurusu ön inceleme] çalıştırma #41.289.494süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Her ruhsat kararı ruhsat memuru tarafından verilir; sistem yalnızca ön inceleme yapar Bu çalıştırmadaki ajanlar Ruhsat Süpervizörüsupervisorplanlar, yönlendirir, onaylar Belge Ajanıaginies-konrul-1 · on-premEklerin türünü, imzasını ve eksiklerini kontrol eder Mevzuat Ajanıaginies-tulpar-4 · on-premİmar ve ruhsat mevzuatıyla uyumu değerlendirir Dosya Ajanıaginies-talay-2 · on-premMemur için özet hazırlar, dosyayı günceller Denetim izi00:00.00 - ◦BR-2026-7731 işyeri açma ruhsatı başvurusu 9 ekle alındı - ∿11 zorunlu belgeden 9 mevcut; eksik: itfaiye raporu; kira sözleşmesi imzasız - ⚙Parsel 412/7: imar "ticari", zemin kat, 96 m² - ⌕İşyeri Açma ve Çalışma Ruhsatları Yönetmeliği md. 5, 6 ve ek 1 getirildi - ∿Kafe 2. sınıf: okula mesafe 210 m (> 100 m uygun), harç 4.750 TL, 2 şart - ✓Sonuç: eksik (2 kalem); politika gereği otomatik ret yapılamaz - ⇢Dosyaya kontrol listesi ve memur için ön inceleme özeti eklendi - ⏸Ruhsat memuru inceledi ve karar verdi: "eksiklik yazısı" (her ruhsatta zorunlu) - ➤Başvurana eksiklik listesi e-postalandı, 15 günlük tamamlama süresi başladı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici e-Devlet üzerinden işyeri açma ruhsatı başvurusu Sonuç Memur ön inceleme özetiyle karar verdi, başvurana eksiklik listesi ve süre bildirildi ## Ajanlar SÜPERVİZÖR ### Ruhsat Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### Belge Ajanı Eklerin türünü, imzasını ve eksiklerini kontrol eder extract_document_typecheck_signatureread_form_fields aginies-tulpar-4 · on-prem ### Mevzuat Ajanı İmar ve ruhsat mevzuatıyla uyumu değerlendirir get_parcel_infosearch_regulationscheck_zoning_rulescompute_fee aginies-talay-2 · on-prem ### Dosya Ajanı Memur için özet hazırlar, dosyayı günceller build_review_summaryupdate_casesend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Ruhsat Süpervizörü | BR-2026-7731 işyeri açma ruhsatı başvurusu 9 ekle alındı | — | 200 | — | | 02 | akıl yürütme | Belge Ajanı | 11 zorunlu belgeden 9 mevcut; eksik: itfaiye raporu; kira sözleşmesi imzasız | — | 940 | 880 | | 03 | araç | Mevzuat Ajanı | Parsel 412/7: imar "ticari", zemin kat, 96 m² | GIS / Zoning Plan | 420 | — | | 04 | bilgi | Mevzuat Ajanı | İşyeri Açma ve Çalışma Ruhsatları Yönetmeliği md. 5, 6 ve ek 1 getirildi | Regulation Vector Store | 460 | 620 | | 05 | akıl yürütme | Mevzuat Ajanı | Kafe 2. sınıf: okula mesafe 210 m (> 100 m uygun), harç 4.750 TL, 2 şart | — | 760 | 1.040 | | 06 | doğrulama | Dosya Ajanı | Sonuç: eksik (2 kalem); politika gereği otomatik ret yapılamaz | — | 320 | 240 | | 07 | yazma | Dosya Ajanı | Dosyaya kontrol listesi ve memur için ön inceleme özeti eklendi | Case Management | 480 | — | | 08 | insan onayı | Ruhsat Süpervizörü | Ruhsat memuru inceledi ve karar verdi: "eksiklik yazısı" (her ruhsatta zorunlu) | — | 3.200 | — | | 09 | bildirim | Dosya Ajanı | Başvurana eksiklik listesi e-postalandı, 15 günlük tamamlama süresi başladı | Exchange / M365 Mail | 260 | — | Simülasyon metrikleri (tahmini) 3.100çalıştırma / ay %95,8başarı 7 sp50 $0,0079tahmini maliyet / çalıştırma Onay kuralı Her ruhsat kararı ruhsat memuru tarafından verilir; sistem yalnızca ön inceleme yapar Adım profili giriş · 1bilgi · 1akıl yürütme · 2araç · 1doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - e-Devlet Gatewayapi - Case Managementticketing - GIS / Zoning Plandb - Regulation Vector Storedb - Exchange / M365 Mailemail Çalıştırmak için gerekenler - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Singapur ve İskandinav belediyelerinin ruhsat ön inceleme pilotlarını yansıtır Bu senaryonun yer aldığı paketler Bilgi yönetimi ve karar desteği →Vatandaş hizmetleri ve dosya ön inceleme → Sektör: Kamu ve Hükümet → İlgili ajanlar Kamu ve HükümetHITL ### Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi Vatandaş yazar ya da arar; uygunluk kontrol edilir, belgeler okunur, başvuru açılır, randevu verilir, her adım izlenebilir. 230.000 çalıştırma / ay %93,8 başarı 5 ajan Kamu ve HükümetHITL ### Vatandaş başvurusu yönlendirme e-Devlet başvurusu saniyeler içinde sınıflandırılır, yasal süresi hesaplanır ve doğru birime SLA ile atanır. 54.000 çalıştırma / ay %98,1 başarı 4 ajan Kamu ve HükümetHITL ### Kamu ihalesi teklif dosyası ön kontrolü Teklif dosyaları şartnameyle karşılaştırılır, eksik ve uygunsuz belgeler listelenir; değerlendirme kararı komisyonun. 380 çalıştırma / ay %94,8 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Tarife yükseltme danışmanı URL: https://www.aginies.com/tr/ajanlar/plan-upgrade-advisor Language: tr Summary: Kullanım analiz edilir, doğru tarife karşılaştırmalı sunulur; onayla değişiklik anında provizyonlanır. "İnternetim hep bitiyor" mesajı ya da 3 ay üst üste ek paket → Yeni tarife provizyonlandı, rıza zaman damgalı, müşteri yürürlük tarihini biliyor Telekomappwhatsappvoice # Tarife yükseltme danışmanı Kullanım analiz edilir, doğru tarife karşılaştırmalı sunulur; onayla değişiklik anında provizyonlanır. Kısaca Tarife yükseltme danışmanı, Telekom sektöründe app, whatsapp, voice kanalında çalışan bir ajan senaryosudur. Tetikleyici: "İnternetim hep bitiyor" mesajı ya da 3 ay üst üste ek paket. Tarife Süpervizörü 3 uzman ajanı (Kullanım Ajanı, Teklif Ajanı, Provizyon Ajanı) 5 sistem üzerinde yönetir (Billing (Amdocs), Product Catalogue / Offer Engine, HLR / Provisioning, WhatsApp Business API, SMS Gateway); 10 adımın 1 tanesi sistemde işlem, 1 tanesi doğrulama veya onaydır. Sonuç: Yeni tarife provizyonlandı, rıza zaman damgalı, müşteri yürürlük tarihini biliyor. Onay kuralı: Taahhütlü veya indirimli özel teklif → kampanya yöneticisi onayı. Telekom ### Tarife yükseltme danışmanı [Tarife yükseltme danışmanı] çalıştırma #41.286.660süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Taahhütlü veya indirimli özel teklif → kampanya yöneticisi onayı Bu çalıştırmadaki ajanlar Tarife Süpervizörüsupervisorplanlar, yönlendirir, onaylar Kullanım Ajanıaginies-talay-2 · on-premKullanım geçmişini çeker ve gelecek dönemi tahmin eder Teklif Ajanıaginies-tulpar-4 · on-premUygun tarifeleri karşılaştırır, teklifi sunar, yanıtı yorumlar Provizyon Ajanıaginies-sunkar-1 · on-premRızayı kaydeder, tarifeyi değiştirir, müşteriyi bilgilendirir Denetim izi00:00.00 - ◦WhatsApp: "her ay internetim bitiyor, ne yapmalıyım?" - ⚙Son 6 ay: ortalama 28 GB kullanım, 3 ayda ek paket alındı (toplam 210 TL) - ∿Mevcut tarife 20 GB / 349 TL; tahmin 32 GB'a çıkıyor → yükseltme uygun - ⌘Plan: uygun tarifeler → karşılaştırma → guardrail → teklif → rıza → provizyon - ⚙Uygun tarifeler: 40 GB / 419 TL, 60 GB / 479 TL; taahhüt yok - ∿Öneri: 40 GB / 419 TL → ek paketlere kıyasla ayda yaklaşık 70 TL tasarruf - ✓Guardrail: teklif kampanya kurallarına uygun, fiyat ve taahhüt bilgisi doğru - ➤WhatsApp: karşılaştırma kartı ve "Onaylıyorum" butonu; müşteri onayladı - ⇢Rıza kaydedildi, tarife değişikliği bir sonraki dönemden geçerli provizyonlandı - ➤SMS: yeni tarife özeti ve yürürlük tarihi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici "İnternetim hep bitiyor" mesajı ya da 3 ay üst üste ek paket Sonuç Yeni tarife provizyonlandı, rıza zaman damgalı, müşteri yürürlük tarihini biliyor ## Ajanlar SÜPERVİZÖR ### Tarife Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Kullanım Ajanı Kullanım geçmişini çeker ve gelecek dönemi tahmin eder get_usage_historyget_current_planforecast_usage aginies-tulpar-4 · on-prem ### Teklif Ajanı Uygun tarifeleri karşılaştırır, teklifi sunar, yanıtı yorumlar get_eligible_planscompare_planscheck_campaign_rulessend_whatsappparse_reply aginies-sunkar-1 · on-prem ### Provizyon Ajanı Rızayı kaydeder, tarifeyi değiştirir, müşteriyi bilgilendirir record_consentchange_plansend_sms ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Tarife Süpervizörü | WhatsApp: "her ay internetim bitiyor, ne yapmalıyım?" | — | 180 | — | | 02 | araç | Kullanım Ajanı | Son 6 ay: ortalama 28 GB kullanım, 3 ayda ek paket alındı (toplam 210 TL) | Billing (Amdocs) | 460 | — | | 03 | akıl yürütme | Kullanım Ajanı | Mevcut tarife 20 GB / 349 TL; tahmin 32 GB'a çıkıyor → yükseltme uygun | — | 620 | 560 | | 04 | plan | Tarife Süpervizörü | Plan: uygun tarifeler → karşılaştırma → guardrail → teklif → rıza → provizyon | — | 240 | 180 | | 05 | araç | Teklif Ajanı | Uygun tarifeler: 40 GB / 419 TL, 60 GB / 479 TL; taahhüt yok | Product Catalogue / Offer Engine | 380 | — | | 06 | akıl yürütme | Teklif Ajanı | Öneri: 40 GB / 419 TL → ek paketlere kıyasla ayda yaklaşık 70 TL tasarruf | — | 680 | 640 | | 07 | doğrulama | Teklif Ajanı | Guardrail: teklif kampanya kurallarına uygun, fiyat ve taahhüt bilgisi doğru | — | 360 | 280 | | 08 | bildirim | Teklif Ajanı | WhatsApp: karşılaştırma kartı ve "Onaylıyorum" butonu; müşteri onayladı | WhatsApp Business API | 1.400 | 300 | | 09 | yazma | Provizyon Ajanı | Rıza kaydedildi, tarife değişikliği bir sonraki dönemden geçerli provizyonlandı | HLR / Provisioning | 520 | — | | 10 | bildirim | Provizyon Ajanı | SMS: yeni tarife özeti ve yürürlük tarihi | SMS Gateway | 220 | — | Simülasyon metrikleri (tahmini) 52.000çalıştırma / ay %95,8başarı 5,1 sp50 $0,001tahmini maliyet / çalıştırma Onay kuralı Taahhütlü veya indirimli özel teklif → kampanya yöneticisi onayı Adım profili giriş · 1plan · 1akıl yürütme · 2araç · 2doğrulama · 1yazma · 1bildirim · 2 SİSTEMLER - Billing (Amdocs)core - Product Catalogue / Offer Engineapi - HLR / Provisioningcore - WhatsApp Business APIchannel - SMS Gatewaychannel Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Operatörlerin kullanım verisiyle "doğru tarife" önerisi deseninin provizyonu da yapan sürümü Bu senaryonun yer aldığı paketler Satış ve lead yönetimi →Abone hizmetleri ve şebeke desteği → Sektör: Telekom → İlgili ajanlar TelekomHITL ### Abone hizmet asistanı: fatura, paket, arıza ve taahhüt Abone arar ya da yazar; fatura itirazı, paket değişikliği ve hat arızası aynı oturumda BSS ve şebeke sistemlerinde çözülür. 410.000 çalıştırma / ay %94,3 başarı 5 ajan TelekomHITL ### Fatura itirazı çözümü Fatura kalemleri ve kullanım kayıtları saniyede açıklanır; iyi niyet kredisi politikaya göre uygulanır. 88.000 çalıştırma / ay %96,3 başarı 4 ajan TelekomHITL ### SIM değişimi dolandırıcılık kontrolü Her SIM değişimi provizyondan önce risk skorlanır; şüpheli talepler bekletilir ve bankalara sinyal iletilir. 41.000 çalıştırma / ay %97,7 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Platform bakım (MRO) iş emri hazırlığı URL: https://www.aginies.com/tr/ajanlar/platform-mro-work-order Language: tr Summary: Arıza raporu teknik yayınlarla eşlenir, görev kartı ve parça listesi çıkar, iş emri planlayıcı onayıyla açılır. Uçuş sonrası arıza kaydı ya da bakım yönetim sisteminden zamanlı bakım tetiği → İş emri doğru görev kartları ve rezerve parçalarla açıldı; teknisyen ekibe atandı Savunma Sanayiiappapibatch # Platform bakım (MRO) iş emri hazırlığı Arıza raporu teknik yayınlarla eşlenir, görev kartı ve parça listesi çıkar, iş emri planlayıcı onayıyla açılır. Kısaca Platform bakım (MRO) iş emri hazırlığı, Savunma Sanayii sektöründe app, api, batch kanalında çalışan bir ajan senaryosudur. Tetikleyici: Uçuş sonrası arıza kaydı ya da bakım yönetim sisteminden zamanlı bakım tetiği. Bakım Süpervizörü 4 uzman ajanı (Arıza Ajanı, Teknik Yayın Ajanı, Parça Ajanı, Planlama Ajanı) 4 sistem üzerinde yönetir (Maintenance Management System, Technical Publications (S1000D vector), Spares Inventory, Technician App); 9 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: İş emri doğru görev kartları ve rezerve parçalarla açıldı; teknisyen ekibe atandı. Onay kuralı: Uçuşa elverişliliği etkileyen her iş paketi → bakım planlayıcısı onayı. Savunma Sanayii ### Platform bakım (MRO) iş emri hazırlığı [Platform bakım (MRO) iş emri hazırlığı] çalıştırma #41.286.988süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Uçuşa elverişliliği etkileyen her iş paketi → bakım planlayıcısı onayı Bu çalıştırmadaki ajanlar Bakım Süpervizörüsupervisorplanlar, yönlendirir, onaylar Arıza Ajanıaginies-talay-2 · on-premArıza kaydını ve geçmişi yorumlar Teknik Yayın Ajanıaginies-tulpar-4 · on-premBakım el kitabından görev kartlarını bulur Parça Ajanıaginies-sunkar-1 · on-premParça, alet ve stok uygunluğunu kontrol eder Planlama Ajanıaginies-talay-2 · on-premİş emrini açar ve ekibi atar Denetim izi00:00.00 - ◦Arıza kaydı F-30917: kuyruk numarası 07, “hidrolik basınç uyarısı, sistem 2” - ⚙Bileşen geçmişi çekildi: pompa P2 son 400 saatte 2 benzer uyarı - ∿Arıza sınıfı: hidrolik pompa performans düşüşü; olası neden: iç sızıntı, öncelik uçuşa elverişlilik - ⌕Görev kartları bulundu: 29-11-05 pompa test, 29-11-07 değişim; revizyon güncel - ⚙IPC parça listesi: pompa P/N 4211-88, 2 conta; stokta 1 pompa, 6 conta - ✓Alet listesi ve kalibrasyon tarihleri uygun; tahmini 6,5 adam-saat - ⏸Uçuşa elverişlilik etkisi → bakım planlayıcısı iş paketini onayladı - ⇢İş emri WO-71822 açıldı, parçalar rezerve edildi, ekip B vardiyasına atandı - ➤Teknisyen uygulamasına görev kartları, parça listesi ve emniyet notları gönderildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Uçuş sonrası arıza kaydı ya da bakım yönetim sisteminden zamanlı bakım tetiği Sonuç İş emri doğru görev kartları ve rezerve parçalarla açıldı; teknisyen ekibe atandı ## Ajanlar SÜPERVİZÖR ### Bakım Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Arıza Ajanı Arıza kaydını ve geçmişi yorumlar get_fault_recordget_component_historyclassify_fault aginies-tulpar-4 · on-prem ### Teknik Yayın Ajanı Bakım el kitabından görev kartlarını bulur search_tech_pubsget_task_cardcheck_revision aginies-sunkar-1 · on-prem ### Parça Ajanı Parça, alet ve stok uygunluğunu kontrol eder get_ipc_partscheck_stockreserve_parts aginies-talay-2 · on-prem ### Planlama Ajanı İş emrini açar ve ekibi atar create_work_orderassign_crewnotify_crew ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Bakım Süpervizörü | Arıza kaydı F-30917: kuyruk numarası 07, “hidrolik basınç uyarısı, sistem 2” | — | 200 | — | | 02 | araç | Arıza Ajanı | Bileşen geçmişi çekildi: pompa P2 son 400 saatte 2 benzer uyarı | Maintenance Management System | 440 | — | | 03 | akıl yürütme | Arıza Ajanı | Arıza sınıfı: hidrolik pompa performans düşüşü; olası neden: iç sızıntı, öncelik uçuşa elverişlilik | — | 780 | 980 | | 04 | bilgi | Teknik Yayın Ajanı | Görev kartları bulundu: 29-11-05 pompa test, 29-11-07 değişim; revizyon güncel | Technical Publications (S1000D vector) | 620 | 760 | | 05 | araç | Parça Ajanı | IPC parça listesi: pompa P/N 4211-88, 2 conta; stokta 1 pompa, 6 conta | Spares Inventory | 420 | — | | 06 | doğrulama | Parça Ajanı | Alet listesi ve kalibrasyon tarihleri uygun; tahmini 6,5 adam-saat | — | 360 | 240 | | 07 | insan onayı | Bakım Süpervizörü | Uçuşa elverişlilik etkisi → bakım planlayıcısı iş paketini onayladı | — | 2.800 | — | | 08 | yazma | Planlama Ajanı | İş emri WO-71822 açıldı, parçalar rezerve edildi, ekip B vardiyasına atandı | Maintenance Management System | 520 | — | | 09 | bildirim | Planlama Ajanı | Teknisyen uygulamasına görev kartları, parça listesi ve emniyet notları gönderildi | Technician App | 260 | — | Simülasyon metrikleri (tahmini) 4.100çalıştırma / ay %95,7başarı 6,4 sp50 $0,007tahmini maliyet / çalıştırma Onay kuralı Uçuşa elverişliliği etkileyen her iş paketi → bakım planlayıcısı onayı Adım profili giriş · 1bilgi · 1akıl yürütme · 1araç · 2doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - Maintenance Management Systemerp - Technical Publications (S1000D vector)db - Spares Inventoryerp - Technician Appchannel Çalıştırmak için gerekenler - ERP/MES için kayıt ve sorgu API erişimi - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Havacılık ve savunma MRO organizasyonlarının S1000D tabanlı görev kartı ve iş emri otomasyonunu yansıtır Bu senaryonun yer aldığı paketler Saha servis ve sevk otomasyonu →Savunma ihracat kontrolü ve tedarik zinciri güvenliği → Sektör: Savunma Sanayii → İlgili ajanlar Savunma SanayiiHITL ### Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolü Teknisyen hangarda sorar; doğru revizyondaki prosedür, parça numarası ve emniyet notu yetki kontrolüyle gelir, her erişim iz bırakır. 9.800 çalıştırma / ay %95,7 başarı 5 ajan Savunma SanayiiHITL ### İhracat kontrolü ve son kullanıcı taraması Parça, alıcı ve son kullanım tek akışta sınıflandırılır; lisans gereği tespit edilir, karar her zaman uyum yöneticisinde kalır. 2.600 çalıştırma / ay %96,1 başarı 4 ajan Savunma SanayiiHITL ### Tedarikçi güvenlik ve uyum değerlendirmesi Tedarikçi anketleri ve belgeleri okunur, tesis güvenlik gereklilikleriyle karşılaştırılır, eksikler listelenir; nitelendirme kararı güvenlik müdürünün. 900 çalıştırma / ay %95,2 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Poliçe yenileme teklifi URL: https://www.aginies.com/tr/ajanlar/policy-renewal Language: tr Summary: Poliçeler yeniden fiyatlanır, sert artışlar onaya gider; müşteri tek dokunuşla onaylar, poliçe anında kesilir. Toplu iş: 30 gün içinde süresi dolacak poliçeler → Yeni poliçe tanzim edildi, rıza zaman damgalı, PDF ve ödeme linki müşteride Sigortabatchwhatsappemailapp # Poliçe yenileme teklifi Poliçeler yeniden fiyatlanır, sert artışlar onaya gider; müşteri tek dokunuşla onaylar, poliçe anında kesilir. Kısaca Poliçe yenileme teklifi, Sigorta sektöründe batch, whatsapp, email, app kanalında çalışan bir ajan senaryosudur. Tetikleyici: Toplu iş: 30 gün içinde süresi dolacak poliçeler. Yenileme Süpervizörü 3 uzman ajanı (Fiyatlama Ajanı, Teklif Ajanı, Tanzim Ajanı) 4 sistem üzerinde yönetir (Policy Admin System, Rating Engine, WhatsApp Business API, Email (Exchange)); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Yeni poliçe tanzim edildi, rıza zaman damgalı, PDF ve ödeme linki müşteride. Onay kuralı: Prim artışı > %20 veya indirim > %10 → yenileme ekibi lideri onayı. Sigorta ### Poliçe yenileme teklifi [Poliçe yenileme teklifi] çalıştırma #41.288.023süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Prim artışı > %20 veya indirim > %10 → yenileme ekibi lideri onayı Bu çalıştırmadaki ajanlar Yenileme Süpervizörüsupervisorplanlar, yönlendirir, onaylar Fiyatlama Ajanıaginies-talay-2 · on-premHasar geçmişi ve fiyatlama motoruyla yeni primi hesaplar Teklif Ajanıaginies-tulpar-4 · on-premTeklifi yazar, gönderir, yanıtı ve rızayı yorumlar Tanzim Ajanıaginies-sunkar-1 · on-premRızayı kaydeder, poliçeyi tanzim eder, ödeme linki üretir Denetim izi00:00.00 - ◦Toplu iş: 30 gün içinde bitecek 8.120 kasko poliçesi listelendi - ⚙Poliçe #KS-1187720: 2 yıl hasarsız, 2021 model araç, hasarsızlık %30 - ⚙Fiyatlama motoru: yeni prim 19.700 TL (geçen yıl 15.900 TL, +%23,9) - ⌘Plan: artış kontrolü → onay → teklif → rıza → tanzim - ✓Artış %23,9 > %20 eşiği → sadakat indirimi önerisi %8, onay gerekli - ⏸Artış > %20 → yenileme ekibi lideri %8 indirimi onayladı → 18.124 TL - ➤WhatsApp'tan teklif özeti ve "Onaylıyorum" butonu gönderildi - ∿Müşteri onayladı; açık rıza metni ve KVKK bilgilendirmesi kabul edildi - ⇢Rıza zaman damgasıyla kaydedildi, poliçe #KS-1187720/2 tanzim edildi - ➤Poliçe PDF'i ve ödeme linki e-posta ile gönderildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Toplu iş: 30 gün içinde süresi dolacak poliçeler Sonuç Yeni poliçe tanzim edildi, rıza zaman damgalı, PDF ve ödeme linki müşteride ## Ajanlar SÜPERVİZÖR ### Yenileme Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Fiyatlama Ajanı Hasar geçmişi ve fiyatlama motoruyla yeni primi hesaplar get_policyget_claim_historycalculate_premiumapply_discounts aginies-tulpar-4 · on-prem ### Teklif Ajanı Teklifi yazar, gönderir, yanıtı ve rızayı yorumlar draft_offersend_whatsappsend_emailparse_reply aginies-sunkar-1 · on-prem ### Tanzim Ajanı Rızayı kaydeder, poliçeyi tanzim eder, ödeme linki üretir record_consentissue_policycreate_payment_link ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Yenileme Süpervizörü | Toplu iş: 30 gün içinde bitecek 8.120 kasko poliçesi listelendi | — | 320 | — | | 02 | araç | Fiyatlama Ajanı | Poliçe #KS-1187720: 2 yıl hasarsız, 2021 model araç, hasarsızlık %30 | Policy Admin System | 410 | — | | 03 | araç | Fiyatlama Ajanı | Fiyatlama motoru: yeni prim 19.700 TL (geçen yıl 15.900 TL, +%23,9) | Rating Engine | 520 | — | | 04 | plan | Yenileme Süpervizörü | Plan: artış kontrolü → onay → teklif → rıza → tanzim | — | 260 | 180 | | 05 | doğrulama | Fiyatlama Ajanı | Artış %23,9 > %20 eşiği → sadakat indirimi önerisi %8, onay gerekli | — | 580 | 520 | | 06 | insan onayı | Yenileme Süpervizörü | Artış > %20 → yenileme ekibi lideri %8 indirimi onayladı → 18.124 TL | — | 2.600 | — | | 07 | bildirim | Teklif Ajanı | WhatsApp'tan teklif özeti ve "Onaylıyorum" butonu gönderildi | WhatsApp Business API | 260 | 340 | | 08 | akıl yürütme | Teklif Ajanı | Müşteri onayladı; açık rıza metni ve KVKK bilgilendirmesi kabul edildi | — | 420 | 210 | | 09 | yazma | Tanzim Ajanı | Rıza zaman damgasıyla kaydedildi, poliçe #KS-1187720/2 tanzim edildi | Policy Admin System | 540 | — | | 10 | bildirim | Tanzim Ajanı | Poliçe PDF'i ve ödeme linki e-posta ile gönderildi | Email (Exchange) | 240 | — | Simülasyon metrikleri (tahmini) 22.000çalıştırma / ay %94,4başarı 6,2 sp50 $0,0015tahmini maliyet / çalıştırma Onay kuralı Prim artışı > %20 veya indirim > %10 → yenileme ekibi lideri onayı Adım profili giriş · 1plan · 1akıl yürütme · 1araç · 2doğrulama · 1insan onayı · 1yazma · 1bildirim · 2 SİSTEMLER - Policy Admin Systemcore - Rating Engineapi - WhatsApp Business APIchannel - Email (Exchange)email Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek AB sigortacılarının fiyat şeffaflığı kurallarına göre kurgulanmış otomatik yenileme deseni Bu senaryonun yer aldığı paketler Underwriting ve yenileme operasyonları → Sektör: Sigorta → İlgili ajanlar SigortaHITL ### Uçtan uca kasko hasarı: ihbardan ödemeye Sesli ihbar, fotoğraf ekspertizi, anlaşmalı servis ataması ve ödeme talimatı tek akışta; eksper yalnızca eşik üstünde devreye girer. 21.000 çalıştırma / ay %91,2 başarı 5 ajan SigortaHITL ### Sesli hasar ihbarı (FNOL) Müşteri arar ve anlatır; poliçe plakadan bulunur, teminat kontrol edilir, dosya açılır ve eksper atanır. 34.000 çalıştırma / ay %95,8 başarı 5 ajan SigortaHITL ### Underwriting belge çıkarımı Sağlık raporu ve tahliller okunur, kılavuza göre değerlendirilir; underwriter yalnızca eşik üstünü görür. 4.800 çalıştırma / ay %94,9 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Ön yetkilendirme (prior authorization) URL: https://www.aginies.com/tr/ajanlar/prior-authorization Language: tr Summary: İstem düşer düşmez ödeyici kuralları taranır, klinik kanıt toplanır, dosya dakikalar içinde gönderime hazır. EHR'de ön yetki gerektiren bir işlem istemi oluşturulması → Ödeyiciye eksiksiz dosya gönderildi, yetki numarası EHR'ye yazıldı, ekip bilgilendirildi Sağlıkapiemailweb # Ön yetkilendirme (prior authorization) İstem düşer düşmez ödeyici kuralları taranır, klinik kanıt toplanır, dosya dakikalar içinde gönderime hazır. Kısaca Ön yetkilendirme (prior authorization), Sağlık sektöründe api, email, web kanalında çalışan bir ajan senaryosudur. Tetikleyici: EHR'de ön yetki gerektiren bir işlem istemi oluşturulması. Ön Yetki Süpervizörü 3 uzman ajanı (İstem Ajanı, Ödeyici Kural Ajanı, Gönderim Ajanı) 4 sistem üzerinde yönetir (EHR (HL7/FHIR), Payer Policy Vector Store, Payer Portal, Exchange / M365 Mail); 10 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Ödeyiciye eksiksiz dosya gönderildi, yetki numarası EHR'ye yazıldı, ekip bilgilendirildi. Onay kuralı: Cerrahi işlem veya maliyet > 25.000 TL → kullanım yönetimi hemşiresi onayı. Sağlık ### Ön yetkilendirme (prior authorization) [Ön yetkilendirme (prior authorization)] çalıştırma #41.287.578süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Cerrahi işlem veya maliyet > 25.000 TL → kullanım yönetimi hemşiresi onayı Bu çalıştırmadaki ajanlar Ön Yetki Süpervizörüsupervisorplanlar, yönlendirir, onaylar İstem Ajanıaginies-konrul-1 · on-premİstem ve klinik belgelerden alanları çıkarır Ödeyici Kural Ajanıaginies-tulpar-4 · on-premÖdeyici politikasını arar ve tıbbi gerekliliği değerlendirir Gönderim Ajanıaginies-talay-2 · on-premDosyayı derler, portala gönderir ve kaydı günceller Denetim izi00:00.00 - ◦EHR'den OR-58213 numaralı lomber füzyon istemi alındı, hasta kimliği eşleşti - ∿CPT 22612 · ICD-10 M43.16 · Hekim: Dr. Yılmaz · Ödeyici: Plan A · Maliyet 185.000 TL - ⌘Plan: klinik notlar → ödeyici kuralı → gereklilik kontrolü → gönderim - ⌕Son 6 ayın klinik notları, MR raporu ve 8 haftalık fizyoterapi kaydı çekildi - ⌕Ödeyici politikası SP-L-2026 §4.2: ≥ 6 hafta konservatif tedavi şartı bulundu - ✓Gereklilik: 8 hafta FT belgeli, nörolojik bulgu mevcut → kriter sağlandı (güven 0,93) - ⏸Cerrahi işlem → kullanım yönetimi hemşiresi dosyayı inceledi ve onayladı - ⇢Dosya (12 sayfa) ödeyici portalına gönderildi, referans PA-2026-118402 alındı - ⇢Yetki numarası ve 90 günlük geçerlilik encounter kaydına yazıldı - ➤Hekim ofisi ve ameliyat planlama bilgilendirildi, karar izi kaydedildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici EHR'de ön yetki gerektiren bir işlem istemi oluşturulması Sonuç Ödeyiciye eksiksiz dosya gönderildi, yetki numarası EHR'ye yazıldı, ekip bilgilendirildi ## Ajanlar SÜPERVİZÖR ### Ön Yetki Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### İstem Ajanı İstem ve klinik belgelerden alanları çıkarır extract_order_fieldsextract_clinical_notes aginies-tulpar-4 · on-prem ### Ödeyici Kural Ajanı Ödeyici politikasını arar ve tıbbi gerekliliği değerlendirir search_payer_policycheck_medical_necessity aginies-talay-2 · on-prem ### Gönderim Ajanı Dosyayı derler, portala gönderir ve kaydı günceller build_auth_packetsubmit_prior_authupdate_encountersend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Ön Yetki Süpervizörü | EHR'den OR-58213 numaralı lomber füzyon istemi alındı, hasta kimliği eşleşti | — | 200 | — | | 02 | akıl yürütme | İstem Ajanı | CPT 22612 · ICD-10 M43.16 · Hekim: Dr. Yılmaz · Ödeyici: Plan A · Maliyet 185.000 TL | — | 720 | 640 | | 03 | plan | Ön Yetki Süpervizörü | Plan: klinik notlar → ödeyici kuralı → gereklilik kontrolü → gönderim | — | 300 | 210 | | 04 | bilgi | İstem Ajanı | Son 6 ayın klinik notları, MR raporu ve 8 haftalık fizyoterapi kaydı çekildi | EHR (HL7/FHIR) | 460 | — | | 05 | bilgi | Ödeyici Kural Ajanı | Ödeyici politikası SP-L-2026 §4.2: ≥ 6 hafta konservatif tedavi şartı bulundu | Payer Policy Vector Store | 520 | 880 | | 06 | doğrulama | Ödeyici Kural Ajanı | Gereklilik: 8 hafta FT belgeli, nörolojik bulgu mevcut → kriter sağlandı (güven 0,93) | — | 840 | 1.100 | | 07 | insan onayı | Ön Yetki Süpervizörü | Cerrahi işlem → kullanım yönetimi hemşiresi dosyayı inceledi ve onayladı | — | 2.600 | — | | 08 | yazma | Gönderim Ajanı | Dosya (12 sayfa) ödeyici portalına gönderildi, referans PA-2026-118402 alındı | Payer Portal | 900 | — | | 09 | yazma | Gönderim Ajanı | Yetki numarası ve 90 günlük geçerlilik encounter kaydına yazıldı | EHR (HL7/FHIR) | 380 | — | | 10 | bildirim | Gönderim Ajanı | Hekim ofisi ve ameliyat planlama bilgilendirildi, karar izi kaydedildi | Exchange / M365 Mail | 240 | — | Simülasyon metrikleri (tahmini) 6.200çalıştırma / ay %94,1başarı 7,2 sp50 $0,0111tahmini maliyet / çalıştırma Onay kuralı Cerrahi işlem veya maliyet > 25.000 TL → kullanım yönetimi hemşiresi onayı Adım profili giriş · 1plan · 1bilgi · 2akıl yürütme · 1doğrulama · 1insan onayı · 1yazma · 2bildirim · 1 SİSTEMLER - EHR (HL7/FHIR)core - Payer Policy Vector Storedb - Payer Portalapi - Exchange / M365 Mailemail Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek ABD sağlık sistemlerinin ve ödeyicilerin ön yetki otomasyonu programlarını yansıtır Bu senaryonun yer aldığı paketler Bilgi yönetimi ve karar desteği →Sağlık operasyonları ve klinik yönlendirme → Sektör: Sağlık → İlgili ajanlar SağlıkHITL ### Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuç Hasta yazar ya da arar; randevu, ön hazırlık talimatı, sigorta ön onayı ve sonuç bildirimi tek koordinatörde birleşir. 58.000 çalıştırma / ay %95,1 başarı 5 ajan SağlıkHITL ### Hasta ön kabul ve triyaj Hasta yazar ya da anlatır; anamnez ve aciliyet skoru saniyeler sürer, son kararı her zaman hemşire verir. 24.500 çalıştırma / ay %95,8 başarı 5 ajan Sağlık ### Randevu planlama ve no-show azaltma Gelmeme riski yüksek hastalar WhatsApp'tan önceden aranır, boşalan slotlar bekleme listesinden anında dolar. 62.000 çalıştırma / ay %98,2 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada URL: https://www.aginies.com/tr/ajanlar/procure-to-pay-desk Language: tr Summary: Talep e-postayla gelir; teklifler karşılaştırılır, sözleşme maddeleri kontrol edilir, sipariş açılır, fatura üç yönlü eşleşir ve ödeme planlanır. Satın alma talebi e-postası veya ERP'de açılan talep formu → Sipariş açıldı, sözleşme sapmaları işaretlendi, fatura eşleşti, ödeme vadesinde planlandı Kurumsal Fonksiyonlaremailapi # Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada Talep e-postayla gelir; teklifler karşılaştırılır, sözleşme maddeleri kontrol edilir, sipariş açılır, fatura üç yönlü eşleşir ve ödeme planlanır. Kısaca Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada, Kurumsal Fonksiyonlar sektöründe email, api kanalında çalışan bir ajan senaryosudur. Tetikleyici: Satın alma talebi e-postası veya ERP'de açılan talep formu. Satın Alma Süpervizörü 4 uzman ajanı (Talep Ajanı, Tedarik Ajanı, Sözleşme Ajanı, Fatura Ajanı) 5 sistem üzerinde yönetir (Procurement Inbox, ERP, Supplier Portal, Contract Repository, Payment Hub); 10 adımın 3 tanesi sistemde işlem, 4 tanesi doğrulama veya onaydır. Sonuç: Sipariş açıldı, sözleşme sapmaları işaretlendi, fatura eşleşti, ödeme vadesinde planlandı. Onay kuralı: Tutar > 250.000 TL, tek teklif veya çerçeve dışı tedarikçi → satın alma komitesi. Kurumsal Fonksiyonlar ### Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada [Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada] çalıştırma #41.287.971süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Tutar > 250.000 TL, tek teklif veya çerçeve dışı tedarikçi → satın alma komitesi Bu çalıştırmadaki ajanlar Satın Alma Süpervizörüsupervisorplanlar, yönlendirir, onaylar Talep Ajanıaginies-sunkar-1 · on-premTalebi sınıflandırır, bütçe kalemini ve onay zincirini belirler Tedarik Ajanıaginies-tulpar-4 · on-premTeklifleri toplar, normalize eder, toplam sahip olma maliyetiyle karşılaştırır Sözleşme Ajanıaginies-talay-2 · on-premTeklif şartlarını şirket şablonu ve mevcut çerçeve sözleşmeyle karşılaştırır Fatura Ajanıaginies-konrul-1 · on-premFaturayı okur, sipariş ve teslimatla eşleştirir, ödemeyi planlar Denetim izi00:00.00 - ◦E-posta: "40 adet dizüstü bilgisayar, Q4 bütçesinden, 3 hafta içinde" + şartname eki - ✓Kategori BT donanım, bütçe kalemi IT-CAPEX-Q4 uygun, onay zinciri: BT müdürü + finans - ⌘Plan: 3 tedarikçiden teklif → normalize → sözleşme kontrolü → onay → sipariş → fatura → ödeme - ⇢Çerçeve sözleşmeli 3 tedarikçiye teklif talebi portaldan gönderildi, son tarih 3 iş günü - ∿Teklifler: 1,84M / 1,91M / 1,79M TL; garanti ve teslimat dahil TCO sıralaması: B, A, C - ✓Tedarikçi B: ödeme vadesi 30 gün (çerçeve 45), ceza maddesi eksik → 2 sapma işaretlendi - ⏸Karşılaştırma ve sapmalar BT müdürü ve finansa sunuldu; B, 45 gün vadeyle onaylandı - ⇢Sipariş #PO-88213 açıldı, tedarikçiye iletildi, teslimat 18 gün - ✓Fatura okundu: 40 adet, 1,79M TL; sipariş ve irsaliye ile üç yönlü eşleşme tam - ⇢Ödeme 45. güne planlandı, erken ödeme indirimi yok; talep sahibine kapanış e-postası - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Satın alma talebi e-postası veya ERP'de açılan talep formu Sonuç Sipariş açıldı, sözleşme sapmaları işaretlendi, fatura eşleşti, ödeme vadesinde planlandı ## Ajanlar SÜPERVİZÖR ### Satın Alma Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Talep Ajanı Talebi sınıflandırır, bütçe kalemini ve onay zincirini belirler classify_requestcheck_budgetroute_approval aginies-tulpar-4 · on-prem ### Tedarik Ajanı Teklifleri toplar, normalize eder, toplam sahip olma maliyetiyle karşılaştırır request_quotesnormalise_quotescompare_tco aginies-talay-2 · on-prem ### Sözleşme Ajanı Teklif şartlarını şirket şablonu ve mevcut çerçeve sözleşmeyle karşılaştırır get_frameworkcompare_clausesflag_deviation aginies-konrul-1 · on-prem ### Fatura Ajanı Faturayı okur, sipariş ve teslimatla eşleştirir, ödemeyi planlar extract_invoicethree_way_matchschedule_payment ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Satın Alma Süpervizörü | E-posta: "40 adet dizüstü bilgisayar, Q4 bütçesinden, 3 hafta içinde" + şartname eki | Procurement Inbox | 180 | — | | 02 | doğrulama | Talep Ajanı | Kategori BT donanım, bütçe kalemi IT-CAPEX-Q4 uygun, onay zinciri: BT müdürü + finans | ERP | 600 | 260 | | 03 | plan | Satın Alma Süpervizörü | Plan: 3 tedarikçiden teklif → normalize → sözleşme kontrolü → onay → sipariş → fatura → ödeme | — | 240 | 220 | | 04 | yazma | Tedarik Ajanı | Çerçeve sözleşmeli 3 tedarikçiye teklif talebi portaldan gönderildi, son tarih 3 iş günü | Supplier Portal | 700 | — | | 05 | akıl yürütme | Tedarik Ajanı | Teklifler: 1,84M / 1,91M / 1,79M TL; garanti ve teslimat dahil TCO sıralaması: B, A, C | Supplier Portal | 2.400 | 1.400 | | 06 | doğrulama | Sözleşme Ajanı | Tedarikçi B: ödeme vadesi 30 gün (çerçeve 45), ceza maddesi eksik → 2 sapma işaretlendi | Contract Repository | 1.500 | 820 | | 07 | insan onayı | Talep Ajanı | Karşılaştırma ve sapmalar BT müdürü ve finansa sunuldu; B, 45 gün vadeyle onaylandı | ERP | 1.200 | — | | 08 | yazma | Tedarik Ajanı | Sipariş #PO-88213 açıldı, tedarikçiye iletildi, teslimat 18 gün | ERP | 560 | — | | 09 | doğrulama | Fatura Ajanı | Fatura okundu: 40 adet, 1,79M TL; sipariş ve irsaliye ile üç yönlü eşleşme tam | ERP | 1.800 | 900 | | 10 | yazma | Fatura Ajanı | Ödeme 45. güne planlandı, erken ödeme indirimi yok; talep sahibine kapanış e-postası | Payment Hub | 480 | — | Simülasyon metrikleri (tahmini) 14.000çalıştırma / ay %92,7başarı 9,1 sp50 $0,0055tahmini maliyet / çalıştırma Onay kuralı Tutar > 250.000 TL, tek teklif veya çerçeve dışı tedarikçi → satın alma komitesi Adım profili giriş · 1plan · 1akıl yürütme · 1doğrulama · 3insan onayı · 1yazma · 3 SİSTEMLER - Procurement Inboxemail - ERPerp - Supplier Portalapi - Contract Repositorystorage - Payment Hubpayments Çalıştırmak için gerekenler - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - ERP/MES için kayıt ve sorgu API erişimi - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Doküman deposuna okuma erişimi ve belge tipleri - Ödeme servisleri için yetkili API erişimi ve limitler - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük kurumların kaynak-ödeme (S2P) otomasyon deseni; sözleşme kontrolü ve üç yönlü eşleşmeyle Bu senaryonun yer aldığı paketler Fatura ve doküman veri çıkarımı → İlgili ajanlar Kurumsal Fonksiyonlar ### E-posta sınıflandırma ve ek kontrolü Ortak kutuya gelen her e-posta sınıflandırılır, ekler listeyle karşılaştırılır, eksikler anında sorulur. 86.000 çalıştırma / ay %97,6 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Çağrı merkezi kalite izleme Örneklem değil, tüm çağrılar rubriğe göre puanlanır; uyum ihlali ve koçluk fırsatı sabah liderin önünde. 90.000 çalıştırma / ay %98,8 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Sözleşme madde çıkarımı Sözleşme maddeleri çıkarılır, hukuk playbook'u ile karşılaştırılır; sapmalar redline önerisiyle avukata gider. 1.400 çalıştırma / ay %95,1 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Ürün soru-cevap asistanı URL: https://www.aginies.com/tr/ajanlar/product-qa-assistant Language: tr Summary: Beden, uyumluluk ve özellik soruları katalog ve iade verisiyle yanıtlanır; stok ve fiyat doğrulanır. Ürün sayfasındaki sohbet penceresine yazılan soru → Kaynaklı yanıt ve doğru beden/ürün kartı sohbette, sepete ekleme tek dokunuş Perakende & E-ticaretwebappwhatsapp # Ürün soru-cevap asistanı Beden, uyumluluk ve özellik soruları katalog ve iade verisiyle yanıtlanır; stok ve fiyat doğrulanır. Kısaca Ürün soru-cevap asistanı, Perakende & E-ticaret sektöründe web, app, whatsapp kanalında çalışan bir ajan senaryosudur. Tetikleyici: Ürün sayfasındaki sohbet penceresine yazılan soru. Alışveriş Süpervizörü 3 uzman ajanı (Niyet Ajanı, Katalog Ajanı, Yanıt Ajanı) 4 sistem üzerinde yönetir (Vector Store, PIM / Catalogue, Shopify / OMS, Web Chat Widget); 9 adımın 0 tanesi sistemde işlem, 1 tanesi doğrulama veya onaydır. Sonuç: Kaynaklı yanıt ve doğru beden/ürün kartı sohbette, sepete ekleme tek dokunuş. Perakende & E-ticaret ### Ürün soru-cevap asistanı [Ürün soru-cevap asistanı] çalıştırma #41.287.575süre 0 mstoken 0tahmini maliyet $0,000durum — Bu çalıştırmadaki ajanlar Alışveriş Süpervizörüsupervisorplanlar, yönlendirir, onaylar Niyet Ajanıaginies-sunkar-1 · on-premSoruyu ve kısıtları (boy, kilo, cihaz modeli) çıkarır Katalog Ajanıaginies-tulpar-4 · on-premKatalog, beden tablosu ve iade verisiyle öneri üretir Yanıt Ajanıaginies-sunkar-1 · on-premYanıtı kaynaklara bağlar ve ürün kartıyla gönderir Denetim izi00:00.00 - ◦Web sohbeti: "Bu montun 42 bedeni bana olur mu? 1,78 boy, 80 kg" - ∿Niyet: beden uyumu · ürün SKU 88-4410 · boy 178 · kilo 80 - ⌘Plan: beden tablosu → iade verisi → stok → kaynak kontrolü → yanıt - ⌕Beden tablosu ve ürün kesim notu getirildi ("dar kalıp") - ⚙PIM: %80 yün, astarlı; iade verisi: 42 beden "küçük geldi" %31 - ∿Öneri: 44 beden (dar kalıp + iade verisi), güven %86 - ⚙Stok: 44 beden 7 adet, İstanbul deposu, yarın teslim - ✓Yanıt yalnızca katalog kaynaklarına dayanıyor; fiyat ve teslimat vaadi doğrulandı - ➤Yanıt ve 44 beden ürün kartı "sepete ekle" butonuyla sohbete gönderildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Ürün sayfasındaki sohbet penceresine yazılan soru Sonuç Kaynaklı yanıt ve doğru beden/ürün kartı sohbette, sepete ekleme tek dokunuş ## Ajanlar SÜPERVİZÖR ### Alışveriş Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Niyet Ajanı Soruyu ve kısıtları (boy, kilo, cihaz modeli) çıkarır detect_intentextract_product_refextract_constraints aginies-tulpar-4 · on-prem ### Katalog Ajanı Katalog, beden tablosu ve iade verisiyle öneri üretir search_catalogueget_product_specsget_size_chartcheck_compatibilityget_stock aginies-sunkar-1 · on-prem ### Yanıt Ajanı Yanıtı kaynaklara bağlar ve ürün kartıyla gönderir check_groundingcompose_answeradd_product_cardssend_reply ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Alışveriş Süpervizörü | Web sohbeti: "Bu montun 42 bedeni bana olur mu? 1,78 boy, 80 kg" | — | 160 | — | | 02 | akıl yürütme | Niyet Ajanı | Niyet: beden uyumu · ürün SKU 88-4410 · boy 178 · kilo 80 | — | 380 | 240 | | 03 | plan | Alışveriş Süpervizörü | Plan: beden tablosu → iade verisi → stok → kaynak kontrolü → yanıt | — | 220 | 160 | | 04 | bilgi | Katalog Ajanı | Beden tablosu ve ürün kesim notu getirildi ("dar kalıp") | Vector Store | 340 | — | | 05 | araç | Katalog Ajanı | PIM: %80 yün, astarlı; iade verisi: 42 beden "küçük geldi" %31 | PIM / Catalogue | 420 | — | | 06 | akıl yürütme | Katalog Ajanı | Öneri: 44 beden (dar kalıp + iade verisi), güven %86 | — | 720 | 810 | | 07 | araç | Katalog Ajanı | Stok: 44 beden 7 adet, İstanbul deposu, yarın teslim | Shopify / OMS | 300 | — | | 08 | doğrulama | Yanıt Ajanı | Yanıt yalnızca katalog kaynaklarına dayanıyor; fiyat ve teslimat vaadi doğrulandı | — | 360 | 280 | | 09 | bildirim | Yanıt Ajanı | Yanıt ve 44 beden ürün kartı "sepete ekle" butonuyla sohbete gönderildi | Web Chat Widget | 240 | 210 | Simülasyon metrikleri (tahmini) 120.000çalıştırma / ay %98,7başarı 3,2 sp50 $0,0004tahmini maliyet / çalıştırma Adım profili giriş · 1plan · 1bilgi · 1akıl yürütme · 2araç · 2doğrulama · 1bildirim · 1 SİSTEMLER - Vector Storedb - PIM / Cataloguedb - Shopify / OMSerp - Web Chat Widgetchannel Çalıştırmak için gerekenler - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - ERP/MES için kayıt ve sorgu API erişimi - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük moda perakendecilerinin iade verisiyle beslenen beden öneri motoru deseni Bu senaryonun yer aldığı paketler E-ticaret müşteri desteği ve aksiyon →Satış ve lead yönetimi → Sektör: Perakende & E-ticaret → İlgili ajanlar Perakende & E-ticaretHITL ### Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin Müşteri bir fotoğraf ve bir cümle gönderir; ajan stokta olan, bedenine uyan ve bütçeye sığan üç kombin kurar, seçileni sepete ekler. 142.000 çalıştırma / ay %96,1 başarı 5 ajan Perakende & E-ticaret ### Siparişim nerede? (WISMO) Müşteri sorar, sipariş ve kargo olayları anında birleşir; gecikmede yeni tarih verilir, kargo araştırılır. 110.000 çalıştırma / ay %98,1 başarı 4 ajan Perakende & E-ticaretHITL ### İade ve para iadesi Politika ve müşteri geçmişi saniyede kontrol edilir, etiket üretilir; iade tutarı kargo taramasıyla ödenir. 48.000 çalıştırma / ay %96,7 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Kamu ihalesi teklif dosyası ön kontrolü URL: https://www.aginies.com/tr/ajanlar/public-procurement-bid-check Language: tr Summary: Teklif dosyaları şartnameyle karşılaştırılır, eksik ve uygunsuz belgeler listelenir; değerlendirme kararı komisyonun. İhale son teklif saatinde elektronik teklif dosyalarının açılması → Her teklif için belge kontrol listesi ve uygunluk özeti hazır; komisyon tutanağı taslağı oluşturuldu Kamu ve Hükümetbatchweb # Kamu ihalesi teklif dosyası ön kontrolü Teklif dosyaları şartnameyle karşılaştırılır, eksik ve uygunsuz belgeler listelenir; değerlendirme kararı komisyonun. Kısaca Kamu ihalesi teklif dosyası ön kontrolü, Kamu ve Hükümet sektöründe batch, web kanalında çalışan bir ajan senaryosudur. Tetikleyici: İhale son teklif saatinde elektronik teklif dosyalarının açılması. İhale Süpervizörü 3 uzman ajanı (Belge Ajanı, Şartname Ajanı, Tutanak Ajanı) 3 sistem üzerinde yönetir (e-Procurement Platform, Specification Vector Store, Document Store); 9 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Her teklif için belge kontrol listesi ve uygunluk özeti hazır; komisyon tutanağı taslağı oluşturuldu. Onay kuralı: Uygunluk ve eleme kararı her zaman ihale komisyonunda; ajan yalnızca hazırlar. Kamu ve Hükümet ### Kamu ihalesi teklif dosyası ön kontrolü [Kamu ihalesi teklif dosyası ön kontrolü] çalıştırma #41.287.213süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Uygunluk ve eleme kararı her zaman ihale komisyonunda; ajan yalnızca hazırlar Bu çalıştırmadaki ajanlar İhale Süpervizörüsupervisorplanlar, yönlendirir, onaylar Belge Ajanıaginies-konrul-1 · on-premTeklif dosyasındaki belgeleri sınıflandırır ve okur Şartname Ajanıaginies-tulpar-4 · on-premİdari ve teknik şartname maddeleriyle karşılaştırır Tutanak Ajanıaginies-talay-2 · on-premKontrol listesini ve tutanak taslağını yazar Denetim izi00:00.00 - ◦İhale 2026/118203: 7 teklif, 214 belge, idari + teknik şartname 96 madde - ⚙Belgeler indirildi ve sınıflandırıldı: teminat, yeterlik, iş deneyimi, teknik ekler - ∿Teklif 3: geçici teminat tutarı %2,9 (min %3); Teklif 5: e-imza doğrulanamadı - ⌕Şartname maddeleri çekildi: yeterlik kriterleri 14, teknik zorunluluk 31 - ∿7 teklif eşlendi: 4 tam uygun, 2 eksik belge, 1 teknik uygunsuzluk adayı - ✓Her bulgu için madde numarası ve kaynak sayfa bağlandı; belirsiz 3 madde komisyona işaretlendi - ⇢Kontrol listeleri ve tutanak taslağı oluşturuldu; ihale kaydına eklendi - ⏸Uygunluk kararı → ihale komisyonu bulguları inceledi, 2 maddeyi düzeltti, tutanağı imzaladı - ➤Komisyon üyelerine imzalı tutanak ve bulgu özeti iletildi; süreç kaydı kapatıldı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici İhale son teklif saatinde elektronik teklif dosyalarının açılması Sonuç Her teklif için belge kontrol listesi ve uygunluk özeti hazır; komisyon tutanağı taslağı oluşturuldu ## Ajanlar SÜPERVİZÖR ### İhale Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### Belge Ajanı Teklif dosyasındaki belgeleri sınıflandırır ve okur extract_documentclassify_documentcheck_signature aginies-tulpar-4 · on-prem ### Şartname Ajanı İdari ve teknik şartname maddeleriyle karşılaştırır search_specificationmatch_requirementlist_gaps aginies-talay-2 · on-prem ### Tutanak Ajanı Kontrol listesini ve tutanak taslağını yazar create_checklistdraft_minutesupdate_tender ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | İhale Süpervizörü | İhale 2026/118203: 7 teklif, 214 belge, idari + teknik şartname 96 madde | e-Procurement Platform | 200 | — | | 02 | araç | Belge Ajanı | Belgeler indirildi ve sınıflandırıldı: teminat, yeterlik, iş deneyimi, teknik ekler | Document Store | 620 | — | | 03 | akıl yürütme | Belge Ajanı | Teklif 3: geçici teminat tutarı %2,9 (min %3); Teklif 5: e-imza doğrulanamadı | — | 1.100 | 1.600 | | 04 | bilgi | Şartname Ajanı | Şartname maddeleri çekildi: yeterlik kriterleri 14, teknik zorunluluk 31 | Specification Vector Store | 520 | 700 | | 05 | akıl yürütme | Şartname Ajanı | 7 teklif eşlendi: 4 tam uygun, 2 eksik belge, 1 teknik uygunsuzluk adayı | — | 1.400 | 2.100 | | 06 | doğrulama | Şartname Ajanı | Her bulgu için madde numarası ve kaynak sayfa bağlandı; belirsiz 3 madde komisyona işaretlendi | — | 480 | 380 | | 07 | yazma | Tutanak Ajanı | Kontrol listeleri ve tutanak taslağı oluşturuldu; ihale kaydına eklendi | e-Procurement Platform | 520 | — | | 08 | insan onayı | İhale Süpervizörü | Uygunluk kararı → ihale komisyonu bulguları inceledi, 2 maddeyi düzeltti, tutanağı imzaladı | — | 3.200 | — | | 09 | bildirim | Tutanak Ajanı | Komisyon üyelerine imzalı tutanak ve bulgu özeti iletildi; süreç kaydı kapatıldı | e-Procurement Platform | 240 | — | Simülasyon metrikleri (tahmini) 380çalıştırma / ay %94,8başarı 8,4 sp50 $0,0201tahmini maliyet / çalıştırma Onay kuralı Uygunluk ve eleme kararı her zaman ihale komisyonunda; ajan yalnızca hazırlar Adım profili giriş · 1bilgi · 1akıl yürütme · 2araç · 1doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - e-Procurement Platformapi - Specification Vector Storedb - Document Storestorage Çalıştırmak için gerekenler - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Doküman deposuna okuma erişimi ve belge tipleri - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Avrupa kamu alım kurumlarının belge kontrolü otomasyonu ve insan komisyon kararı modelini yansıtır Bu senaryonun yer aldığı paketler Kamu alımı ve ihale uyumu → Sektör: Kamu ve Hükümet → İlgili ajanlar Kamu ve HükümetHITL ### Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi Vatandaş yazar ya da arar; uygunluk kontrol edilir, belgeler okunur, başvuru açılır, randevu verilir, her adım izlenebilir. 230.000 çalıştırma / ay %93,8 başarı 5 ajan Kamu ve HükümetHITL ### Vatandaş başvurusu yönlendirme e-Devlet başvurusu saniyeler içinde sınıflandırılır, yasal süresi hesaplanır ve doğru birime SLA ile atanır. 54.000 çalıştırma / ay %98,1 başarı 4 ajan Kamu ve HükümetHITL ### Ruhsat başvurusu ön inceleme Eksik belge ve mevzuat uyumu dakikalar içinde raporlanır; kararı her zaman ruhsat memuru verir. 3.100 çalıştırma / ay %95,8 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Regresyon test döngüsü (kiralık test ajanı) URL: https://www.aginies.com/tr/ajanlar/regression-test-cycle Language: tr Summary: Her sürüm adayında regresyon koşulur, yeni hatalar kanıtla açılır, kapsama raporu yazılır; “go” kararı sürüm yöneticisinde. CI’dan sürüm adayı etiketi → Regresyon sonucu, açılan hatalar ve kapsama raporu sürüm kaydında; iş kaydı güncellendi Yazılım Geliştirmeapibatch # Regresyon test döngüsü (kiralık test ajanı) Her sürüm adayında regresyon koşulur, yeni hatalar kanıtla açılır, kapsama raporu yazılır; “go” kararı sürüm yöneticisinde. Kısaca Regresyon test döngüsü (kiralık test ajanı), Yazılım Geliştirme sektöründe api, batch kanalında çalışan bir ajan senaryosudur. Tetikleyici: CI’dan sürüm adayı etiketi. Test Süpervizörü 3 uzman ajanı (Test Plan Ajanı, Koşum Ajanı, Hata Ajanı) 4 sistem üzerinde yönetir (CI/CD, Device / Browser Grid, Jira / Xray, Workforce Work Log); 9 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Regresyon sonucu, açılan hatalar ve kapsama raporu sürüm kaydında; iş kaydı güncellendi. Onay kuralı: Sürüm “go / no-go” kararı sürüm yöneticisinde. Yazılım Geliştirme ### Regresyon test döngüsü (kiralık test ajanı) [Regresyon test döngüsü (kiralık test ajanı)] çalıştırma #41.289.840süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Sürüm “go / no-go” kararı sürüm yöneticisinde Bu çalıştırmadaki ajanlar Test Süpervizörüsupervisorplanlar, yönlendirir, onaylar Test Plan Ajanıaginies-tulpar-4 · on-premDeğişen alanlara göre test setini seçer ve genişletir Koşum Ajanıaginies-sunkar-1 · on-premTestleri koşar, kararsız testleri ayıklar Hata Ajanıaginies-talay-2 · on-premHataları kanıt ve adımlarla açar, raporu yazar Denetim izi00:00.00 - ◦RC 2.14.0-rc3: 23 commit, değişen alanlar ödeme, profil, bildirimler - ∿412 test seçildi (tam regresyon), 9 yeni senaryo kabul kriterlerinden üretildi - ⚙3 tarayıcı + 4 cihazda koşuldu: 409 geçti, 3 başarısız; 2’si tekrarda geçti (kararsız) - ✓Kalan 1 başarısızlık tutarlı: ödeme özeti yuvarlama farkı (0,01) - ⇢BUG-2291 açıldı: adımlar, ekran kaydı, log; öncelik yüksek, ödeme ekibine atandı - ∿Rapor: kapsama %91, 2 kararsız test karantinaya, kaçan hata 0 (önceki sürüm 1) - ⏸Sürüm yöneticisi: BUG-2291 düzeltilene kadar “no-go”; rc4 için yeniden koşum planlandı - ⇢İş kaydı: 412 test · 1 hata · 9 yeni senaryo · saat karşılığı 11 · kalite 4,8 - ➤Ekibe rapor bağlantısı ve sonraki koşum saati gönderildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici CI’dan sürüm adayı etiketi Sonuç Regresyon sonucu, açılan hatalar ve kapsama raporu sürüm kaydında; iş kaydı güncellendi ## Ajanlar SÜPERVİZÖR ### Test Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-tulpar-4 · on-prem ### Test Plan Ajanı Değişen alanlara göre test setini seçer ve genişletir get_release_diffselect_testsgenerate_cases aginies-sunkar-1 · on-prem ### Koşum Ajanı Testleri koşar, kararsız testleri ayıklar run_playwrightrun_xcuitestretry_flakycollect_artifacts aginies-talay-2 · on-prem ### Hata Ajanı Hataları kanıt ve adımlarla açar, raporu yazar file_bugwrite_reportlog_work ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Test Süpervizörü | RC 2.14.0-rc3: 23 commit, değişen alanlar ödeme, profil, bildirimler | CI/CD | 200 | — | | 02 | akıl yürütme | Test Plan Ajanı | 412 test seçildi (tam regresyon), 9 yeni senaryo kabul kriterlerinden üretildi | — | 900 | 1.300 | | 03 | araç | Koşum Ajanı | 3 tarayıcı + 4 cihazda koşuldu: 409 geçti, 3 başarısız; 2’si tekrarda geçti (kararsız) | Device / Browser Grid | 1.800 | — | | 04 | doğrulama | Koşum Ajanı | Kalan 1 başarısızlık tutarlı: ödeme özeti yuvarlama farkı (0,01) | — | 520 | 380 | | 05 | yazma | Hata Ajanı | BUG-2291 açıldı: adımlar, ekran kaydı, log; öncelik yüksek, ödeme ekibine atandı | Jira / Xray | 520 | — | | 06 | akıl yürütme | Hata Ajanı | Rapor: kapsama %91, 2 kararsız test karantinaya, kaçan hata 0 (önceki sürüm 1) | — | 700 | 640 | | 07 | insan onayı | Test Süpervizörü | Sürüm yöneticisi: BUG-2291 düzeltilene kadar “no-go”; rc4 için yeniden koşum planlandı | — | 2.600 | — | | 08 | yazma | Hata Ajanı | İş kaydı: 412 test · 1 hata · 9 yeni senaryo · saat karşılığı 11 · kalite 4,8 | Workforce Work Log | 300 | — | | 09 | bildirim | Hata Ajanı | Ekibe rapor bağlantısı ve sonraki koşum saati gönderildi | Jira / Xray | 240 | — | Simülasyon metrikleri (tahmini) 260çalıştırma / ay %95,8başarı 8,6 sp50 $0,0165tahmini maliyet / çalıştırma Onay kuralı Sürüm “go / no-go” kararı sürüm yöneticisinde Adım profili giriş · 1akıl yürütme · 2araç · 1doğrulama · 1insan onayı · 1yazma · 2bildirim · 1 SİSTEMLER - CI/CDapi - Device / Browser Gridapi - Jira / Xrayticketing - Workforce Work Logdb Çalıştırmak için gerekenler - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Kiralık test ajanı modeli: dış kaynak testçi yerine her sürümde koşan, kanıtla hata açan ajan Bu senaryonun yer aldığı paketler Ajan kiralama (Agent Outsourcing) → Sektör: Yazılım Geliştirme → İlgili ajanlar Yazılım GeliştirmeHITL ### Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR Jira'daki bir epic'ten başlar; eski modül okunur, davranışı testlerle sabitlenir, yeniden yazılır ve kanıtlı bir PR açılır. 1.900 çalıştırma / ay %90,4 başarı 5 ajan Yazılım GeliştirmeHITL ### Özellik teslimat hattı Bir Jira epic'i ajanlarla kriter, tasarım, kod ve teste dönüşür; insan yalnızca PR'ı inceler ve onaylar. 1.200 çalıştırma / ay %91,2 başarı 5 ajan Yazılım GeliştirmeHITL ### CI/CD sürüm hattı Derleme, test, tarama ve staging otomatik; üretime yalnızca sürüm yöneticisi onayıyla ArgoCD ile çıkılır. 950 çalıştırma / ay %94,7 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # İade ve para iadesi URL: https://www.aginies.com/tr/ajanlar/returns-and-refunds Language: tr Summary: Politika ve müşteri geçmişi saniyede kontrol edilir, etiket üretilir; iade tutarı kargo taramasıyla ödenir. Uygulama, web veya e-postadan gelen iade talebi → İade kaydı ve etiket hazır, para iadesi kargo teslimatıyla tetiklenmiş, müşteri bilgilendirildi Perakende & E-ticaretappwebwhatsappemail # İade ve para iadesi Politika ve müşteri geçmişi saniyede kontrol edilir, etiket üretilir; iade tutarı kargo taramasıyla ödenir. Kısaca İade ve para iadesi, Perakende & E-ticaret sektöründe app, web, whatsapp, email kanalında çalışan bir ajan senaryosudur. Tetikleyici: Uygulama, web veya e-postadan gelen iade talebi. İade Süpervizörü 3 uzman ajanı (Kabul Ajanı, Politika Ajanı, İşlem Ajanı) 4 sistem üzerinde yönetir (Shopify / OMS, Carrier API (Yurtiçi, DHL), Payment Gateway (iyzico), Email (Exchange)); 10 adımın 3 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: İade kaydı ve etiket hazır, para iadesi kargo teslimatıyla tetiklenmiş, müşteri bilgilendirildi. Onay kuralı: İade tutarı > 2.500 TL veya 12 ayda > 5 iade → ekip lideri onayı. Perakende & E-ticaret ### İade ve para iadesi [İade ve para iadesi] çalıştırma #41.288.135süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: İade tutarı > 2.500 TL veya 12 ayda > 5 iade → ekip lideri onayı Bu çalıştırmadaki ajanlar İade Süpervizörüsupervisorplanlar, yönlendirir, onaylar Kabul Ajanıaginies-sunkar-1 · on-premSiparişi bulur, iade nedenini sınıflar Politika Ajanıaginies-tulpar-4 · on-premUygunluğu ve kötüye kullanım sinyallerini değerlendirir İşlem Ajanıaginies-talay-2 · on-premEtiket üretir, iade kaydı açar, para iadesini yapar Denetim izi00:00.00 - ◦Uygulamadan iade talebi: sipariş #100377, 2 kalem, neden "beden uymadı" - ⚙Sipariş 26.08 teslim edildi, 2 kalem toplam 3.180 TL, kartla ödendi - ✓Politika: 14 gün içinde, etiketli, iade edilebilir kategori → uygun - ∿Geçmiş: 12 ayda 11 siparişte 3 iade (%27), kötüye kullanım sinyali yok - ⌘Plan: iade etiketi → onay eşiği → iade kaydı → para iadesi → bildirim - ⇢Ücretsiz iade etiketi oluşturuldu: YK-8812 0034, 3 gün geçerli - ⏸İade > 2.500 TL → ekip lideri 2 dakikada onayladı - ⇢İade kaydı açıldı, stok "beklenen iade" olarak güncellendi - ⇢Kargo teslim alma olayıyla 3.180 TL karta iade edildi - ➤E-posta: iade onayı, etiket ve 3–5 iş günü iade süresi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Uygulama, web veya e-postadan gelen iade talebi Sonuç İade kaydı ve etiket hazır, para iadesi kargo teslimatıyla tetiklenmiş, müşteri bilgilendirildi ## Ajanlar SÜPERVİZÖR ### İade Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Kabul Ajanı Siparişi bulur, iade nedenini sınıflar get_orderextract_return_reasonclassify_reason aginies-tulpar-4 · on-prem ### Politika Ajanı Uygunluğu ve kötüye kullanım sinyallerini değerlendirir get_return_policycheck_eligibilityget_customer_return_history aginies-talay-2 · on-prem ### İşlem Ajanı Etiket üretir, iade kaydı açar, para iadesini yapar create_return_labelcreate_returnissue_refundsend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | İade Süpervizörü | Uygulamadan iade talebi: sipariş #100377, 2 kalem, neden "beden uymadı" | — | 200 | — | | 02 | araç | Kabul Ajanı | Sipariş 26.08 teslim edildi, 2 kalem toplam 3.180 TL, kartla ödendi | Shopify / OMS | 360 | — | | 03 | doğrulama | Politika Ajanı | Politika: 14 gün içinde, etiketli, iade edilebilir kategori → uygun | — | 620 | 540 | | 04 | akıl yürütme | Politika Ajanı | Geçmiş: 12 ayda 11 siparişte 3 iade (%27), kötüye kullanım sinyali yok | — | 480 | 390 | | 05 | plan | İade Süpervizörü | Plan: iade etiketi → onay eşiği → iade kaydı → para iadesi → bildirim | — | 240 | 180 | | 06 | yazma | İşlem Ajanı | Ücretsiz iade etiketi oluşturuldu: YK-8812 0034, 3 gün geçerli | Carrier API (Yurtiçi, DHL) | 420 | — | | 07 | insan onayı | İade Süpervizörü | İade > 2.500 TL → ekip lideri 2 dakikada onayladı | — | 2.400 | — | | 08 | yazma | İşlem Ajanı | İade kaydı açıldı, stok "beklenen iade" olarak güncellendi | Shopify / OMS | 380 | — | | 09 | yazma | İşlem Ajanı | Kargo teslim alma olayıyla 3.180 TL karta iade edildi | Payment Gateway (iyzico) | 540 | — | | 10 | bildirim | İşlem Ajanı | E-posta: iade onayı, etiket ve 3–5 iş günü iade süresi | Email (Exchange) | 240 | — | Simülasyon metrikleri (tahmini) 48.000çalıştırma / ay %96,7başarı 5,9 sp50 $0,0012tahmini maliyet / çalıştırma Onay kuralı İade tutarı > 2.500 TL veya 12 ayda > 5 iade → ekip lideri onayı Adım profili giriş · 1plan · 1akıl yürütme · 1araç · 1doğrulama · 1insan onayı · 1yazma · 3bildirim · 1 SİSTEMLER - Shopify / OMSerp - Carrier API (Yurtiçi, DHL)api - Payment Gateway (iyzico)payments - Email (Exchange)email Çalıştırmak için gerekenler - ERP/MES için kayıt ve sorgu API erişimi - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Ödeme servisleri için yetkili API erişimi ve limitler - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Moda e-ticaretinin self-servis iade deseni; para iadesi kargo tarama olayına bağlanır Bu senaryonun yer aldığı paketler E-ticaret müşteri desteği ve aksiyon → Sektör: Perakende & E-ticaret → İlgili ajanlar Perakende & E-ticaretHITL ### Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin Müşteri bir fotoğraf ve bir cümle gönderir; ajan stokta olan, bedenine uyan ve bütçeye sığan üç kombin kurar, seçileni sepete ekler. 142.000 çalıştırma / ay %96,1 başarı 5 ajan Perakende & E-ticaret ### Siparişim nerede? (WISMO) Müşteri sorar, sipariş ve kargo olayları anında birleşir; gecikmede yeni tarih verilir, kargo araştırılır. 110.000 çalıştırma / ay %98,1 başarı 4 ajan Perakende & E-ticaret ### Ürün soru-cevap asistanı Beden, uyumluluk ve özellik soruları katalog ve iade verisiyle yanıtlanır; stok ve fiyat doğrulanır. 120.000 çalıştırma / ay %98,7 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Yorum yanıtlama ve kalite eskalasyonu URL: https://www.aginies.com/tr/ajanlar/review-response Language: tr Summary: Yorumlar sınıflanır, marka tonunda yanıtlanır; tekrarlayan kusurlar iade verisiyle QA'e Jira kaydı olur. Saatlik yorum çekimi (site, pazar yeri, uygulama mağazası) → Yanıt yayında, müşteride değişim linki, kusur kümesi QA'de kanıtlarıyla açık Perakende & E-ticaretwebapi # Yorum yanıtlama ve kalite eskalasyonu Yorumlar sınıflanır, marka tonunda yanıtlanır; tekrarlayan kusurlar iade verisiyle QA'e Jira kaydı olur. Kısaca Yorum yanıtlama ve kalite eskalasyonu, Perakende & E-ticaret sektöründe web, api kanalında çalışan bir ajan senaryosudur. Tetikleyici: Saatlik yorum çekimi (site, pazar yeri, uygulama mağazası). Müşteri Sesi Süpervizörü 3 uzman ajanı (Sınıflandırma Ajanı, Yanıt Ajanı, Kalite Ajanı) 3 sistem üzerinde yönetir (Review Platform API, Shopify / OMS, Jira (QA)); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Yanıt yayında, müşteride değişim linki, kusur kümesi QA'de kanıtlarıyla açık. Onay kuralı: 1 yıldızlı yorum yanıtları ve kusur eskalasyonları → topluluk yöneticisi onayı. Perakende & E-ticaret ### Yorum yanıtlama ve kalite eskalasyonu [Yorum yanıtlama ve kalite eskalasyonu] çalıştırma #41.288.920süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: 1 yıldızlı yorum yanıtları ve kusur eskalasyonları → topluluk yöneticisi onayı Bu çalıştırmadaki ajanlar Müşteri Sesi Süpervizörüsupervisorplanlar, yönlendirir, onaylar Sınıflandırma Ajanıaginies-sunkar-1 · on-premDuygu, konu ve kusur sinyalini belirler Yanıt Ajanıaginies-tulpar-4 · on-premSipariş bağlamıyla marka tonunda yanıt yazar Kalite Ajanıaginies-talay-2 · on-premKusurları kümeler, iade oranıyla karşılaştırır, QA kaydı açar Denetim izi00:00.00 - ◦Saatlik çekim: 184 yeni yorum, 23'ü ≤ 2 yıldız - ∿Yorum #R-77120: 1 yıldız, "fermuar 2 haftada bozuldu" → kusur sinyali, SKU 55-2210 - ⌘Plan: sipariş bağlamı → yanıt taslağı → kusur kümesi → onay → QA eskalasyonu - ⚙Sipariş #99841 18.08 teslim, garanti kapsamında, değişim hakkı var - ∿Yanıt taslağı: özür + ücretsiz değişim teklifi, marka tonu, 68 kelime - ⚙SKU 55-2210: 30 günde 17 "fermuar" yorumu, iade oranı %9,4 (ortalama %3,1) - ✓Kural: ≥ 10 benzer kusur veya iade oranı > 2× ortalama → QA eskalasyonu - ⏸1 yıldızlı yorum yanıtı → topluluk yöneticisi kuyruktan onayladı - ⇢Jira QA-1184 açıldı: tedarikçi partisi, 17 yorum ve iade verisi eklendi - ➤Yanıt platformda yayınlandı, değişim linki yanıta eklendi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Saatlik yorum çekimi (site, pazar yeri, uygulama mağazası) Sonuç Yanıt yayında, müşteride değişim linki, kusur kümesi QA'de kanıtlarıyla açık ## Ajanlar SÜPERVİZÖR ### Müşteri Sesi Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Sınıflandırma Ajanı Duygu, konu ve kusur sinyalini belirler classify_sentimentdetect_topicsdetect_defect_signal aginies-tulpar-4 · on-prem ### Yanıt Ajanı Sipariş bağlamıyla marka tonunda yanıt yazar get_order_contextdraft_responsecheck_tonepublish_reply aginies-talay-2 · on-prem ### Kalite Ajanı Kusurları kümeler, iade oranıyla karşılaştırır, QA kaydı açar cluster_defectsget_sku_return_ratecreate_qa_ticket ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Müşteri Sesi Süpervizörü | Saatlik çekim: 184 yeni yorum, 23'ü ≤ 2 yıldız | — | 260 | — | | 02 | akıl yürütme | Sınıflandırma Ajanı | Yorum #R-77120: 1 yıldız, "fermuar 2 haftada bozuldu" → kusur sinyali, SKU 55-2210 | — | 520 | 380 | | 03 | plan | Müşteri Sesi Süpervizörü | Plan: sipariş bağlamı → yanıt taslağı → kusur kümesi → onay → QA eskalasyonu | — | 240 | 170 | | 04 | araç | Yanıt Ajanı | Sipariş #99841 18.08 teslim, garanti kapsamında, değişim hakkı var | Shopify / OMS | 340 | — | | 05 | akıl yürütme | Yanıt Ajanı | Yanıt taslağı: özür + ücretsiz değişim teklifi, marka tonu, 68 kelime | — | 680 | 620 | | 06 | araç | Kalite Ajanı | SKU 55-2210: 30 günde 17 "fermuar" yorumu, iade oranı %9,4 (ortalama %3,1) | Shopify / OMS | 380 | — | | 07 | doğrulama | Kalite Ajanı | Kural: ≥ 10 benzer kusur veya iade oranı > 2× ortalama → QA eskalasyonu | — | 420 | 310 | | 08 | insan onayı | Müşteri Sesi Süpervizörü | 1 yıldızlı yorum yanıtı → topluluk yöneticisi kuyruktan onayladı | — | 2.200 | — | | 09 | yazma | Kalite Ajanı | Jira QA-1184 açıldı: tedarikçi partisi, 17 yorum ve iade verisi eklendi | Jira (QA) | 460 | — | | 10 | bildirim | Yanıt Ajanı | Yanıt platformda yayınlandı, değişim linki yanıta eklendi | Review Platform API | 320 | — | Simülasyon metrikleri (tahmini) 26.000çalıştırma / ay %97,2başarı 5,8 sp50 $0,0009tahmini maliyet / çalıştırma Onay kuralı 1 yıldızlı yorum yanıtları ve kusur eskalasyonları → topluluk yöneticisi onayı Adım profili giriş · 1plan · 1akıl yürütme · 2araç · 2doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - Review Platform APIapi - Shopify / OMSerp - Jira (QA)ticketing Çalıştırmak için gerekenler - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - ERP/MES için kayıt ve sorgu API erişimi - Bilet veya iş emri sistemi için API erişimi ve atama kuralları - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Küresel pazar yerlerinin yorumdan ürün kalitesine geri besleme döngüsü deseni Bu senaryonun yer aldığı paketler Çağrı merkezi kalite yönetimi →Satış ve lead yönetimi → Sektör: Perakende & E-ticaret → İlgili ajanlar Perakende & E-ticaretHITL ### Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin Müşteri bir fotoğraf ve bir cümle gönderir; ajan stokta olan, bedenine uyan ve bütçeye sığan üç kombin kurar, seçileni sepete ekler. 142.000 çalıştırma / ay %96,1 başarı 5 ajan Perakende & E-ticaret ### Siparişim nerede? (WISMO) Müşteri sorar, sipariş ve kargo olayları anında birleşir; gecikmede yeni tarih verilir, kargo araştırılır. 110.000 çalıştırma / ay %98,1 başarı 4 ajan Perakende & E-ticaretHITL ### İade ve para iadesi Politika ve müşteri geçmişi saniyede kontrol edilir, etiket üretilir; iade tutarı kargo taramasıyla ödenir. 48.000 çalıştırma / ay %96,7 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Çatı GES bağlantı başvurusu URL: https://www.aginies.com/tr/ajanlar/rooftop-solar-connection-application Language: tr Summary: Başvuru belgeleri okunur, trafo kapasitesi ve mevzuat kontrol edilir, eksikler aynı gün bildirilir; bağlantı görüşü mühendisin. Dağıtım şirketi portalından çatı GES bağlantı başvurusu → Teknik ön değerlendirme ve eksik listesi hazır; bağlantı görüşü mühendis onayıyla yayınlandı Enerjiwebemailapi # Çatı GES bağlantı başvurusu Başvuru belgeleri okunur, trafo kapasitesi ve mevzuat kontrol edilir, eksikler aynı gün bildirilir; bağlantı görüşü mühendisin. Kısaca Çatı GES bağlantı başvurusu, Enerji sektöründe web, email, api kanalında çalışan bir ajan senaryosudur. Tetikleyici: Dağıtım şirketi portalından çatı GES bağlantı başvurusu. Bağlantı Süpervizörü 4 uzman ajanı (Belge Ajanı, Şebeke Ajanı, Mevzuat Ajanı, Dosya Ajanı) 4 sistem üzerinde yönetir (Connection Portal, GIS / Network Model, Regulation Vector Store (EPDK), E-mail); 9 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Teknik ön değerlendirme ve eksik listesi hazır; bağlantı görüşü mühendis onayıyla yayınlandı. Onay kuralı: Bağlantı görüşü her zaman şebeke planlama mühendisinin imzasıyla. Enerji ### Çatı GES bağlantı başvurusu [Çatı GES bağlantı başvurusu] çalıştırma #41.290.299süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Bağlantı görüşü her zaman şebeke planlama mühendisinin imzasıyla Bu çalıştırmadaki ajanlar Bağlantı Süpervizörüsupervisorplanlar, yönlendirir, onaylar Belge Ajanıaginies-konrul-1 · on-premTek hat şeması, tapu ve ruhsatı okur Şebeke Ajanıaginies-tulpar-4 · on-premTrafo ve fider kapasitesini değerlendirir Mevzuat Ajanıaginies-talay-2 · on-premLisanssız üretim mevzuatıyla karşılaştırır Dosya Ajanıaginies-sunkar-1 · on-premBaşvuru kaydını günceller ve başvuranı bilgilendirir Denetim izi00:00.00 - ◦Başvuru B-2026-31877: 48 kWp çatı GES, ticari abone, 6 belge yüklendi - ∿Belgeler okundu: tek hat şeması uygun, tapu uyumlu, yapı ruhsatı eksik sayfa - ⚙Trafo TR-2291: 400 kVA, tepe yük %71, bağlı üretim 96 kWp - ∿Barındırma kapasitesi yeterli: 48 kWp ile toplam üretim trafo gücünün %36’sı - ⌕Mevzuat: tüketim tesisiyle aynı ölçüm noktası, sözleşme gücü sınırı içinde → uygun - ✓Görüş taslağı: olumlu, koruma ayarı ve sayaç değişimi şartlı; 1 eksik belge - ⏸Bağlantı görüşü → şebeke planlama mühendisi inceledi ve imzaladı - ⇢Başvuru “şartlı olumlu” durumuna alındı, görüş belgesi eklendi - ➤Başvurana görüş ve eksik belge listesi e-postalandı; 15 gün tamamlama süresi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Dağıtım şirketi portalından çatı GES bağlantı başvurusu Sonuç Teknik ön değerlendirme ve eksik listesi hazır; bağlantı görüşü mühendis onayıyla yayınlandı ## Ajanlar SÜPERVİZÖR ### Bağlantı Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### Belge Ajanı Tek hat şeması, tapu ve ruhsatı okur extract_documentclassify_documentcheck_completeness aginies-tulpar-4 · on-prem ### Şebeke Ajanı Trafo ve fider kapasitesini değerlendirir get_transformer_loadget_connected_generationassess_hosting_capacity aginies-talay-2 · on-prem ### Mevzuat Ajanı Lisanssız üretim mevzuatıyla karşılaştırır search_regulationcheck_eligibilitydraft_opinion aginies-sunkar-1 · on-prem ### Dosya Ajanı Başvuru kaydını günceller ve başvuranı bilgilendirir update_applicationsend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Bağlantı Süpervizörü | Başvuru B-2026-31877: 48 kWp çatı GES, ticari abone, 6 belge yüklendi | Connection Portal | 200 | — | | 02 | akıl yürütme | Belge Ajanı | Belgeler okundu: tek hat şeması uygun, tapu uyumlu, yapı ruhsatı eksik sayfa | — | 980 | 1.300 | | 03 | araç | Şebeke Ajanı | Trafo TR-2291: 400 kVA, tepe yük %71, bağlı üretim 96 kWp | GIS / Network Model | 460 | — | | 04 | akıl yürütme | Şebeke Ajanı | Barındırma kapasitesi yeterli: 48 kWp ile toplam üretim trafo gücünün %36’sı | — | 700 | 760 | | 05 | bilgi | Mevzuat Ajanı | Mevzuat: tüketim tesisiyle aynı ölçüm noktası, sözleşme gücü sınırı içinde → uygun | Regulation Vector Store (EPDK) | 480 | 520 | | 06 | doğrulama | Mevzuat Ajanı | Görüş taslağı: olumlu, koruma ayarı ve sayaç değişimi şartlı; 1 eksik belge | — | 620 | 640 | | 07 | insan onayı | Bağlantı Süpervizörü | Bağlantı görüşü → şebeke planlama mühendisi inceledi ve imzaladı | — | 3.000 | — | | 08 | yazma | Dosya Ajanı | Başvuru “şartlı olumlu” durumuna alındı, görüş belgesi eklendi | Connection Portal | 440 | — | | 09 | bildirim | Dosya Ajanı | Başvurana görüş ve eksik belge listesi e-postalandı; 15 gün tamamlama süresi | E-mail | 240 | — | Simülasyon metrikleri (tahmini) 5.400çalıştırma / ay %95,3başarı 7,1 sp50 $0,0064tahmini maliyet / çalıştırma Onay kuralı Bağlantı görüşü her zaman şebeke planlama mühendisinin imzasıyla Adım profili giriş · 1bilgi · 1akıl yürütme · 2araç · 1doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - Connection Portalapi - GIS / Network Modeldb - Regulation Vector Store (EPDK)db - E-mailemail Çalıştırmak için gerekenler - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Avrupa dağıtım şirketlerinin dağıtık üretim bağlantı başvurusu otomasyonunu yansıtır Bu senaryonun yer aldığı paketler Enerji müşteri operasyonları → Sektör: Enerji → İlgili ajanlar EnerjiHITL ### Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda Abone arar ya da yazar; yüksek fatura açıklanır, kesinti süresi verilir, sayaç verisi gösterilir, taksit planı kurulur. 520.000 çalıştırma / ay %94,6 başarı 5 ajan EnerjiHITL ### Kesinti bildirimi yönetimi Bildirimler fider bazında kümelenir, tek OSS kaydı açılır, etkilenen herkes ekip yola çıkmadan bilgilenir. 38.000 çalıştırma / ay %97,7 başarı 5 ajan EnerjiHITL ### Sayaç okuma itirazı Müşteri sayaç fotoğrafı yollar; endeks görüntüden okunur, tahmini fatura düzeltilir, büyük iade lidere gider. 9.200 çalıştırma / ay %94,9 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Satış fırsatı nitelendirme URL: https://www.aginies.com/tr/ajanlar/sales-lead-qualification Language: tr Summary: Gelen talep saniyeler içinde zenginleştirilir, skorlanır ve CRM'e işlenir; SDR aramaya bağlam notuyla başlar. Web sitesinde "demo talep et" formunun gönderilmesi → Fırsat SQL aşamasında, SDR atandı, 4 saat içinde arama görevi ve bağlam notu hazır Kurumsal Fonksiyonlarwebapiemail # Satış fırsatı nitelendirme Gelen talep saniyeler içinde zenginleştirilir, skorlanır ve CRM'e işlenir; SDR aramaya bağlam notuyla başlar. Kısaca Satış fırsatı nitelendirme, Kurumsal Fonksiyonlar sektöründe web, api, email kanalında çalışan bir ajan senaryosudur. Tetikleyici: Web sitesinde "demo talep et" formunun gönderilmesi. Satış Süpervizörü 3 uzman ajanı (Zenginleştirme Ajanı, Skorlama Ajanı, CRM Ajanı) 3 sistem üzerinde yönetir (HubSpot, Enrichment API, Microsoft Teams); 9 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Fırsat SQL aşamasında, SDR atandı, 4 saat içinde arama görevi ve bağlam notu hazır. Onay kuralı: Kurumsal (> 1.000 çalışan) veya rakip şüphesi → SDR lideri kabulü. Kurumsal Fonksiyonlar ### Satış fırsatı nitelendirme [Satış fırsatı nitelendirme] çalıştırma #41.289.141süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Kurumsal (> 1.000 çalışan) veya rakip şüphesi → SDR lideri kabulü Bu çalıştırmadaki ajanlar Satış Süpervizörüsupervisorplanlar, yönlendirir, onaylar Zenginleştirme Ajanıaginies-tulpar-4 · on-premŞirket ve kişi verisini zenginleştirir Skorlama Ajanıaginies-tulpar-4 · on-premICP uyumu ve niyet sinyalleriyle skorlar CRM Ajanıaginies-talay-2 · on-premCRM kaydını günceller, SDR'a devreder Denetim izi00:00.00 - ◦Web formu: "demo talep et", ayse.demir@…, şirket alanı "Norm Holding" - ⚙Şirket: 2.400 çalışan, üretim, İzmir; unvan: Dijital Dönüşüm Lideri - ⚙CRM geçmişi: 2024'te 1 kaybedilen fırsat (bütçe), 3 webinar katılımı - ∿ICP uyumu 88/100 · niyet yüksek (fiyat sayfası ×3) · SQL eşiği 75 → nitelikli - ✓Koruma: kurumsal alan adı, rakip değil, KVKK/GDPR onayı mevcut - ⇢Fırsat güncellendi, aşama "SQL", sahip: SDR Marmara ekibi (sırayla atama) - ⏸Kurumsal (> 1.000 çalışan) → SDR lideri aramadan önce kabul etti - ⇢Görev: 4 saat içinde arama; kayda 3 satırlık bağlam notu eklendi - ➤SDR'a Teams'ten not iletildi; kişiselleştirilmiş takip e-postası taslak olarak hazır - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Web sitesinde "demo talep et" formunun gönderilmesi Sonuç Fırsat SQL aşamasında, SDR atandı, 4 saat içinde arama görevi ve bağlam notu hazır ## Ajanlar SÜPERVİZÖR ### Satış Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-tulpar-4 · on-prem ### Zenginleştirme Ajanı Şirket ve kişi verisini zenginleştirir enrich_companyfind_contact_profilecheck_domain aginies-tulpar-4 · on-prem ### Skorlama Ajanı ICP uyumu ve niyet sinyalleriyle skorlar search_crm_historymatch_icpscore_lead aginies-talay-2 · on-prem ### CRM Ajanı CRM kaydını günceller, SDR'a devreder upsert_leadassign_ownercreate_tasksend_teams_message ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Satış Süpervizörü | Web formu: "demo talep et", ayse.demir@…, şirket alanı "Norm Holding" | — | 180 | — | | 02 | araç | Zenginleştirme Ajanı | Şirket: 2.400 çalışan, üretim, İzmir; unvan: Dijital Dönüşüm Lideri | Enrichment API | 520 | — | | 03 | araç | Skorlama Ajanı | CRM geçmişi: 2024'te 1 kaybedilen fırsat (bütçe), 3 webinar katılımı | HubSpot | 380 | — | | 04 | akıl yürütme | Skorlama Ajanı | ICP uyumu 88/100 · niyet yüksek (fiyat sayfası ×3) · SQL eşiği 75 → nitelikli | — | 760 | 820 | | 05 | doğrulama | Skorlama Ajanı | Koruma: kurumsal alan adı, rakip değil, KVKK/GDPR onayı mevcut | — | 300 | 180 | | 06 | yazma | CRM Ajanı | Fırsat güncellendi, aşama "SQL", sahip: SDR Marmara ekibi (sırayla atama) | HubSpot | 480 | — | | 07 | insan onayı | Satış Süpervizörü | Kurumsal (> 1.000 çalışan) → SDR lideri aramadan önce kabul etti | — | 2.200 | — | | 08 | yazma | CRM Ajanı | Görev: 4 saat içinde arama; kayda 3 satırlık bağlam notu eklendi | HubSpot | 360 | — | | 09 | bildirim | CRM Ajanı | SDR'a Teams'ten not iletildi; kişiselleştirilmiş takip e-postası taslak olarak hazır | Microsoft Teams | 240 | — | Simülasyon metrikleri (tahmini) 15.600çalıştırma / ay %97,9başarı 5,4 sp50 $0,0021tahmini maliyet / çalıştırma Onay kuralı Kurumsal (> 1.000 çalışan) veya rakip şüphesi → SDR lideri kabulü Adım profili giriş · 1akıl yürütme · 1araç · 2doğrulama · 1insan onayı · 1yazma · 2bildirim · 1 SİSTEMLER - HubSpotcrm - Enrichment APIapi - Microsoft Teamschannel Çalıştırmak için gerekenler - CRM için API erişimi ve müşteri eşleme anahtarı - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek SaaS gelir ekiplerinin gelen talebe hızlı yanıt (speed-to-lead) otomasyonunu yansıtır Bu senaryonun yer aldığı paketler Satış ve lead yönetimi → İlgili ajanlar Kurumsal FonksiyonlarHITL ### Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada Talep e-postayla gelir; teklifler karşılaştırılır, sözleşme maddeleri kontrol edilir, sipariş açılır, fatura üç yönlü eşleşir ve ödeme planlanır. 14.000 çalıştırma / ay %92,7 başarı 5 ajan Kurumsal Fonksiyonlar ### E-posta sınıflandırma ve ek kontrolü Ortak kutuya gelen her e-posta sınıflandırılır, ekler listeyle karşılaştırılır, eksikler anında sorulur. 86.000 çalıştırma / ay %97,6 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Çağrı merkezi kalite izleme Örneklem değil, tüm çağrılar rubriğe göre puanlanır; uyum ihlali ve koçluk fırsatı sabah liderin önünde. 90.000 çalıştırma / ay %98,8 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Sevkiyat istisna yönetimi URL: https://www.aginies.com/tr/ajanlar/shipment-exception-handling Language: tr Summary: Gecikme ya da hasar geldiğinde etki hesaplanır, taşıyıcı yeniden planlanır, müşteri daha sormadan bilgilenir. Taşıyıcı webhook'undan gecikme veya hasar istisnası → Sevkiyat yeniden planlandı, istisna neden koduyla kapatıldı, müşteri yeni ETA ile bilgilendirildi Lojistikapiemailwhatsapp # Sevkiyat istisna yönetimi Gecikme ya da hasar geldiğinde etki hesaplanır, taşıyıcı yeniden planlanır, müşteri daha sormadan bilgilenir. Kısaca Sevkiyat istisna yönetimi, Lojistik sektöründe api, email, whatsapp kanalında çalışan bir ajan senaryosudur. Tetikleyici: Taşıyıcı webhook'undan gecikme veya hasar istisnası. Sevkiyat Süpervizörü 3 uzman ajanı (İstisna Ajanı, Etki Ajanı, Aksiyon Ajanı) 4 sistem üzerinde yönetir (TMS (Transport Management), Carrier API, Salesforce CRM, WhatsApp Business API); 10 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Sevkiyat yeniden planlandı, istisna neden koduyla kapatıldı, müşteri yeni ETA ile bilgilendirildi. Onay kuralı: Ek maliyet > 1.000 € → sevk şefi onayı. Lojistik ### Sevkiyat istisna yönetimi [Sevkiyat istisna yönetimi] çalıştırma #41.287.262süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Ek maliyet > 1.000 € → sevk şefi onayı Bu çalıştırmadaki ajanlar Sevkiyat Süpervizörüsupervisorplanlar, yönlendirir, onaylar İstisna Ajanıaginies-sunkar-1 · on-premİstisnayı sınıflandırır ve şiddetini belirler Etki Ajanıaginies-tulpar-4 · on-premSLA ve ceza maruziyetini hesaplar Aksiyon Ajanıaginies-talay-2 · on-premTaşıyıcıyı yeniden planlar, müşteriyle iletişim kurar Denetim izi00:00.00 - ◦Taşıyıcı webhook: SH-4471820 "gecikme – aktarma merkezi yoğunluğu", ETA +36 sa - ⚙Sevkiyat: 14 palet, soğuk zincir, müşteri C-2081, teslim Çar 08:00 - ∿İstisna sınıfı: gecikme · şiddet yüksek (soğuk zincir, sözleşmeli SLA) - ⚙Müşteri SLA'sı: %98 zamanında, gecikme cezası palet-gün başına 250 € - ∿Tahmini etki: 3.500 € ceza + reefer riski; alternatif taşıyıcı +1.200 € - ✓Politika: ek maliyet ceza maruziyetinin < %40'ı ise yeniden planlama serbest → %34 - ⏸Ek maliyet > 1.000 € → sevk şefi onayı (6 dakikada onaylandı) - ⇢B taşıyıcısıyla alım yeniden planlandı, yeni ETA Çar 06:30, referans RB-88231 - ⇢Sevkiyat yeniden planlandı, istisna "aktarma yoğunluğu" neden koduyla kapatıldı - ➤Müşteri lojistik yetkilisine yeni ETA iletildi, müşteri temsilcisi e-postada bilgi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Taşıyıcı webhook'undan gecikme veya hasar istisnası Sonuç Sevkiyat yeniden planlandı, istisna neden koduyla kapatıldı, müşteri yeni ETA ile bilgilendirildi ## Ajanlar SÜPERVİZÖR ### Sevkiyat Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### İstisna Ajanı İstisnayı sınıflandırır ve şiddetini belirler classify_exceptionget_shipment_status aginies-tulpar-4 · on-prem ### Etki Ajanı SLA ve ceza maruziyetini hesaplar get_customer_slaestimate_delay_costsearch_contract_terms aginies-talay-2 · on-prem ### Aksiyon Ajanı Taşıyıcıyı yeniden planlar, müşteriyle iletişim kurar rebook_carrierupdate_shipmentcreate_claimsend_whatsappsend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Sevkiyat Süpervizörü | Taşıyıcı webhook: SH-4471820 "gecikme – aktarma merkezi yoğunluğu", ETA +36 sa | — | 180 | — | | 02 | araç | İstisna Ajanı | Sevkiyat: 14 palet, soğuk zincir, müşteri C-2081, teslim Çar 08:00 | TMS (Transport Management) | 400 | — | | 03 | akıl yürütme | İstisna Ajanı | İstisna sınıfı: gecikme · şiddet yüksek (soğuk zincir, sözleşmeli SLA) | — | 540 | 460 | | 04 | araç | Etki Ajanı | Müşteri SLA'sı: %98 zamanında, gecikme cezası palet-gün başına 250 € | Salesforce CRM | 380 | — | | 05 | akıl yürütme | Etki Ajanı | Tahmini etki: 3.500 € ceza + reefer riski; alternatif taşıyıcı +1.200 € | — | 720 | 880 | | 06 | doğrulama | Etki Ajanı | Politika: ek maliyet ceza maruziyetinin 1.000 € → sevk şefi onayı (6 dakikada onaylandı) | — | 2.600 | — | | 08 | yazma | Aksiyon Ajanı | B taşıyıcısıyla alım yeniden planlandı, yeni ETA Çar 06:30, referans RB-88231 | Carrier API | 620 | — | | 09 | yazma | Aksiyon Ajanı | Sevkiyat yeniden planlandı, istisna "aktarma yoğunluğu" neden koduyla kapatıldı | TMS (Transport Management) | 420 | — | | 10 | bildirim | Aksiyon Ajanı | Müşteri lojistik yetkilisine yeni ETA iletildi, müşteri temsilcisi e-postada bilgi | WhatsApp Business API | 260 | — | Simülasyon metrikleri (tahmini) 12.800çalıştırma / ay %95,5başarı 6,5 sp50 $0,0037tahmini maliyet / çalıştırma Onay kuralı Ek maliyet > 1.000 € → sevk şefi onayı Adım profili giriş · 1akıl yürütme · 2araç · 2doğrulama · 1insan onayı · 1yazma · 2bildirim · 1 SİSTEMLER - TMS (Transport Management)core - Carrier APIapi - Salesforce CRMcrm - WhatsApp Business APIchannel Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - CRM için API erişimi ve müşteri eşleme anahtarı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Küresel 3PL'lerin ve kargo ağlarının kontrol kulesi istisna yönetimini yansıtır Bu senaryonun yer aldığı paketler Lojistik kontrol kulesi → Sektör: Lojistik → İlgili ajanlar LojistikHITL ### Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA Telematik sinyali gecikmeyi erken yakalar; alternatif rota ve taşıyıcı seçilir, müşteri bilgilendirilir, SLA tazmini hesaplanır. 96.000 çalıştırma / ay %93,4 başarı 5 ajan LojistikHITL ### Gümrük belgesi hazırlama GTİP kodu, fatura ve çeki listesi dakikalarda hazırlanır; düşük güvenli satırlar gümrük uzmanına gider. 4.300 çalıştırma / ay %94,7 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # SIM değişimi dolandırıcılık kontrolü URL: https://www.aginies.com/tr/ajanlar/sim-swap-fraud-check Language: tr Summary: Her SIM değişimi provizyondan önce risk skorlanır; şüpheli talepler bekletilir ve bankalara sinyal iletilir. Bayi, uygulama veya API üzerinden gelen SIM değişimi talebi → Talep provizyonlandı ya da bekletildi, bankalar sinyal aldı, müşteri uyarıldı Telekombranchappapi # SIM değişimi dolandırıcılık kontrolü Her SIM değişimi provizyondan önce risk skorlanır; şüpheli talepler bekletilir ve bankalara sinyal iletilir. Kısaca SIM değişimi dolandırıcılık kontrolü, Telekom sektöründe branch, app, api kanalında çalışan bir ajan senaryosudur. Tetikleyici: Bayi, uygulama veya API üzerinden gelen SIM değişimi talebi. SIM Güvenlik Süpervizörü 3 uzman ajanı (Risk Ajanı, Doğrulama Ajanı, Provizyon Ajanı) 4 sistem üzerinde yönetir (Salesforce CRM, HLR / Provisioning, Bank SIM-Swap Signal API, SMS Gateway); 10 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Talep provizyonlandı ya da bekletildi, bankalar sinyal aldı, müşteri uyarıldı. Onay kuralı: Risk skoru ≥ 70 veya 72 saat içinde kimlik bilgisi değişikliği → dolandırıcılık masası onayı. Telekom ### SIM değişimi dolandırıcılık kontrolü [SIM değişimi dolandırıcılık kontrolü] çalıştırma #41.287.007süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Risk skoru ≥ 70 veya 72 saat içinde kimlik bilgisi değişikliği → dolandırıcılık masası onayı Bu çalıştırmadaki ajanlar SIM Güvenlik Süpervizörüsupervisorplanlar, yönlendirir, onaylar Risk Ajanıaginies-talay-2 · on-premHesap olayları ve cihaz geçmişiyle risk skorlar Doğrulama Ajanıaginies-sunkar-1 · on-premEski SIM'e OTP gönderir, kimlik belgesini doğrular Provizyon Ajanıaginies-sunkar-1 · on-premSIM'i aktive eder ya da bekletir, bankaları ve müşteriyi bilgilendirir Denetim izi00:00.00 - ◦Bayi talebi: 0532 *** 8810 için SIM değişimi, neden "kayıp" - ⚙Hesap olayları: 2 gün önce şifre sıfırlama, 6 saat önce e-posta değişikliği - ⚙Cihaz geçmişi: aynı IMEI 14 aydır aktif, son 24 saatte 3 giriş denemesi - ∿Risk skoru 78/100: yakın tarihli kimlik değişiklikleri + bayi kanalı → yüksek - ⌘Plan: eski SIM'e OTP → kimlik belgesi → onay → provizyon ya da bekletme - ✓Eski SIM'e OTP yanıtsız; bayide kimlik OCR eşleşmesi %71 → yetersiz - ⏸Risk ≥ 70 → dolandırıcılık masası: 24 saat bekletme, video doğrulama istendi - ⇢SIM değişimi "beklemede" olarak kaydedildi, aktivasyon engellendi - ⇢Bankalara hat bazlı SIM değişimi girişimi sinyali gönderildi - ➤Eski SIM'e ve kayıtlı e-postaya uyarı: "talebi siz yapmadıysanız 444'ü arayın" - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Bayi, uygulama veya API üzerinden gelen SIM değişimi talebi Sonuç Talep provizyonlandı ya da bekletildi, bankalar sinyal aldı, müşteri uyarıldı ## Ajanlar SÜPERVİZÖR ### SIM Güvenlik Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Risk Ajanı Hesap olayları ve cihaz geçmişiyle risk skorlar get_account_eventsget_device_historyscore_sim_swap_risk aginies-sunkar-1 · on-prem ### Doğrulama Ajanı Eski SIM'e OTP gönderir, kimlik belgesini doğrular send_otp_old_simverify_id_documentcheck_recent_password_reset aginies-sunkar-1 · on-prem ### Provizyon Ajanı SIM'i aktive eder ya da bekletir, bankaları ve müşteriyi bilgilendirir activate_simhold_requestnotify_banks_apisend_smssend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | SIM Güvenlik Süpervizörü | Bayi talebi: 0532 *** 8810 için SIM değişimi, neden "kayıp" | — | 200 | — | | 02 | araç | Risk Ajanı | Hesap olayları: 2 gün önce şifre sıfırlama, 6 saat önce e-posta değişikliği | Salesforce CRM | 420 | — | | 03 | araç | Risk Ajanı | Cihaz geçmişi: aynı IMEI 14 aydır aktif, son 24 saatte 3 giriş denemesi | HLR / Provisioning | 380 | — | | 04 | akıl yürütme | Risk Ajanı | Risk skoru 78/100: yakın tarihli kimlik değişiklikleri + bayi kanalı → yüksek | — | 680 | 720 | | 05 | plan | SIM Güvenlik Süpervizörü | Plan: eski SIM'e OTP → kimlik belgesi → onay → provizyon ya da bekletme | — | 240 | 180 | | 06 | doğrulama | Doğrulama Ajanı | Eski SIM'e OTP yanıtsız; bayide kimlik OCR eşleşmesi %71 → yetersiz | — | 1.600 | 420 | | 07 | insan onayı | SIM Güvenlik Süpervizörü | Risk ≥ 70 → dolandırıcılık masası: 24 saat bekletme, video doğrulama istendi | — | 3.000 | — | | 08 | yazma | Provizyon Ajanı | SIM değişimi "beklemede" olarak kaydedildi, aktivasyon engellendi | HLR / Provisioning | 400 | — | | 09 | yazma | Provizyon Ajanı | Bankalara hat bazlı SIM değişimi girişimi sinyali gönderildi | Bank SIM-Swap Signal API | 360 | — | | 10 | bildirim | Provizyon Ajanı | Eski SIM'e ve kayıtlı e-postaya uyarı: "talebi siz yapmadıysanız 444'ü arayın" | SMS Gateway | 240 | — | Simülasyon metrikleri (tahmini) 41.000çalıştırma / ay %97,7başarı 7,5 sp50 $0,001tahmini maliyet / çalıştırma Onay kuralı Risk skoru ≥ 70 veya 72 saat içinde kimlik bilgisi değişikliği → dolandırıcılık masası onayı Adım profili giriş · 1plan · 1akıl yürütme · 1araç · 2doğrulama · 1insan onayı · 1yazma · 2bildirim · 1 SİSTEMLER - Salesforce CRMcrm - HLR / Provisioningcore - Bank SIM-Swap Signal APIapi - SMS Gatewaychannel Çalıştırmak için gerekenler - CRM için API erişimi ve müşteri eşleme anahtarı - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek İngiltere/AB operatörlerinin bankalarla paylaştığı SIM değişimi risk sinyali deseni Bu senaryonun yer aldığı paketler Abone hizmetleri ve şebeke desteği → Sektör: Telekom → İlgili ajanlar TelekomHITL ### Abone hizmet asistanı: fatura, paket, arıza ve taahhüt Abone arar ya da yazar; fatura itirazı, paket değişikliği ve hat arızası aynı oturumda BSS ve şebeke sistemlerinde çözülür. 410.000 çalıştırma / ay %94,3 başarı 5 ajan TelekomHITL ### Fatura itirazı çözümü Fatura kalemleri ve kullanım kayıtları saniyede açıklanır; iyi niyet kredisi politikaya göre uygulanır. 88.000 çalıştırma / ay %96,3 başarı 4 ajan TelekomHITL ### Şebeke kesintisi iletişimi Kesinti kaydı açılınca etkilenen aboneler bulunur, mesaj yazılıp onaylanır; müşteri aramadan bilgilendirilir. 1.400 çalıştırma / ay %98,5 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Sosyal yardım başvurusu ön değerlendirme URL: https://www.aginies.com/tr/ajanlar/social-assistance-eligibility Language: tr Summary: Başvuru kurum kayıtlarıyla doğrulanır, uygunluk kriterleri puanlanır, eksikler vatandaşa bildirilir; karar sosyal inceleme görevlisinin. e-Devlet ya da sosyal hizmet merkezi üzerinden yardım başvurusu → Doğrulanmış dosya ve uygunluk özeti görevliye ulaştı; vatandaş eksik ve süre bilgisi aldı Kamu ve Hükümetwebappvoicebatch # Sosyal yardım başvurusu ön değerlendirme Başvuru kurum kayıtlarıyla doğrulanır, uygunluk kriterleri puanlanır, eksikler vatandaşa bildirilir; karar sosyal inceleme görevlisinin. Kısaca Sosyal yardım başvurusu ön değerlendirme, Kamu ve Hükümet sektöründe web, app, voice, batch kanalında çalışan bir ajan senaryosudur. Tetikleyici: e-Devlet ya da sosyal hizmet merkezi üzerinden yardım başvurusu. Sosyal Yardım Süpervizörü 3 uzman ajanı (Doğrulama Ajanı, Uygunluk Ajanı, Vatandaş İletişim Ajanı) 4 sistem üzerinde yönetir (Social Assistance Information System, Civil Registry / Income Services, Eligibility Criteria (vector), SMS Gateway); 9 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Doğrulanmış dosya ve uygunluk özeti görevliye ulaştı; vatandaş eksik ve süre bilgisi aldı. Onay kuralı: Yardım kararı her zaman sosyal inceleme görevlisinde; ajan ön değerlendirme yapar. Kamu ve Hükümet ### Sosyal yardım başvurusu ön değerlendirme [Sosyal yardım başvurusu ön değerlendirme] çalıştırma #41.286.904süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Yardım kararı her zaman sosyal inceleme görevlisinde; ajan ön değerlendirme yapar Bu çalıştırmadaki ajanlar Sosyal Yardım Süpervizörüsupervisorplanlar, yönlendirir, onaylar Doğrulama Ajanıaginies-sunkar-1 · on-premNüfus, gelir ve sosyal güvenlik kayıtlarını sorgular Uygunluk Ajanıaginies-tulpar-4 · on-premKriterleri puanlar ve gerekçe yazar Vatandaş İletişim Ajanıaginies-talay-2 · on-premBaşvuranı sade dille bilgilendirir Denetim izi00:00.00 - ◦Başvuru SY-2026-77410: 4 kişilik hane, düzenli yardım talebi, kimlik doğrulandı - ⚙Hane bileşimi ve adres doğrulandı; 2 yetişkin, 2 çocuk (7 ve 11) - ⚙Gelir kayıtları: 1 kayıtlı çalışan, aylık brüt 21.400 TL; SGK durumu aktif - ⌕Kriterler: kişi başı gelir eşiği, mülkiyet, hane durumu, önceki yardımlar - ∿Kişi başı gelir eşiğin %8 altında; mülkiyet kaydı yok → ön uygunluk: olumlu, hane ziyareti gerekli - ✓Gerekçe sade dille yazıldı; hassas alanlar görevli ekranında maskelendi - ⇢Dosya “ön uygun” durumuna alındı, hane ziyareti planlandı: 3 iş günü - ⏸Yardım kararı → sosyal inceleme görevlisi ziyaret sonrası dosyayı onayladı - ➤Vatandaşa SMS: karar, ilk ödeme tarihi ve itiraz hakkı; kayıt denetim iziyle kapandı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici e-Devlet ya da sosyal hizmet merkezi üzerinden yardım başvurusu Sonuç Doğrulanmış dosya ve uygunluk özeti görevliye ulaştı; vatandaş eksik ve süre bilgisi aldı ## Ajanlar SÜPERVİZÖR ### Sosyal Yardım Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Doğrulama Ajanı Nüfus, gelir ve sosyal güvenlik kayıtlarını sorgular get_householdget_income_recordsget_social_security_status aginies-tulpar-4 · on-prem ### Uygunluk Ajanı Kriterleri puanlar ve gerekçe yazar search_criteriascore_eligibilityexplain_decision aginies-talay-2 · on-prem ### Vatandaş İletişim Ajanı Başvuranı sade dille bilgilendirir send_smsupdate_caseschedule_home_visit ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Sosyal Yardım Süpervizörü | Başvuru SY-2026-77410: 4 kişilik hane, düzenli yardım talebi, kimlik doğrulandı | Social Assistance Information System | 200 | — | | 02 | araç | Doğrulama Ajanı | Hane bileşimi ve adres doğrulandı; 2 yetişkin, 2 çocuk (7 ve 11) | Civil Registry / Income Services | 480 | — | | 03 | araç | Doğrulama Ajanı | Gelir kayıtları: 1 kayıtlı çalışan, aylık brüt 21.400 TL; SGK durumu aktif | Civil Registry / Income Services | 520 | — | | 04 | bilgi | Uygunluk Ajanı | Kriterler: kişi başı gelir eşiği, mülkiyet, hane durumu, önceki yardımlar | Eligibility Criteria (vector) | 460 | 520 | | 05 | akıl yürütme | Uygunluk Ajanı | Kişi başı gelir eşiğin %8 altında; mülkiyet kaydı yok → ön uygunluk: olumlu, hane ziyareti gerekli | — | 900 | 1.200 | | 06 | doğrulama | Uygunluk Ajanı | Gerekçe sade dille yazıldı; hassas alanlar görevli ekranında maskelendi | — | 420 | 340 | | 07 | yazma | Vatandaş İletişim Ajanı | Dosya “ön uygun” durumuna alındı, hane ziyareti planlandı: 3 iş günü | Social Assistance Information System | 480 | — | | 08 | insan onayı | Sosyal Yardım Süpervizörü | Yardım kararı → sosyal inceleme görevlisi ziyaret sonrası dosyayı onayladı | — | 3.200 | — | | 09 | bildirim | Vatandaş İletişim Ajanı | Vatandaşa SMS: karar, ilk ödeme tarihi ve itiraz hakkı; kayıt denetim iziyle kapandı | SMS Gateway | 260 | — | Simülasyon metrikleri (tahmini) 21.000çalıştırma / ay %96,6başarı 6,9 sp50 $0,0034tahmini maliyet / çalıştırma Onay kuralı Yardım kararı her zaman sosyal inceleme görevlisinde; ajan ön değerlendirme yapar Adım profili giriş · 1bilgi · 1akıl yürütme · 1araç · 2doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - Social Assistance Information Systemcore - Civil Registry / Income Servicesapi - Eligibility Criteria (vector)db - SMS Gatewaychannel Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Avrupa sosyal hizmet kurumlarının kayıt doğrulama ve açıklanabilir ön değerlendirme modellerini yansıtır Bu senaryonun yer aldığı paketler Vatandaş hizmetleri ve dosya ön inceleme → Sektör: Kamu ve Hükümet → İlgili ajanlar Kamu ve HükümetHITL ### Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi Vatandaş yazar ya da arar; uygunluk kontrol edilir, belgeler okunur, başvuru açılır, randevu verilir, her adım izlenebilir. 230.000 çalıştırma / ay %93,8 başarı 5 ajan Kamu ve HükümetHITL ### Vatandaş başvurusu yönlendirme e-Devlet başvurusu saniyeler içinde sınıflandırılır, yasal süresi hesaplanır ve doğru birime SLA ile atanır. 54.000 çalıştırma / ay %98,1 başarı 4 ajan Kamu ve HükümetHITL ### Ruhsat başvurusu ön inceleme Eksik belge ve mevzuat uyumu dakikalar içinde raporlanır; kararı her zaman ruhsat memuru verir. 3.100 çalıştırma / ay %95,8 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Rücu ve tahsilat URL: https://www.aginies.com/tr/ajanlar/subrogation-recovery Language: tr Summary: Kapanmış dosyalar rücu için taranır, kusurlu tarafın sigortacısı bulunur, talep avukat onayıyla gönderilir. Haftalık toplu iş: son 90 günde kapanmış kasko dosyaları → Rücu dosyası açık, talep yazısı karşı sigortacıda, 30 günlük takip kuruldu Sigortabatchemailapi # Rücu ve tahsilat Kapanmış dosyalar rücu için taranır, kusurlu tarafın sigortacısı bulunur, talep avukat onayıyla gönderilir. Kısaca Rücu ve tahsilat, Sigorta sektöründe batch, email, api kanalında çalışan bir ajan senaryosudur. Tetikleyici: Haftalık toplu iş: son 90 günde kapanmış kasko dosyaları. Rücu Süpervizörü 3 uzman ajanı (Tarama Ajanı, Kanıt Ajanı, Tahsilat Ajanı) 4 sistem üzerinde yönetir (Guidewire ClaimCenter, Document Store (S3), Insurance Registry (SBM), Email (Exchange)); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Rücu dosyası açık, talep yazısı karşı sigortacıda, 30 günlük takip kuruldu. Onay kuralı: Rücu talebi > 50.000 TL → rücu avukatı onayı. Sigorta ### Rücu ve tahsilat [Rücu ve tahsilat] çalıştırma #41.288.620süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Rücu talebi > 50.000 TL → rücu avukatı onayı Bu çalıştırmadaki ajanlar Rücu Süpervizörüsupervisorplanlar, yönlendirir, onaylar Tarama Ajanıaginies-talay-2 · on-premKapanmış dosyalarda üçüncü taraf kusurunu skorlar Kanıt Ajanıaginies-tulpar-4 · on-premKanıtları özetler, yasal dayanakla talep yazısını yazar Tahsilat Ajanıaginies-sunkar-1 · on-premKarşı sigortacıyı bulur, rücu dosyasını açar, yazıyı gönderir Denetim izi00:00.00 - ◦Haftalık toplu iş: 1.240 kapanmış kasko dosyası rücu için tarandı - ∿Dosya #CL-2026-104877: ödenen 86.000 TL, karşı taraf kusuru %100 (tutanak) - ⌘Plan: kanıt paketi → karşı sigortacı → talep yazısı → onay → gönder - ⌕Tutanak, fotoğraflar ve onarım faturaları alındı (11 belge) - ⚙Karşı araç 06 XYZ 456 → trafik sigortacısı sicilde bulundu, poliçe aktif - ∿Rücu talebi taslağı: 86.000 TL, faiz başlangıcı 14.04, yasal dayanak eklendi - ✓Zamanaşımı: 2 yıl içinde · tutar dosyayla mutabık · yasal dayanak doğru - ⏸Talep > 50.000 TL → rücu avukatı yazıyı onayladı - ⇢Rücu dosyası #SR-8817 açıldı, beklenen tahsilat 86.000 TL - ➤Talep yazısı karşı sigortacıya e-postalandı, 30 gün takip kuruldu - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Haftalık toplu iş: son 90 günde kapanmış kasko dosyaları Sonuç Rücu dosyası açık, talep yazısı karşı sigortacıda, 30 günlük takip kuruldu ## Ajanlar SÜPERVİZÖR ### Rücu Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Tarama Ajanı Kapanmış dosyalarda üçüncü taraf kusurunu skorlar get_closed_claimsscore_third_party_liabilityget_accident_report aginies-tulpar-4 · on-prem ### Kanıt Ajanı Kanıtları özetler, yasal dayanakla talep yazısını yazar fetch_claim_documentssummarise_evidencedraft_demand_lettercheck_limitation_period aginies-sunkar-1 · on-prem ### Tahsilat Ajanı Karşı sigortacıyı bulur, rücu dosyasını açar, yazıyı gönderir lookup_insurercreate_recovery_casesend_emailschedule_followup ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Rücu Süpervizörü | Haftalık toplu iş: 1.240 kapanmış kasko dosyası rücu için tarandı | — | 340 | — | | 02 | akıl yürütme | Tarama Ajanı | Dosya #CL-2026-104877: ödenen 86.000 TL, karşı taraf kusuru %100 (tutanak) | — | 820 | 760 | | 03 | plan | Rücu Süpervizörü | Plan: kanıt paketi → karşı sigortacı → talep yazısı → onay → gönder | — | 280 | 200 | | 04 | bilgi | Kanıt Ajanı | Tutanak, fotoğraflar ve onarım faturaları alındı (11 belge) | Document Store (S3) | 420 | — | | 05 | araç | Tahsilat Ajanı | Karşı araç 06 XYZ 456 → trafik sigortacısı sicilde bulundu, poliçe aktif | Insurance Registry (SBM) | 460 | — | | 06 | akıl yürütme | Kanıt Ajanı | Rücu talebi taslağı: 86.000 TL, faiz başlangıcı 14.04, yasal dayanak eklendi | — | 1.300 | 1.580 | | 07 | doğrulama | Kanıt Ajanı | Zamanaşımı: 2 yıl içinde · tutar dosyayla mutabık · yasal dayanak doğru | — | 520 | 440 | | 08 | insan onayı | Rücu Süpervizörü | Talep > 50.000 TL → rücu avukatı yazıyı onayladı | — | 3.200 | — | | 09 | yazma | Tahsilat Ajanı | Rücu dosyası #SR-8817 açıldı, beklenen tahsilat 86.000 TL | Guidewire ClaimCenter | 480 | — | | 10 | bildirim | Tahsilat Ajanı | Talep yazısı karşı sigortacıya e-postalandı, 30 gün takip kuruldu | Email (Exchange) | 260 | — | Simülasyon metrikleri (tahmini) 3.400çalıştırma / ay %95,5başarı 8,1 sp50 $0,0087tahmini maliyet / çalıştırma Onay kuralı Rücu talebi > 50.000 TL → rücu avukatı onayı Adım profili giriş · 1plan · 1bilgi · 1akıl yürütme · 2araç · 1doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - Guidewire ClaimCentercore - Document Store (S3)storage - Insurance Registry (SBM)api - Email (Exchange)email Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Doküman deposuna okuma erişimi ve belge tipleri - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek ABD sigortacılarının "kaçırılan rücu" fırsatlarını yakalayan analitik deseni Bu senaryonun yer aldığı paketler Hasar operasyonları uçtan uca → Sektör: Sigorta → İlgili ajanlar SigortaHITL ### Uçtan uca kasko hasarı: ihbardan ödemeye Sesli ihbar, fotoğraf ekspertizi, anlaşmalı servis ataması ve ödeme talimatı tek akışta; eksper yalnızca eşik üstünde devreye girer. 21.000 çalıştırma / ay %91,2 başarı 5 ajan SigortaHITL ### Sesli hasar ihbarı (FNOL) Müşteri arar ve anlatır; poliçe plakadan bulunur, teminat kontrol edilir, dosya açılır ve eksper atanır. 34.000 çalıştırma / ay %95,8 başarı 5 ajan SigortaHITL ### Poliçe yenileme teklifi Poliçeler yeniden fiyatlanır, sert artışlar onaya gider; müşteri tek dokunuşla onaylar, poliçe anında kesilir. 22.000 çalıştırma / ay %94,4 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Abone hizmet asistanı: fatura, paket, arıza ve taahhüt URL: https://www.aginies.com/tr/ajanlar/subscriber-care-assistant Language: tr Summary: Abone arar ya da yazar; fatura itirazı, paket değişikliği ve hat arızası aynı oturumda BSS ve şebeke sistemlerinde çözülür. IVR, uygulama sohbeti veya WhatsApp üzerinden gelen çok konulu talep → Fatura düzeltildi, paket değişti, modem uzaktan onarıldı, taahhüt bilgisi verildi Telekomvoiceappwhatsapp # Abone hizmet asistanı: fatura, paket, arıza ve taahhüt Abone arar ya da yazar; fatura itirazı, paket değişikliği ve hat arızası aynı oturumda BSS ve şebeke sistemlerinde çözülür. Kısaca Abone hizmet asistanı: fatura, paket, arıza ve taahhüt, Telekom sektöründe voice, app, whatsapp kanalında çalışan bir ajan senaryosudur. Tetikleyici: IVR, uygulama sohbeti veya WhatsApp üzerinden gelen çok konulu talep. Abone Hizmet Süpervizörü 4 uzman ajanı (Diyalog Ajanı, Hesap Ajanı, Şebeke Ajanı, Taahhüt Ajanı) 5 sistem üzerinde yönetir (IVR / SIP, BSS / Billing, OSS Network, CRM, SMS Gateway); 10 adımın 3 tanesi sistemde işlem, 1 tanesi doğrulama veya onaydır. Sonuç: Fatura düzeltildi, paket değişti, modem uzaktan onarıldı, taahhüt bilgisi verildi. Onay kuralı: Taahhüt iptali, 500 TL üzeri fatura düzeltmesi veya teknisyen önceliği → temsilci onayı. Telekom ### Abone hizmet asistanı: fatura, paket, arıza ve taahhüt [Abone hizmet asistanı: fatura, paket, arıza ve taahhüt] çalıştırma #41.288.418süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Taahhüt iptali, 500 TL üzeri fatura düzeltmesi veya teknisyen önceliği → temsilci onayı Bu çalıştırmadaki ajanlar Abone Hizmet Süpervizörüsupervisorplanlar, yönlendirir, onaylar Diyalog Ajanıaginies-talay-2 · on-premKonuşmayı yürütür, konuları ayrıştırır, sesli ve yazılı yanıt verir Hesap Ajanıaginies-sunkar-1 · on-premFatura kalemlerini açıklar, düzeltme ve paket değişikliğini uygular Şebeke Ajanıaginies-tulpar-4 · on-premHat ve modem tanısı yapar, uzaktan onarım ve teknisyen ataması kararı verir Taahhüt Ajanıaginies-sunkar-1 · on-premTaahhüt ve cayma koşullarını hesaplar, uygun teklifi seçer Denetim izi00:00.00 - ◦Çağrı: "Faturam iki katı gelmiş, internet de iki gündür kopuyor, paketi değiştirmek istiyorum" - ∿Üç konu: fatura itirazı · arıza · paket değişikliği; kimlik IVR PIN ile doğrulandı - ⌘Plan: fatura kalemleri → düzeltme → hat testi → onarım → taahhüt → paket → özet - ⚙Fatura 486 TL: yurt dışı dolaşım 212 TL (3 gün), paket dışı kalem yok - ⇢Dolaşım paketi geriye dönük tanımlandı, 212 → 89 TL; fark bir sonraki faturada - ⚙Hat testi: sinyal/gürültü sınırda, modem 41 gündür yeniden başlamamış, firmware eski - ⇢Modem uzaktan güncellendi ve yeniden başlatıldı; test: 94 Mbps kararlı, teknisyen gerekmedi - ✓Taahhüt 7 ay kaldı; üst pakete geçişte cayma yok, aylık +60 TL, 3 ay indirimli - ⇢Paket değişikliği onaylandı, bir sonraki fatura döneminden geçerli, ref #PL-99102 - ➤Sesli özet okundu, SMS ile üç işlemin referansı gönderildi, memnuniyet anketi tetiklendi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici IVR, uygulama sohbeti veya WhatsApp üzerinden gelen çok konulu talep Sonuç Fatura düzeltildi, paket değişti, modem uzaktan onarıldı, taahhüt bilgisi verildi ## Ajanlar SÜPERVİZÖR ### Abone Hizmet Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Diyalog Ajanı Konuşmayı yürütür, konuları ayrıştırır, sesli ve yazılı yanıt verir detect_intentscompose_replytts_speak aginies-sunkar-1 · on-prem ### Hesap Ajanı Fatura kalemlerini açıklar, düzeltme ve paket değişikliğini uygular get_billadjust_billchange_planget_commitment aginies-tulpar-4 · on-prem ### Şebeke Ajanı Hat ve modem tanısı yapar, uzaktan onarım ve teknisyen ataması kararı verir line_testmodem_resetschedule_technician aginies-sunkar-1 · on-prem ### Taahhüt Ajanı Taahhüt ve cayma koşullarını hesaplar, uygun teklifi seçer calc_exit_feeget_offerspolicy_check ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Abone Hizmet Süpervizörü | Çağrı: "Faturam iki katı gelmiş, internet de iki gündür kopuyor, paketi değiştirmek istiyorum" | IVR / SIP | 200 | — | | 02 | akıl yürütme | Diyalog Ajanı | Üç konu: fatura itirazı · arıza · paket değişikliği; kimlik IVR PIN ile doğrulandı | — | 640 | 380 | | 03 | plan | Abone Hizmet Süpervizörü | Plan: fatura kalemleri → düzeltme → hat testi → onarım → taahhüt → paket → özet | — | 240 | 220 | | 04 | araç | Hesap Ajanı | Fatura 486 TL: yurt dışı dolaşım 212 TL (3 gün), paket dışı kalem yok | BSS / Billing | 440 | — | | 05 | yazma | Hesap Ajanı | Dolaşım paketi geriye dönük tanımlandı, 212 → 89 TL; fark bir sonraki faturada | BSS / Billing | 560 | — | | 06 | araç | Şebeke Ajanı | Hat testi: sinyal/gürültü sınırda, modem 41 gündür yeniden başlamamış, firmware eski | OSS Network | 1.600 | — | | 07 | yazma | Şebeke Ajanı | Modem uzaktan güncellendi ve yeniden başlatıldı; test: 94 Mbps kararlı, teknisyen gerekmedi | OSS Network | 2.400 | — | | 08 | doğrulama | Taahhüt Ajanı | Taahhüt 7 ay kaldı; üst pakete geçişte cayma yok, aylık +60 TL, 3 ay indirimli | CRM | 480 | 260 | | 09 | yazma | Hesap Ajanı | Paket değişikliği onaylandı, bir sonraki fatura döneminden geçerli, ref #PL-99102 | BSS / Billing | 420 | — | | 10 | bildirim | Diyalog Ajanı | Sesli özet okundu, SMS ile üç işlemin referansı gönderildi, memnuniyet anketi tetiklendi | SMS Gateway | 400 | 200 | Simülasyon metrikleri (tahmini) 410.000çalıştırma / ay %94,3başarı 7,4 sp50 $0,0019tahmini maliyet / çalıştırma Onay kuralı Taahhüt iptali, 500 TL üzeri fatura düzeltmesi veya teknisyen önceliği → temsilci onayı Adım profili giriş · 1plan · 1akıl yürütme · 1araç · 2doğrulama · 1yazma · 3bildirim · 1 SİSTEMLER - IVR / SIPvoice - BSS / Billingcore - OSS Networkapi - CRMcrm - SMS Gatewaychannel Çalıştırmak için gerekenler - Santral ile SIP entegrasyonu ve kayıt erişimi - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - CRM için API erişimi ve müşteri eşleme anahtarı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Operatörlerin dijital asistanlarının işlem yapan sürümü; şebeke tanısıyla birleştirilmiş Bu senaryonun yer aldığı paketler Ses analizi ve kalite →Abone hizmetleri ve şebeke desteği → Sektör: Telekom → İlgili ajanlar TelekomHITL ### Fatura itirazı çözümü Fatura kalemleri ve kullanım kayıtları saniyede açıklanır; iyi niyet kredisi politikaya göre uygulanır. 88.000 çalıştırma / ay %96,3 başarı 4 ajan TelekomHITL ### SIM değişimi dolandırıcılık kontrolü Her SIM değişimi provizyondan önce risk skorlanır; şüpheli talepler bekletilir ve bankalara sinyal iletilir. 41.000 çalıştırma / ay %97,7 başarı 4 ajan TelekomHITL ### Şebeke kesintisi iletişimi Kesinti kaydı açılınca etkilenen aboneler bulunur, mesaj yazılıp onaylanır; müşteri aramadan bilgilendirilir. 1.400 çalıştırma / ay %98,5 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Tedarikçi faturası üçlü eşleştirme URL: https://www.aginies.com/tr/ajanlar/supplier-invoice-three-way-match Language: tr Summary: Fatura, sipariş ve mal kabul SAP'de otomatik eşleşir; yalnızca tolerans dışı kalemler satın almacıya gider. Tedarikçi faturasının ap@ posta kutusuna PDF olarak gelmesi → Fatura SAP'de kaydedildi, ödeme bloğu kaldırıldı, sapma gerekçesi denetim izinde Üretim & Otomotivemailapibatch # Tedarikçi faturası üçlü eşleştirme Fatura, sipariş ve mal kabul SAP'de otomatik eşleşir; yalnızca tolerans dışı kalemler satın almacıya gider. Kısaca Tedarikçi faturası üçlü eşleştirme, Üretim & Otomotiv sektöründe email, api, batch kanalında çalışan bir ajan senaryosudur. Tetikleyici: Tedarikçi faturasının ap@ posta kutusuna PDF olarak gelmesi. Fatura Süpervizörü 3 uzman ajanı (Yakalama Ajanı, Eşleştirme Ajanı, Kayıt Ajanı) 3 sistem üzerinde yönetir (SAP S/4HANA, Exchange / M365 Mail, Tolerance Policy Vector Store); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Fatura SAP'de kaydedildi, ödeme bloğu kaldırıldı, sapma gerekçesi denetim izinde. Onay kuralı: Fiyat sapması > %2 veya > 1.000 TL → satın alma sorumlusu onayı. Üretim & Otomotiv ### Tedarikçi faturası üçlü eşleştirme [Tedarikçi faturası üçlü eşleştirme] çalıştırma #41.290.297süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Fiyat sapması > %2 veya > 1.000 TL → satın alma sorumlusu onayı Bu çalıştırmadaki ajanlar Fatura Süpervizörüsupervisorplanlar, yönlendirir, onaylar Yakalama Ajanıaginies-konrul-1 · on-premOCR ile fatura alanlarını çıkarır Eşleştirme Ajanıaginies-tulpar-4 · on-premSipariş ve mal kabulle karşılaştırır, tolerans uygular Kayıt Ajanıaginies-talay-2 · on-premFaturayı SAP'ye kaydeder ya da park eder Denetim izi00:00.00 - ◦Demir Metal A.Ş.'den ap@ kutusuna fatura PDF'i alındı - ∿OCR: fatura DM-2026-4471 · 184.320 TL · 3 kalem · SAS 4500123877 referanslı - ⌘Plan: sipariş → mal kabul → tolerans kontrolü → kayıt ya da park - ⚙SAS 4500123877 çekildi: 3 kalem, 180.000 TL, tedarikçi 100428 - ⚙Mal kabul 5000871122: 3 kalem, 500/500/250 adet, 2 Eylül'de teslim - ⌕Tolerans politikası: fiyat ±%2, miktar ±%0, navlun en çok 3.000 TL - ✓Kalem 2 fiyat +%2,4 (> %2 tolerans) → 4.320 TL sapma işaretlendi - ⏸Sapma > %2 ve > 1.000 TL → satın almacı onayı istendi, notla onaylandı - ⇢Fatura kaydedildi (belge 5105634201), ödeme bloğu kaldırıldı, vade 2 Ekim - ➤Tedarikçiye alındı bilgisi gönderildi, OCR sayfaları ve not denetim izine eklendi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Tedarikçi faturasının ap@ posta kutusuna PDF olarak gelmesi Sonuç Fatura SAP'de kaydedildi, ödeme bloğu kaldırıldı, sapma gerekçesi denetim izinde ## Ajanlar SÜPERVİZÖR ### Fatura Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### Yakalama Ajanı OCR ile fatura alanlarını çıkarır ocr_documentextract_invoice_fields aginies-tulpar-4 · on-prem ### Eşleştirme Ajanı Sipariş ve mal kabulle karşılaştırır, tolerans uygular get_purchase_orderget_goods_receiptsearch_tolerance_policyapply_tolerance_rules aginies-talay-2 · on-prem ### Kayıt Ajanı Faturayı SAP'ye kaydeder ya da park eder post_invoicepark_invoicesend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Fatura Süpervizörü | Demir Metal A.Ş.'den ap@ kutusuna fatura PDF'i alındı | — | 200 | — | | 02 | akıl yürütme | Yakalama Ajanı | OCR: fatura DM-2026-4471 · 184.320 TL · 3 kalem · SAS 4500123877 referanslı | — | 880 | 720 | | 03 | plan | Fatura Süpervizörü | Plan: sipariş → mal kabul → tolerans kontrolü → kayıt ya da park | — | 280 | 190 | | 04 | araç | Eşleştirme Ajanı | SAS 4500123877 çekildi: 3 kalem, 180.000 TL, tedarikçi 100428 | SAP S/4HANA | 430 | — | | 05 | araç | Eşleştirme Ajanı | Mal kabul 5000871122: 3 kalem, 500/500/250 adet, 2 Eylül'de teslim | SAP S/4HANA | 410 | — | | 06 | bilgi | Eşleştirme Ajanı | Tolerans politikası: fiyat ±%2, miktar ±%0, navlun en çok 3.000 TL | Tolerance Policy Vector Store | 360 | 300 | | 07 | doğrulama | Eşleştirme Ajanı | Kalem 2 fiyat +%2,4 (> %2 tolerans) → 4.320 TL sapma işaretlendi | — | 760 | 890 | | 08 | insan onayı | Fatura Süpervizörü | Sapma > %2 ve > 1.000 TL → satın almacı onayı istendi, notla onaylandı | — | 2.600 | — | | 09 | yazma | Kayıt Ajanı | Fatura kaydedildi (belge 5105634201), ödeme bloğu kaldırıldı, vade 2 Ekim | SAP S/4HANA | 560 | — | | 10 | bildirim | Kayıt Ajanı | Tedarikçiye alındı bilgisi gönderildi, OCR sayfaları ve not denetim izine eklendi | Exchange / M365 Mail | 240 | — | Simülasyon metrikleri (tahmini) 21.000çalıştırma / ay %97,1başarı 6,7 sp50 $0,0042tahmini maliyet / çalıştırma Onay kuralı Fiyat sapması > %2 veya > 1.000 TL → satın alma sorumlusu onayı Adım profili giriş · 1plan · 1bilgi · 1akıl yürütme · 1araç · 2doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - SAP S/4HANAerp - Exchange / M365 Mailemail - Tolerance Policy Vector Storedb Çalıştırmak için gerekenler - ERP/MES için kayıt ve sorgu API erişimi - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük ortak hizmet merkezlerinin borçlar hesabı otomasyon modelini yansıtır Bu senaryonun yer aldığı paketler Fatura ve doküman veri çıkarımı →Finansal analiz ve operasyonel zekâ →Üretim ve bakım operasyonları → Sektör: Üretim ve Otomotiv → İlgili ajanlar Üretim & OtomotivHITL ### Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna MES alarmı düşer; kök neden tarihsel veriyle bulunur, yedek parça kontrol edilir, iş emri ve plan güncellemesi dakikalar içinde hazırdır. 3.100 çalıştırma / ay %92,2 başarı 5 ajan Üretim & OtomotivHITL ### Bakım talebi triyajı Sensör alarmı ve teknisyen sesli notu tek iş emrine dönüşür; yedek parça ve güvenli duruş penceresi hazırdır. 3.400 çalıştırma / ay %95,2 başarı 5 ajan Üretim & OtomotivHITL ### Garanti talebi işleme Bayi talebi dakikalarda kapsam, parça ve işçilik yönünden doğrulanır; eşik üstü tutar bölge müdürüne gider. 8.600 çalıştırma / ay %96,3 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Tedarikçi güvenlik ve uyum değerlendirmesi URL: https://www.aginies.com/tr/ajanlar/supplier-security-assessment Language: tr Summary: Tedarikçi anketleri ve belgeleri okunur, tesis güvenlik gereklilikleriyle karşılaştırılır, eksikler listelenir; nitelendirme kararı güvenlik müdürünün. Satın almadan yeni tedarikçi nitelendirme talebi → Değerlendirme raporu, eksik listesi ve risk sınıfı hazır; karar kayıt altında Savunma Sanayiiemailwebapi # Tedarikçi güvenlik ve uyum değerlendirmesi Tedarikçi anketleri ve belgeleri okunur, tesis güvenlik gereklilikleriyle karşılaştırılır, eksikler listelenir; nitelendirme kararı güvenlik müdürünün. Kısaca Tedarikçi güvenlik ve uyum değerlendirmesi, Savunma Sanayii sektöründe email, web, api kanalında çalışan bir ajan senaryosudur. Tetikleyici: Satın almadan yeni tedarikçi nitelendirme talebi. Tedarikçi Güvenlik Süpervizörü 3 uzman ajanı (Belge Ajanı, Değerlendirme Ajanı, Dosya Ajanı) 4 sistem üzerinde yönetir (Supplier Management (SRM), Security Requirements (vector), Document Store, E-mail); 9 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Değerlendirme raporu, eksik listesi ve risk sınıfı hazır; karar kayıt altında. Onay kuralı: Nitelendirme kararı her zaman tesis güvenlik müdüründe; ajan yalnızca hazırlar. Savunma Sanayii ### Tedarikçi güvenlik ve uyum değerlendirmesi [Tedarikçi güvenlik ve uyum değerlendirmesi] çalıştırma #41.289.948süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Nitelendirme kararı her zaman tesis güvenlik müdüründe; ajan yalnızca hazırlar Bu çalıştırmadaki ajanlar Tedarikçi Güvenlik Süpervizörüsupervisorplanlar, yönlendirir, onaylar Belge Ajanıaginies-konrul-1 · on-premAnket, sertifika ve politika belgelerini okur Değerlendirme Ajanıaginies-tulpar-4 · on-premCevapları gereklilik setiyle karşılaştırır Dosya Ajanıaginies-talay-2 · on-premRaporu yazar, tedarikçiye eksikleri iletir Denetim izi00:00.00 - ◦Nitelendirme talebi: kompozit parça tedarikçisi, 3 belge + 84 soruluk anket - ⚙Belgeler alındı: ISO 27001 sertifikası, bilgi güvenliği politikası, tesis planı - ∿Sertifika geçerli (2027-03); politika sürümü 2024, erişim kontrolü bölümü eksik - ⌕Gereklilik seti: tesis güvenliği, personel güvenliği, bilgi güvenliği, alt yüklenici kontrolü - ∿84 cevap puanlandı: 71 uygun, 9 kısmi, 4 eksik; risk sınıfı: orta - ✓Kritik kontrol: alt yüklenici bildirimi yok → nitelendirme öncesi zorunlu - ⏸Orta risk + kritik eksik → tesis güvenlik müdürü şartlı nitelendirmeyi onayladı - ⇢Değerlendirme SA-2026-117 kaydedildi; tedarikçi durumu “şartlı”, 30 gün süre - ➤Tedarikçiye eksik listesi ve kanıt şablonları gönderildi; satın alma bilgilendirildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Satın almadan yeni tedarikçi nitelendirme talebi Sonuç Değerlendirme raporu, eksik listesi ve risk sınıfı hazır; karar kayıt altında ## Ajanlar SÜPERVİZÖR ### Tedarikçi Güvenlik Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### Belge Ajanı Anket, sertifika ve politika belgelerini okur extract_documentclassify_documentcheck_certificate_validity aginies-tulpar-4 · on-prem ### Değerlendirme Ajanı Cevapları gereklilik setiyle karşılaştırır search_requirementsscore_controllist_gaps aginies-talay-2 · on-prem ### Dosya Ajanı Raporu yazar, tedarikçiye eksikleri iletir create_assessmentupdate_supplier_statussend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Tedarikçi Güvenlik Süpervizörü | Nitelendirme talebi: kompozit parça tedarikçisi, 3 belge + 84 soruluk anket | — | 200 | — | | 02 | araç | Belge Ajanı | Belgeler alındı: ISO 27001 sertifikası, bilgi güvenliği politikası, tesis planı | Document Store | 460 | — | | 03 | akıl yürütme | Belge Ajanı | Sertifika geçerli (2027-03); politika sürümü 2024, erişim kontrolü bölümü eksik | — | 900 | 1.200 | | 04 | bilgi | Değerlendirme Ajanı | Gereklilik seti: tesis güvenliği, personel güvenliği, bilgi güvenliği, alt yüklenici kontrolü | Security Requirements (vector) | 520 | 640 | | 05 | akıl yürütme | Değerlendirme Ajanı | 84 cevap puanlandı: 71 uygun, 9 kısmi, 4 eksik; risk sınıfı: orta | — | 1.100 | 1.800 | | 06 | doğrulama | Değerlendirme Ajanı | Kritik kontrol: alt yüklenici bildirimi yok → nitelendirme öncesi zorunlu | — | 380 | 260 | | 07 | insan onayı | Tedarikçi Güvenlik Süpervizörü | Orta risk + kritik eksik → tesis güvenlik müdürü şartlı nitelendirmeyi onayladı | — | 3.000 | — | | 08 | yazma | Dosya Ajanı | Değerlendirme SA-2026-117 kaydedildi; tedarikçi durumu “şartlı”, 30 gün süre | Supplier Management (SRM) | 480 | — | | 09 | bildirim | Dosya Ajanı | Tedarikçiye eksik listesi ve kanıt şablonları gönderildi; satın alma bilgilendirildi | E-mail | 260 | — | Simülasyon metrikleri (tahmini) 900çalıştırma / ay %95,2başarı 7,8 sp50 $0,0147tahmini maliyet / çalıştırma Onay kuralı Nitelendirme kararı her zaman tesis güvenlik müdüründe; ajan yalnızca hazırlar Adım profili giriş · 1bilgi · 1akıl yürütme · 2araç · 1doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - Supplier Management (SRM)erp - Security Requirements (vector)db - Document Storestorage - E-mailemail Çalıştırmak için gerekenler - ERP/MES için kayıt ve sorgu API erişimi - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Doküman deposuna okuma erişimi ve belge tipleri - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Savunma ana yüklenicilerinin tedarik zinciri güvenlik programlarını (anket + belge + insan kararı) yansıtır Bu senaryonun yer aldığı paketler Kamu alımı ve ihale uyumu →Savunma ihracat kontrolü ve tedarik zinciri güvenliği → Sektör: Savunma Sanayii → İlgili ajanlar Savunma SanayiiHITL ### Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolü Teknisyen hangarda sorar; doğru revizyondaki prosedür, parça numarası ve emniyet notu yetki kontrolüyle gelir, her erişim iz bırakır. 9.800 çalıştırma / ay %95,7 başarı 5 ajan Savunma SanayiiHITL ### İhracat kontrolü ve son kullanıcı taraması Parça, alıcı ve son kullanım tek akışta sınıflandırılır; lisans gereği tespit edilir, karar her zaman uyum yöneticisinde kalır. 2.600 çalıştırma / ay %96,1 başarı 4 ajan Savunma SanayiiHITL ### Platform bakım (MRO) iş emri hazırlığı Arıza raporu teknik yayınlarla eşlenir, görev kartı ve parça listesi çıkar, iş emri planlayıcı onayıyla açılır. 4.100 çalıştırma / ay %95,7 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolü URL: https://www.aginies.com/tr/ajanlar/technical-publication-assistant Language: tr Summary: Teknisyen hangarda sorar; doğru revizyondaki prosedür, parça numarası ve emniyet notu yetki kontrolüyle gelir, her erişim iz bırakır. Kapalı ağdaki tablet uygulamasından gelen bakım sorusu → Prosedür ve parça bilgisi verildi, revizyon doğrulandı, iş paketine bağlandı, erişim kaydedildi Savunma Sanayiiapp # Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolü Teknisyen hangarda sorar; doğru revizyondaki prosedür, parça numarası ve emniyet notu yetki kontrolüyle gelir, her erişim iz bırakır. Kısaca Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolü, Savunma Sanayii sektöründe app kanalında çalışan bir ajan senaryosudur. Tetikleyici: Kapalı ağdaki tablet uygulamasından gelen bakım sorusu. Teknik Yayın Süpervizörü 4 uzman ajanı (Yetki Ajanı, Yayın Ajanı, Prosedür Ajanı, Parça Ajanı) 5 sistem üzerinde yönetir (Hangar Tablet App, Access Control, Tech Pub Repository (S1000D), MRO System, Audit Log); 10 adımın 2 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Prosedür ve parça bilgisi verildi, revizyon doğrulandı, iş paketine bağlandı, erişim kaydedildi. Onay kuralı: Prosedürden sapma önerisi → yetkili mühendis onayı; ihracat kontrollü içerik → yetki seviyesi şart. Savunma Sanayii ### Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolü [Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolü] çalıştırma #41.289.789süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Prosedürden sapma önerisi → yetkili mühendis onayı; ihracat kontrollü içerik → yetki seviyesi şart Bu çalıştırmadaki ajanlar Teknik Yayın Süpervizörüsupervisorplanlar, yönlendirir, onaylar Yetki Ajanıaginies-sunkar-1 · on-premKullanıcının yetki seviyesini ve içeriğin ihracat kontrol işaretini karşılaştırır Yayın Ajanıaginies-sunkar-1 · on-premS1000D veri modüllerinde doğru revizyonu bulur Prosedür Ajanıaginies-tulpar-4 · on-premProsedürü platform konfigürasyonuna göre yorumlar, emniyet notlarını öne çıkarır Parça Ajanıaginies-sunkar-1 · on-premParça numarasını, alternatiflerini ve stok durumunu MRO sisteminden getirir Denetim izi00:00.00 - ◦Tablet: "Kuyruk no 07-14, hidrolik pompa değişimi, tork değerleri ve emniyet adımları?" - ✓Kullanıcı yetkisi: seviye 2; modül işareti seviye 2, ihracat kontrollü değil → erişim uygun - ⌘Plan: uygulanabilirlik → güncel revizyon → konfigürasyon → prosedür → parça → iş paketi → iz - ⌕Veri modülü DM-29-10-05 bulundu; revizyon 14 (3 ay önce), kuyruk 07-14 için uygulanabilir - ✓Revizyon 13 ile fark: tork değeri güncellenmiş, ek bir uyarı eklenmiş → fark özeti hazırlandı - ∿Prosedür 07-14 konfigürasyonuna uyarlandı: 11 adım, 3 UYARI, 2 DİKKAT; tork 48 ± 2 Nm - ⚙Pompa P/N 7742-A2: depoda 1 adet, alternatif 7742-A3 (bülten gerekli); contalar stokta - ⇢Parça ve prosedür iş paketi #WP-3391'e bağlandı; kalite kontrol imza adımı eklendi - ➤Tablete: adım listesi, uyarılar öne çıkarılmış, revizyon farkı ve parça kartı - ⇢Erişim kaydı: kullanıcı, modül, revizyon, zaman ve iş paketi denetim günlüğüne yazıldı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Kapalı ağdaki tablet uygulamasından gelen bakım sorusu Sonuç Prosedür ve parça bilgisi verildi, revizyon doğrulandı, iş paketine bağlandı, erişim kaydedildi ## Ajanlar SÜPERVİZÖR ### Teknik Yayın Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-sunkar-1 · on-prem ### Yetki Ajanı Kullanıcının yetki seviyesini ve içeriğin ihracat kontrol işaretini karşılaştırır get_clearancecheck_markinglog_access aginies-sunkar-1 · on-prem ### Yayın Ajanı S1000D veri modüllerinde doğru revizyonu bulur search_modulesget_revisionget_applicability aginies-tulpar-4 · on-prem ### Prosedür Ajanı Prosedürü platform konfigürasyonuna göre yorumlar, emniyet notlarını öne çıkarır apply_configurationextract_warningscompose_answer aginies-sunkar-1 · on-prem ### Parça Ajanı Parça numarasını, alternatiflerini ve stok durumunu MRO sisteminden getirir lookup_partget_alternatescheck_stock ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Teknik Yayın Süpervizörü | Tablet: "Kuyruk no 07-14, hidrolik pompa değişimi, tork değerleri ve emniyet adımları?" | Hangar Tablet App | 160 | — | | 02 | doğrulama | Yetki Ajanı | Kullanıcı yetkisi: seviye 2; modül işareti seviye 2, ihracat kontrollü değil → erişim uygun | Access Control | 380 | — | | 03 | plan | Teknik Yayın Süpervizörü | Plan: uygulanabilirlik → güncel revizyon → konfigürasyon → prosedür → parça → iş paketi → iz | — | 240 | 220 | | 04 | bilgi | Yayın Ajanı | Veri modülü DM-29-10-05 bulundu; revizyon 14 (3 ay önce), kuyruk 07-14 için uygulanabilir | Tech Pub Repository (S1000D) | 900 | 320 | | 05 | doğrulama | Yayın Ajanı | Revizyon 13 ile fark: tork değeri güncellenmiş, ek bir uyarı eklenmiş → fark özeti hazırlandı | — | 700 | 380 | | 06 | akıl yürütme | Prosedür Ajanı | Prosedür 07-14 konfigürasyonuna uyarlandı: 11 adım, 3 UYARI, 2 DİKKAT; tork 48 ± 2 Nm | — | 2.200 | 1.500 | | 07 | araç | Parça Ajanı | Pompa P/N 7742-A2: depoda 1 adet, alternatif 7742-A3 (bülten gerekli); contalar stokta | MRO System | 520 | — | | 08 | yazma | Parça Ajanı | Parça ve prosedür iş paketi #WP-3391'e bağlandı; kalite kontrol imza adımı eklendi | MRO System | 480 | — | | 09 | bildirim | Prosedür Ajanı | Tablete: adım listesi, uyarılar öne çıkarılmış, revizyon farkı ve parça kartı | Hangar Tablet App | 420 | 260 | | 10 | yazma | Yetki Ajanı | Erişim kaydı: kullanıcı, modül, revizyon, zaman ve iş paketi denetim günlüğüne yazıldı | Audit Log | 200 | — | Simülasyon metrikleri (tahmini) 9.800çalıştırma / ay %95,7başarı 6,2 sp50 $0,0024tahmini maliyet / çalıştırma Onay kuralı Prosedürden sapma önerisi → yetkili mühendis onayı; ihracat kontrollü içerik → yetki seviyesi şart Adım profili giriş · 1plan · 1bilgi · 1akıl yürütme · 1araç · 1doğrulama · 2yazma · 2bildirim · 1 SİSTEMLER - Hangar Tablet Appchannel - Access Controlapi - Tech Pub Repository (S1000D)storage - MRO Systemerp - Audit Logdb Çalıştırmak için gerekenler - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Doküman deposuna okuma erişimi ve belge tipleri - ERP/MES için kayıt ve sorgu API erişimi - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Havacılık ve savunma MRO'larının S1000D tabanlı akıllı teknik yayın deseni; kapalı ağda Bu senaryonun yer aldığı paketler Savunma ihracat kontrolü ve tedarik zinciri güvenliği → Sektör: Savunma Sanayii → İlgili ajanlar Savunma SanayiiHITL ### İhracat kontrolü ve son kullanıcı taraması Parça, alıcı ve son kullanım tek akışta sınıflandırılır; lisans gereği tespit edilir, karar her zaman uyum yöneticisinde kalır. 2.600 çalıştırma / ay %96,1 başarı 4 ajan Savunma SanayiiHITL ### Platform bakım (MRO) iş emri hazırlığı Arıza raporu teknik yayınlarla eşlenir, görev kartı ve parça listesi çıkar, iş emri planlayıcı onayıyla açılır. 4.100 çalıştırma / ay %95,7 başarı 5 ajan Savunma SanayiiHITL ### Tedarikçi güvenlik ve uyum değerlendirmesi Tedarikçi anketleri ve belgeleri okunur, tesis güvenlik gereklilikleriyle karşılaştırılır, eksikler listelenir; nitelendirme kararı güvenlik müdürünün. 900 çalıştırma / ay %95,2 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Underwriting belge çıkarımı URL: https://www.aginies.com/tr/ajanlar/underwriting-document-extraction Language: tr Summary: Sağlık raporu ve tahliller okunur, kılavuza göre değerlendirilir; underwriter yalnızca eşik üstünü görür. Acenteden e-postayla gelen hayat sigortası başvurusu ve ekleri → Başvuru kararı ve ek prim taslağı iş istasyonunda, gerekçe ve alanlar dosyada Sigortaemailwebbatch # Underwriting belge çıkarımı Sağlık raporu ve tahliller okunur, kılavuza göre değerlendirilir; underwriter yalnızca eşik üstünü görür. Kısaca Underwriting belge çıkarımı, Sigorta sektöründe email, web, batch kanalında çalışan bir ajan senaryosudur. Tetikleyici: Acenteden e-postayla gelen hayat sigortası başvurusu ve ekleri. Underwriting Süpervizörü 3 uzman ajanı (Belge Ajanı, Tıbbi Değerlendirme Ajanı, Underwriting Ajanı) 4 sistem üzerinde yönetir (Underwriting Workbench, Document Store (S3), Vector Store, Email (Exchange)); 9 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Başvuru kararı ve ek prim taslağı iş istasyonunda, gerekçe ve alanlar dosyada. Onay kuralı: Ek prim ≥ %50 veya teminat > 5.000.000 TL → kıdemli underwriter onayı. Sigorta ### Underwriting belge çıkarımı [Underwriting belge çıkarımı] çalıştırma #41.288.927süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Ek prim ≥ %50 veya teminat > 5.000.000 TL → kıdemli underwriter onayı Bu çalıştırmadaki ajanlar Underwriting Süpervizörüsupervisorplanlar, yönlendirir, onaylar Belge Ajanıaginies-konrul-1 · on-premSayfaları sınıflar, alanları ve tahlil tablolarını çıkarır Tıbbi Değerlendirme Ajanıaginies-tulpar-4 · on-premBulguları ICD kodlarına eşler, tıbbi özet çıkarır Underwriting Ajanıaginies-talay-2 · on-premKılavuzu getirir, kuralları uygular, başvuruyu günceller Denetim izi00:00.00 - ◦E-posta: hayat başvurusu #UW-55120, 6 ek (sağlık raporu, tahliller, beyan) - ⌕Ekler S3'e alındı; 14 sayfa sınıflandı, 32 satırlık tahlil tablosu çıkarıldı (%98,4) - ⌘Plan: tıbbi özet → UW kılavuzu → kural seti → onay → karar - ∿Bulgular: HbA1c 6,9 · BMI 31 · tansiyon ilacı → ICD E11, I10 - ⌕UW kılavuzu v7.2: tip 2 diyabet ve hipertansiyon bölümleri getirildi - ✓Kural: HbA1c 6,5–7,5 + kontrollü tansiyon → +%75 ek prim, red yok - ⏸Ek prim ≥ %50 → kıdemli underwriter kararı onayladı - ⇢Başvuru "ek primle kabul" olarak güncellendi, alanlar ve gerekçe eklendi - ➤Acenteye karar özeti ve ek primli teklif e-posta ile gönderildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Acenteden e-postayla gelen hayat sigortası başvurusu ve ekleri Sonuç Başvuru kararı ve ek prim taslağı iş istasyonunda, gerekçe ve alanlar dosyada ## Ajanlar SÜPERVİZÖR ### Underwriting Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### Belge Ajanı Sayfaları sınıflar, alanları ve tahlil tablolarını çıkarır classify_documentextract_fieldsextract_tables aginies-tulpar-4 · on-prem ### Tıbbi Değerlendirme Ajanı Bulguları ICD kodlarına eşler, tıbbi özet çıkarır map_icd_codessummarise_historyscore_medical_risk aginies-talay-2 · on-prem ### Underwriting Ajanı Kılavuzu getirir, kuralları uygular, başvuruyu günceller get_uw_manualapply_uw_rulesupdate_applicationsend_email ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Underwriting Süpervizörü | E-posta: hayat başvurusu #UW-55120, 6 ek (sağlık raporu, tahliller, beyan) | — | 220 | — | | 02 | bilgi | Belge Ajanı | Ekler S3'e alındı; 14 sayfa sınıflandı, 32 satırlık tahlil tablosu çıkarıldı (%98,4) | Document Store (S3) | 1.600 | 3.100 | | 03 | plan | Underwriting Süpervizörü | Plan: tıbbi özet → UW kılavuzu → kural seti → onay → karar | — | 280 | 200 | | 04 | akıl yürütme | Tıbbi Değerlendirme Ajanı | Bulgular: HbA1c 6,9 · BMI 31 · tansiyon ilacı → ICD E11, I10 | — | 940 | 1.120 | | 05 | bilgi | Underwriting Ajanı | UW kılavuzu v7.2: tip 2 diyabet ve hipertansiyon bölümleri getirildi | Vector Store | 360 | — | | 06 | doğrulama | Underwriting Ajanı | Kural: HbA1c 6,5–7,5 + kontrollü tansiyon → +%75 ek prim, red yok | — | 720 | 860 | | 07 | insan onayı | Underwriting Süpervizörü | Ek prim ≥ %50 → kıdemli underwriter kararı onayladı | — | 3.000 | — | | 08 | yazma | Underwriting Ajanı | Başvuru "ek primle kabul" olarak güncellendi, alanlar ve gerekçe eklendi | Underwriting Workbench | 480 | — | | 09 | bildirim | Underwriting Ajanı | Acenteye karar özeti ve ek primli teklif e-posta ile gönderildi | Email (Exchange) | 240 | — | Simülasyon metrikleri (tahmini) 4.800çalıştırma / ay %94,9başarı 7,8 sp50 $0,0128tahmini maliyet / çalıştırma Onay kuralı Ek prim ≥ %50 veya teminat > 5.000.000 TL → kıdemli underwriter onayı Adım profili giriş · 1plan · 1bilgi · 2akıl yürütme · 1doğrulama · 1insan onayı · 1yazma · 1bildirim · 1 SİSTEMLER - Underwriting Workbenchcore - Document Store (S3)storage - Vector Storedb - Email (Exchange)email Çalıştırmak için gerekenler - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Doküman deposuna okuma erişimi ve belge tipleri - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Posta kutusu erişimi (Microsoft 365 veya Google Workspace) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Küresel hayat sigortacılarının otomatik underwriting (STP) oranını artıran belge çıkarımı deseni Bu senaryonun yer aldığı paketler Fatura ve doküman veri çıkarımı →KVKK kişisel veri koruma →Underwriting ve yenileme operasyonları → Sektör: Sigorta → İlgili ajanlar SigortaHITL ### Uçtan uca kasko hasarı: ihbardan ödemeye Sesli ihbar, fotoğraf ekspertizi, anlaşmalı servis ataması ve ödeme talimatı tek akışta; eksper yalnızca eşik üstünde devreye girer. 21.000 çalıştırma / ay %91,2 başarı 5 ajan SigortaHITL ### Sesli hasar ihbarı (FNOL) Müşteri arar ve anlatır; poliçe plakadan bulunur, teminat kontrol edilir, dosya açılır ve eksper atanır. 34.000 çalıştırma / ay %95,8 başarı 5 ajan SigortaHITL ### Poliçe yenileme teklifi Poliçeler yeniden fiyatlanır, sert artışlar onaya gider; müşteri tek dokunuşla onaylar, poliçe anında kesilir. 22.000 çalıştırma / ay %94,4 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda URL: https://www.aginies.com/tr/ajanlar/utility-customer-assistant Language: tr Summary: Abone arar ya da yazar; yüksek fatura açıklanır, kesinti süresi verilir, sayaç verisi gösterilir, taksit planı kurulur. IVR, WhatsApp veya abone uygulaması üzerinden gelen çok konulu talep → Fatura açıklandı, kesinti bilgisi verildi, taksit planı aktif, hatırlatma kuruldu Enerjivoicewhatsappapp # Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda Abone arar ya da yazar; yüksek fatura açıklanır, kesinti süresi verilir, sayaç verisi gösterilir, taksit planı kurulur. Kısaca Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda, Enerji sektöründe voice, whatsapp, app kanalında çalışan bir ajan senaryosudur. Tetikleyici: IVR, WhatsApp veya abone uygulaması üzerinden gelen çok konulu talep. Abone Hizmet Süpervizörü 4 uzman ajanı (Diyalog Ajanı, Fatura Ajanı, Şebeke Ajanı, Tahsilat Ajanı) 5 sistem üzerinde yönetir (IVR / SIP, Billing (CIS), Outage Management, Smart Meter Data, SMS Gateway); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Fatura açıklandı, kesinti bilgisi verildi, taksit planı aktif, hatırlatma kuruldu. Onay kuralı: Taksit > 6 ay, fatura iptali veya sayaç şüphesi → müşteri hizmetleri onayı. Enerji ### Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda [Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda] çalıştırma #41.289.459süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Taksit > 6 ay, fatura iptali veya sayaç şüphesi → müşteri hizmetleri onayı Bu çalıştırmadaki ajanlar Abone Hizmet Süpervizörüsupervisorplanlar, yönlendirir, onaylar Diyalog Ajanıaginies-talay-2 · on-premAboneyi doğrular, konuları ayrıştırır, sade yanıt üretir Fatura Ajanıaginies-sunkar-1 · on-premFatura kalemlerini ve tüketim profilini açıklar, düzeltme uygular Şebeke Ajanıaginies-tulpar-4 · on-premKesinti kaydını ve tahmini süreyi OMS'den alır, planlı çalışmaları bildirir Tahsilat Ajanıaginies-sunkar-1 · on-premTaksit politikasını uygular, planı kurar, hatırlatmayı planlar Denetim izi00:00.00 - ◦Çağrı: "Faturam neden 3 katı geldi, mahallede elektrik de yok, ödemeyi bölebilir miyim?" - ✓Abone numarası ve kayıtlı telefonla doğrulandı; üç konu ayrıştırıldı - ⌘Plan: kesinti durumu → fatura analizi → sayaç verisi → taksit → sesli özet - ⚙OMS: trafo arızası 11:20, 1.260 abone etkilendi, ekip sahada, tahmini 13:45 - ⚙Fatura 2.140 TL: 1.980 kWh, geçen yıl aynı dönem 690 kWh; okuma gerçek, tahmini değil - ⌕Sayaç serisi: 14 Ağustos'tan itibaren gece tüketimi 4 kat; sabit yük profili (su ısıtıcı / klima) - ∿Sonuç: fatura doğru, tüketim gerçek; aboneye tarih ve saatli tüketim özeti hazırlandı - ✓Taksit politikası: 2.140 TL → 4 taksit faizsiz uygun; önceki borç yok - ⇢Taksit planı oluşturuldu, ilk taksit 15 gün sonra; hatırlatma SMS planlandı - ➤Sesli özet: kesinti 13:45, tüketim açıklaması, taksit planı; SMS ile tüketim grafiği bağlantısı - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici IVR, WhatsApp veya abone uygulaması üzerinden gelen çok konulu talep Sonuç Fatura açıklandı, kesinti bilgisi verildi, taksit planı aktif, hatırlatma kuruldu ## Ajanlar SÜPERVİZÖR ### Abone Hizmet Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-talay-2 · on-prem ### Diyalog Ajanı Aboneyi doğrular, konuları ayrıştırır, sade yanıt üretir verify_customerdetect_intentscompose_replytts_speak aginies-sunkar-1 · on-prem ### Fatura Ajanı Fatura kalemlerini ve tüketim profilini açıklar, düzeltme uygular get_billget_meter_seriescompare_periodsadjust_bill aginies-tulpar-4 · on-prem ### Şebeke Ajanı Kesinti kaydını ve tahmini süreyi OMS'den alır, planlı çalışmaları bildirir outage_lookupget_etaplanned_works aginies-sunkar-1 · on-prem ### Tahsilat Ajanı Taksit politikasını uygular, planı kurar, hatırlatmayı planlar policy_checkcreate_instalment_planschedule_reminder ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Abone Hizmet Süpervizörü | Çağrı: "Faturam neden 3 katı geldi, mahallede elektrik de yok, ödemeyi bölebilir miyim?" | IVR / SIP | 200 | — | | 02 | doğrulama | Diyalog Ajanı | Abone numarası ve kayıtlı telefonla doğrulandı; üç konu ayrıştırıldı | — | 600 | 320 | | 03 | plan | Abone Hizmet Süpervizörü | Plan: kesinti durumu → fatura analizi → sayaç verisi → taksit → sesli özet | — | 240 | 220 | | 04 | araç | Şebeke Ajanı | OMS: trafo arızası 11:20, 1.260 abone etkilendi, ekip sahada, tahmini 13:45 | Outage Management | 520 | — | | 05 | araç | Fatura Ajanı | Fatura 2.140 TL: 1.980 kWh, geçen yıl aynı dönem 690 kWh; okuma gerçek, tahmini değil | Billing (CIS) | 480 | — | | 06 | bilgi | Fatura Ajanı | Sayaç serisi: 14 Ağustos'tan itibaren gece tüketimi 4 kat; sabit yük profili (su ısıtıcı / klima) | Smart Meter Data | 1.100 | 540 | | 07 | akıl yürütme | Fatura Ajanı | Sonuç: fatura doğru, tüketim gerçek; aboneye tarih ve saatli tüketim özeti hazırlandı | — | 900 | 480 | | 08 | doğrulama | Tahsilat Ajanı | Taksit politikası: 2.140 TL → 4 taksit faizsiz uygun; önceki borç yok | — | 420 | 180 | | 09 | yazma | Tahsilat Ajanı | Taksit planı oluşturuldu, ilk taksit 15 gün sonra; hatırlatma SMS planlandı | Billing (CIS) | 560 | — | | 10 | bildirim | Diyalog Ajanı | Sesli özet: kesinti 13:45, tüketim açıklaması, taksit planı; SMS ile tüketim grafiği bağlantısı | SMS Gateway | 440 | 240 | Simülasyon metrikleri (tahmini) 520.000çalıştırma / ay %94,6başarı 6,4 sp50 $0,0015tahmini maliyet / çalıştırma Onay kuralı Taksit > 6 ay, fatura iptali veya sayaç şüphesi → müşteri hizmetleri onayı Adım profili giriş · 1plan · 1bilgi · 1akıl yürütme · 1araç · 2doğrulama · 2yazma · 1bildirim · 1 SİSTEMLER - IVR / SIPvoice - Billing (CIS)core - Outage Managementapi - Smart Meter Datadb - SMS Gatewaychannel Çalıştırmak için gerekenler - Santral ile SIP entegrasyonu ve kayıt erişimi - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Dağıtım şirketlerinin dijital abone merkezi deseni; akıllı sayaç verisiyle fatura açıklaması Bu senaryonun yer aldığı paketler Sesli AI ile hasar ve talep kaydı →Enerji müşteri operasyonları → Sektör: Enerji → İlgili ajanlar EnerjiHITL ### Kesinti bildirimi yönetimi Bildirimler fider bazında kümelenir, tek OSS kaydı açılır, etkilenen herkes ekip yola çıkmadan bilgilenir. 38.000 çalıştırma / ay %97,7 başarı 5 ajan EnerjiHITL ### Sayaç okuma itirazı Müşteri sayaç fotoğrafı yollar; endeks görüntüden okunur, tahmini fatura düzeltilir, büyük iade lidere gider. 9.200 çalıştırma / ay %94,9 başarı 4 ajan EnerjiHITL ### Çatı GES bağlantı başvurusu Başvuru belgeleri okunur, trafo kapasitesi ve mevzuat kontrol edilir, eksikler aynı gün bildirilir; bağlantı görüşü mühendisin. 5.400 çalıştırma / ay %95,3 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Sesli bankacılıkla para transferi URL: https://www.aginies.com/tr/ajanlar/voice-banking-transfer Language: tr Summary: Müşteri arar ve konuşur; ses izi, güvenlik sorusu ve OTP ile doğrulanır, transfer temsilcisiz tamamlanır. Çağrı merkezine SIP üzerinden gelen "para göndermek istiyorum" çağrısı → Transfer çekirdek sistemde gerçekleşti, sesli ve SMS teyit verildi, kayıt arşivlendi Bankacılıkvoice # Sesli bankacılıkla para transferi Müşteri arar ve konuşur; ses izi, güvenlik sorusu ve OTP ile doğrulanır, transfer temsilcisiz tamamlanır. Kısaca Sesli bankacılıkla para transferi, Bankacılık sektöründe voice kanalında çalışan bir ajan senaryosudur. Tetikleyici: Çağrı merkezine SIP üzerinden gelen "para göndermek istiyorum" çağrısı. Sesli İşlem Süpervizörü 4 uzman ajanı (Konuşma Ajanı, Kimlik Ajanı, Transfer Ajanı, Diyalog Ajanı) 4 sistem üzerinde yönetir (Genesys SIP, Core Banking, Fraud Engine, SMS Gateway); 10 adımın 1 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Transfer çekirdek sistemde gerçekleşti, sesli ve SMS teyit verildi, kayıt arşivlendi. Onay kuralı: Tutar > 50.000 TL veya yeni alıcı → çağrı merkezi temsilcisine sıcak aktarım. Bankacılık ### Sesli bankacılıkla para transferi [Sesli bankacılıkla para transferi] çalıştırma #41.289.869süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Tutar > 50.000 TL veya yeni alıcı → çağrı merkezi temsilcisine sıcak aktarım Bu çalıştırmadaki ajanlar Sesli İşlem Süpervizörüsupervisorplanlar, yönlendirir, onaylar Konuşma Ajanıaginies-voice-tr-v3 · on-premCanlı sesi yazıya döker, dili ve güveni belirler Kimlik Ajanıaginies-sunkar-1 · on-premSes izi, güvenlik sorusu ve OTP ile kimliği doğrular Transfer Ajanıaginies-talay-2 · on-premAlıcıyı çözümler, limit ve dolandırıcılık kontrolü yapar, transferi gerçekleştirir Diyalog Ajanıaginies-tulpar-4 · on-premNiyeti çıkarır, detayları teyit ettirir, sesli yanıt üretir Denetim izi00:00.00 - ◦Çağrı SIP üzerinden bağlandı, arayan numara müşteri kaydıyla eşleşti - ∿Transkript: "Ayşe Demir'e 12.500 lira gönder" · dil: TR · güven 0,96 - ✓Ses izi eşleşmesi %94 + güvenlik sorusu doğru → kimlik doğrulandı - ⌘Plan: alıcıyı çözümle → limit/dolandırıcılık → OTP → transfer → sesli teyit - ⚙Kayıtlı alıcı "Ayşe Demir" TR12 0001 …8834 bulundu, günlük limit 50.000 TL - ⚙Dolandırıcılık motoru: skor 12/100, alıcı 8 aydır kayıtlı - ✓Politika: tutar ≤ 50.000 TL ve alıcı kayıtlı → uygun, OTP zorunlu - ➤OTP SMS gönderildi, müşteri 6 haneli kodu sesle okudu → doğrulandı - ⇢12.500 TL transfer çekirdek sistemde gerçekleşti, ref #TR-5583021 - ➤Sesli teyit okundu, özet SMS gönderildi, çağrı kaydı arşivlendi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Çağrı merkezine SIP üzerinden gelen "para göndermek istiyorum" çağrısı Sonuç Transfer çekirdek sistemde gerçekleşti, sesli ve SMS teyit verildi, kayıt arşivlendi ## Ajanlar SÜPERVİZÖR ### Sesli İşlem Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-voice-tr-v3 · on-prem ### Konuşma Ajanı Canlı sesi yazıya döker, dili ve güveni belirler transcribe_streamdetect_language aginies-sunkar-1 · on-prem ### Kimlik Ajanı Ses izi, güvenlik sorusu ve OTP ile kimliği doğrular verify_voiceprintask_security_questionsend_otpverify_otp aginies-talay-2 · on-prem ### Transfer Ajanı Alıcıyı çözümler, limit ve dolandırıcılık kontrolü yapar, transferi gerçekleştirir get_accountsresolve_beneficiarycheck_limitsget_fraud_scoreexecute_transfer aginies-tulpar-4 · on-prem ### Diyalog Ajanı Niyeti çıkarır, detayları teyit ettirir, sesli yanıt üretir extract_intentconfirm_detailssynthesize_speech ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Sesli İşlem Süpervizörü | Çağrı SIP üzerinden bağlandı, arayan numara müşteri kaydıyla eşleşti | — | 200 | — | | 02 | akıl yürütme | Konuşma Ajanı | Transkript: "Ayşe Demir'e 12.500 lira gönder" · dil: TR · güven 0,96 | — | 540 | 180 | | 03 | doğrulama | Kimlik Ajanı | Ses izi eşleşmesi %94 + güvenlik sorusu doğru → kimlik doğrulandı | — | 900 | 260 | | 04 | plan | Sesli İşlem Süpervizörü | Plan: alıcıyı çözümle → limit/dolandırıcılık → OTP → transfer → sesli teyit | — | 280 | 200 | | 05 | araç | Transfer Ajanı | Kayıtlı alıcı "Ayşe Demir" TR12 0001 …8834 bulundu, günlük limit 50.000 TL | Core Banking | 380 | — | | 06 | araç | Transfer Ajanı | Dolandırıcılık motoru: skor 12/100, alıcı 8 aydır kayıtlı | Fraud Engine | 360 | — | | 07 | doğrulama | Transfer Ajanı | Politika: tutar ≤ 50.000 TL ve alıcı kayıtlı → uygun, OTP zorunlu | — | 420 | 300 | | 08 | bildirim | Kimlik Ajanı | OTP SMS gönderildi, müşteri 6 haneli kodu sesle okudu → doğrulandı | SMS Gateway | 1.800 | — | | 09 | yazma | Transfer Ajanı | 12.500 TL transfer çekirdek sistemde gerçekleşti, ref #TR-5583021 | Core Banking | 560 | — | | 10 | bildirim | Diyalog Ajanı | Sesli teyit okundu, özet SMS gönderildi, çağrı kaydı arşivlendi | Genesys SIP | 420 | 160 | Simülasyon metrikleri (tahmini) 27.000çalıştırma / ay %93,1başarı 5,9 sp50 $0,0033tahmini maliyet / çalıştırma Onay kuralı Tutar > 50.000 TL veya yeni alıcı → çağrı merkezi temsilcisine sıcak aktarım Adım profili giriş · 1plan · 1akıl yürütme · 1araç · 2doğrulama · 2yazma · 1bildirim · 2 SİSTEMLER - Genesys SIPvoice - Core Bankingcore - Fraud Engineapi - SMS Gatewaychannel Çalıştırmak için gerekenler - Santral ile SIP entegrasyonu ve kayıt erişimi - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük bankaların IVR yerine konuşan asistan deseninin işlem yapan sürümü Bu senaryonun yer aldığı paketler Diyalog bankacılığı → Sektör: Bankacılık → İlgili ajanlar BankacılıkHITL ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı Müşteri konuşur ya da yazar; fatura, limit, itiraz ve transfer aynı oturumda çekirdek sistemde tamamlanır. 186.000 çalıştırma / ay %94,7 başarı 5 ajan BankacılıkHITL ### Kredi kartı limit artışı Müşteri mobilde yazar ya da çağrı merkezinde konuşur; karar saniyeler içinde verilir, eşik üstü yönetici onayına gider. 18.400 çalıştırma / ay %96,4 başarı 4 ajan BankacılıkHITL ### Dijital müşteri edinimi (KYC) Kimlik, canlılık ve yaptırım kontrolü tek akışta; temiz başvurular şubeye gitmeden dakikalarda hesap açar. 42.000 çalıştırma / ay %93,8 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Sesli hasar ihbarı (FNOL) URL: https://www.aginies.com/tr/ajanlar/voice-claims-intake Language: tr Summary: Müşteri arar ve anlatır; poliçe plakadan bulunur, teminat kontrol edilir, dosya açılır ve eksper atanır. Hasar hattına gelen "aracıma çarptılar" çağrısı → Hasar dosyası açıldı, rezerv ayrıldı, eksper randevusu ve fotoğraf linki müşteride Sigortavoicewhatsapp # Sesli hasar ihbarı (FNOL) Müşteri arar ve anlatır; poliçe plakadan bulunur, teminat kontrol edilir, dosya açılır ve eksper atanır. Kısaca Sesli hasar ihbarı (FNOL), Sigorta sektöründe voice, whatsapp kanalında çalışan bir ajan senaryosudur. Tetikleyici: Hasar hattına gelen "aracıma çarptılar" çağrısı. Hasar Süpervizörü 4 uzman ajanı (Konuşma Ajanı, İhbar Ajanı, Poliçe Ajanı, Hasar Kayıt Ajanı) 4 sistem üzerinde yönetir (Genesys SIP, Policy Admin System, Guidewire ClaimCenter, WhatsApp Business API); 10 adımın 2 tanesi sistemde işlem, 1 tanesi doğrulama veya onaydır. Sonuç: Hasar dosyası açıldı, rezerv ayrıldı, eksper randevusu ve fotoğraf linki müşteride. Onay kuralı: Yaralanma veya rezerv > 250.000 TL → hasar uzmanına sıcak aktarım. Sigorta ### Sesli hasar ihbarı (FNOL) [Sesli hasar ihbarı (FNOL)] çalıştırma #41.288.348süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Yaralanma veya rezerv > 250.000 TL → hasar uzmanına sıcak aktarım Bu çalıştırmadaki ajanlar Hasar Süpervizörüsupervisorplanlar, yönlendirir, onaylar Konuşma Ajanıaginies-voice-tr-v3 · on-premCanlı sesi yazıya döker, gürültüyü filtreler İhbar Ajanıaginies-tulpar-4 · on-premOlay bilgilerini toplar, eksikleri sorar, sesli özet verir Poliçe Ajanıaginies-talay-2 · on-premPoliçeyi bulur, teminatı ve hasar geçmişini kontrol eder Hasar Kayıt Ajanıaginies-sunkar-1 · on-premDosyayı açar, rezerv koyar, eksper atar Denetim izi00:00.00 - ◦Çağrı bağlandı: "dün gece arabama çarptılar" · plaka 34 ABC 123 - ∿Canlı transkript, gürültü filtresi; ton: sakin, güven 0,95 - ⚙Plakayla poliçe bulundu: kasko #KS-2291045, aktif, muafiyet 2.000 TL - ⌘Plan: olay bilgilerini topla → teminat → dosya aç → eksper ata - ∿Olay: park hâlinde çarpma, 07.09 23:30, Kadıköy · yaralı yok · 3. taraf bilinmiyor - ✓Teminat: çarpma kasko kapsamında · 12 ayda 0 hasar · bekleme süresi yok - ⇢Hasar dosyası #CL-2026-118342 açıldı, rezerv 45.000 TL - ⇢En yakın anlaşmalı eksper (Kadıköy) atandı, randevu yarın 10:00 - ➤Sesli özet okundu, müşteri dosya numarasını teyit etti - ➤WhatsApp'a dosya numarası ve fotoğraf yükleme linki gönderildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Hasar hattına gelen "aracıma çarptılar" çağrısı Sonuç Hasar dosyası açıldı, rezerv ayrıldı, eksper randevusu ve fotoğraf linki müşteride ## Ajanlar SÜPERVİZÖR ### Hasar Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-voice-tr-v3 · on-prem ### Konuşma Ajanı Canlı sesi yazıya döker, gürültüyü filtreler transcribe_streamdetect_emotion aginies-tulpar-4 · on-prem ### İhbar Ajanı Olay bilgilerini toplar, eksikleri sorar, sesli özet verir extract_claim_factsask_followupsynthesize_speech aginies-talay-2 · on-prem ### Poliçe Ajanı Poliçeyi bulur, teminatı ve hasar geçmişini kontrol eder get_policy_by_platecheck_coverageget_claim_history aginies-sunkar-1 · on-prem ### Hasar Kayıt Ajanı Dosyayı açar, rezerv koyar, eksper atar create_claimset_reserveassign_adjustersend_whatsapp ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Hasar Süpervizörü | Çağrı bağlandı: "dün gece arabama çarptılar" · plaka 34 ABC 123 | — | 200 | — | | 02 | akıl yürütme | Konuşma Ajanı | Canlı transkript, gürültü filtresi; ton: sakin, güven 0,95 | — | 480 | 170 | | 03 | araç | Poliçe Ajanı | Plakayla poliçe bulundu: kasko #KS-2291045, aktif, muafiyet 2.000 TL | Policy Admin System | 420 | — | | 04 | plan | Hasar Süpervizörü | Plan: olay bilgilerini topla → teminat → dosya aç → eksper ata | — | 280 | 210 | | 05 | akıl yürütme | İhbar Ajanı | Olay: park hâlinde çarpma, 07.09 23:30, Kadıköy · yaralı yok · 3. taraf bilinmiyor | — | 1.100 | 1.240 | | 06 | doğrulama | Poliçe Ajanı | Teminat: çarpma kasko kapsamında · 12 ayda 0 hasar · bekleme süresi yok | — | 560 | 480 | | 07 | yazma | Hasar Kayıt Ajanı | Hasar dosyası #CL-2026-118342 açıldı, rezerv 45.000 TL | Guidewire ClaimCenter | 620 | — | | 08 | yazma | Hasar Kayıt Ajanı | En yakın anlaşmalı eksper (Kadıköy) atandı, randevu yarın 10:00 | Guidewire ClaimCenter | 380 | — | | 09 | bildirim | İhbar Ajanı | Sesli özet okundu, müşteri dosya numarasını teyit etti | Genesys SIP | 380 | 160 | | 10 | bildirim | Hasar Kayıt Ajanı | WhatsApp'a dosya numarası ve fotoğraf yükleme linki gönderildi | WhatsApp Business API | 240 | — | Simülasyon metrikleri (tahmini) 34.000çalıştırma / ay %95,8başarı 4,7 sp50 $0,0026tahmini maliyet / çalıştırma Onay kuralı Yaralanma veya rezerv > 250.000 TL → hasar uzmanına sıcak aktarım Adım profili giriş · 1plan · 1akıl yürütme · 2araç · 1doğrulama · 1yazma · 2bildirim · 2 SİSTEMLER - Genesys SIPvoice - Policy Admin Systemcore - Guidewire ClaimCentercore - WhatsApp Business APIchannel Çalıştırmak için gerekenler - Santral ile SIP entegrasyonu ve kayıt erişimi - Çekirdek sistem için okuma/yazma API veya MCP sunucusu ve servis hesabı - Kanal entegrasyonu (WhatsApp Business API, SMS geçidi veya uygulama içi widget) - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Büyük ABD/AB sigortacılarının çağrı merkezi yükünü azaltan FNOL otomasyonu deseni Bu senaryonun yer aldığı paketler Sesli AI ile hasar ve talep kaydı →Ses analizi ve kalite →Hasar operasyonları uçtan uca →Underwriting ve yenileme operasyonları → Sektör: Sigorta → İlgili ajanlar SigortaHITL ### Uçtan uca kasko hasarı: ihbardan ödemeye Sesli ihbar, fotoğraf ekspertizi, anlaşmalı servis ataması ve ödeme talimatı tek akışta; eksper yalnızca eşik üstünde devreye girer. 21.000 çalıştırma / ay %91,2 başarı 5 ajan SigortaHITL ### Poliçe yenileme teklifi Poliçeler yeniden fiyatlanır, sert artışlar onaya gider; müşteri tek dokunuşla onaylar, poliçe anında kesilir. 22.000 çalıştırma / ay %94,4 başarı 4 ajan SigortaHITL ### Underwriting belge çıkarımı Sağlık raporu ve tahliller okunur, kılavuza göre değerlendirilir; underwriter yalnızca eşik üstünü görür. 4.800 çalıştırma / ay %94,9 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Garanti talebi işleme URL: https://www.aginies.com/tr/ajanlar/warranty-claim-processing Language: tr Summary: Bayi talebi dakikalarda kapsam, parça ve işçilik yönünden doğrulanır; eşik üstü tutar bölge müdürüne gider. Bayi portalından fotoğraflı garanti talebi gönderilmesi → Talep düzeltilerek onaylandı, SAP'de alacak dekontu kesildi, bayi gerekçeyle bilgilendirildi Üretim & Otomotivwebapiemail # Garanti talebi işleme Bayi talebi dakikalarda kapsam, parça ve işçilik yönünden doğrulanır; eşik üstü tutar bölge müdürüne gider. Kısaca Garanti talebi işleme, Üretim & Otomotiv sektöründe web, api, email kanalında çalışan bir ajan senaryosudur. Tetikleyici: Bayi portalından fotoğraflı garanti talebi gönderilmesi. Garanti Süpervizörü 3 uzman ajanı (Talep Ajanı, Doğrulama Ajanı, Ödeme Ajanı) 4 sistem üzerinde yönetir (Dealer Portal, SAP S/4HANA, Warranty Policy Vector Store, Document Store (S3)); 10 adımın 3 tanesi sistemde işlem, 2 tanesi doğrulama veya onaydır. Sonuç: Talep düzeltilerek onaylandı, SAP'de alacak dekontu kesildi, bayi gerekçeyle bilgilendirildi. Onay kuralı: Garanti talebi > 5.000 € → bölge garanti müdürü onayı. Üretim & Otomotiv ### Garanti talebi işleme [Garanti talebi işleme] çalıştırma #41.289.241süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Garanti talebi > 5.000 € → bölge garanti müdürü onayı Bu çalıştırmadaki ajanlar Garanti Süpervizörüsupervisorplanlar, yönlendirir, onaylar Talep Ajanıaginies-konrul-1 · on-premTalep alanlarını ve hasar fotoğraflarını okur Doğrulama Ajanıaginies-tulpar-4 · on-premKapsam, parça fiyatı ve işçilik standardını doğrular Ödeme Ajanıaginies-talay-2 · on-premAlacak dekontunu keser, bayi kaydını günceller Denetim izi00:00.00 - ◦D-114 bayisinden WC-2026-30917 talebi 4 fotoğrafla alındı - ∿VIN WBA…7712 · 31.400 km · parça: turbo aktüatör · işçilik 2,4 sa · 6.850 € - ⚙Kapsam: 3 yıl / 100.000 km, 03/2027'ye kadar geçerli → uygun - ⌕Politika WP-14: aktüatör işçilik standardı 1,8 sa, parça listesi 2026-Q3 - ✓İşçilik 2,4 sa vs standart 1,8 sa (+%33) → düşürüldü; parça fiyatı ve fotoğraflar uyumlu - ⇢Fotoğraflar ve arıza kaydı talep klasörüne arşivlendi - ⏸Düzeltilmiş talep 6.100 € > 5.000 € → bölge garanti müdürü onayladı - ⇢Bayi hesabına 6.100 € için 1700442310 numaralı alacak dekontu kesildi - ⇢Talep durumu "onaylandı – düzeltilmiş", neden kodu L02 (işçilik) - ➤Bayiye portalda düzeltme dökümü iletildi, denetim izi kaydedildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır Tetikleyici Bayi portalından fotoğraflı garanti talebi gönderilmesi Sonuç Talep düzeltilerek onaylandı, SAP'de alacak dekontu kesildi, bayi gerekçeyle bilgilendirildi ## Ajanlar SÜPERVİZÖR ### Garanti Süpervizörü Hedefi adımlara böler, ajanlara dağıtır, onay noktalarını yönetir, sonucu birleştirir. aginies-konrul-1 · on-prem ### Talep Ajanı Talep alanlarını ve hasar fotoğraflarını okur extract_claim_fieldsanalyse_damage_photosarchive_documents aginies-tulpar-4 · on-prem ### Doğrulama Ajanı Kapsam, parça fiyatı ve işçilik standardını doğrular get_vehicle_coverageget_labour_standardcheck_parts_cataloguesearch_warranty_policy aginies-talay-2 · on-prem ### Ödeme Ajanı Alacak dekontunu keser, bayi kaydını günceller post_credit_memoupdate_claim_statussend_dealer_message ## Adımlar | # | Tür | Ajan | Ne olur | Sistem | ms | tok | | --- | --- | --- | --- | --- | --- | --- | | 01 | giriş | Garanti Süpervizörü | D-114 bayisinden WC-2026-30917 talebi 4 fotoğrafla alındı | — | 200 | — | | 02 | akıl yürütme | Talep Ajanı | VIN WBA…7712 · 31.400 km · parça: turbo aktüatör · işçilik 2,4 sa · 6.850 € | — | 860 | 690 | | 03 | araç | Doğrulama Ajanı | Kapsam: 3 yıl / 100.000 km, 03/2027'ye kadar geçerli → uygun | SAP S/4HANA | 420 | — | | 04 | bilgi | Doğrulama Ajanı | Politika WP-14: aktüatör işçilik standardı 1,8 sa, parça listesi 2026-Q3 | Warranty Policy Vector Store | 400 | 520 | | 05 | doğrulama | Doğrulama Ajanı | İşçilik 2,4 sa vs standart 1,8 sa (+%33) → düşürüldü; parça fiyatı ve fotoğraflar uyumlu | — | 820 | 1.040 | | 06 | yazma | Talep Ajanı | Fotoğraflar ve arıza kaydı talep klasörüne arşivlendi | Document Store (S3) | 300 | — | | 07 | insan onayı | Garanti Süpervizörü | Düzeltilmiş talep 6.100 € > 5.000 € → bölge garanti müdürü onayladı | — | 2.800 | — | | 08 | yazma | Ödeme Ajanı | Bayi hesabına 6.100 € için 1700442310 numaralı alacak dekontu kesildi | SAP S/4HANA | 540 | — | | 09 | yazma | Ödeme Ajanı | Talep durumu "onaylandı – düzeltilmiş", neden kodu L02 (işçilik) | Dealer Portal | 380 | — | | 10 | bildirim | Ödeme Ajanı | Bayiye portalda düzeltme dökümü iletildi, denetim izi kaydedildi | Dealer Portal | 220 | — | Simülasyon metrikleri (tahmini) 8.600çalıştırma / ay %96,3başarı 7 sp50 $0,0062tahmini maliyet / çalıştırma Onay kuralı Garanti talebi > 5.000 € → bölge garanti müdürü onayı Adım profili giriş · 1bilgi · 1akıl yürütme · 1araç · 1doğrulama · 1insan onayı · 1yazma · 3bildirim · 1 SİSTEMLER - Dealer Portalapi - SAP S/4HANAerp - Warranty Policy Vector Storedb - Document Store (S3)storage Çalıştırmak için gerekenler - İlgili kurumsal API’ler için kimlik bilgisi ve test ortamı - ERP/MES için kayıt ve sorgu API erişimi - Veritabanı veya vektör deposuna okuma erişimi; veri sahibi - Doküman deposuna okuma erişimi ve belge tipleri - Onay kuralı için yetkili rol ve onay kanalı (mobil onay kartı, e-posta veya bilet) - Kurumun kimlik sağlayıcısıyla SSO ve senaryoya özel yetki grubu Küresel örnek Otomotiv OEM'leri ve ana tedarikçilerdeki bayi garanti talebi otomasyonunu yansıtır Bu senaryonun yer aldığı paketler Üretim ve bakım operasyonları → Sektör: Üretim ve Otomotiv → İlgili ajanlar Üretim & OtomotivHITL ### Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna MES alarmı düşer; kök neden tarihsel veriyle bulunur, yedek parça kontrol edilir, iş emri ve plan güncellemesi dakikalar içinde hazırdır. 3.100 çalıştırma / ay %92,2 başarı 5 ajan Üretim & OtomotivHITL ### Tedarikçi faturası üçlü eşleştirme Fatura, sipariş ve mal kabul SAP'de otomatik eşleşir; yalnızca tolerans dışı kalemler satın almacıya gider. 21.000 çalıştırma / ay %97,1 başarı 4 ajan Üretim & OtomotivHITL ### Bakım talebi triyajı Sensör alarmı ve teknisyen sesli notu tek iş emrine dönüşür; yedek parça ve güvenli duruş penceresi hazırdır. 3.400 çalıştırma / ay %95,2 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Çözümler URL: https://www.aginies.com/tr/cozumler Language: tr Summary: Aginies çözümleri: müşteri etkileşimi, operasyonel otomasyon, saha servis, IoT filo operasyonu, finansal analiz, bilgi yönetimi, KVKK ve ihracat kontrolü uyumu, kamu alımı, İK ve yazılım geliştirme. Her biri aynı orkestrasyon katmanında, aynı yönetişimle çalışır. Çözümler # Yedi alan, 30 çözüm paketi, tek yürütme katmanı. Müşteri etkileşiminden yazılım geliştirmeye kadar her çözüm aynı orkestrasyon katmanı, aynı yönetişim ve aynı gözlemlenebilirlik üzerinde çalışır. Her çözüm, ajan kütüphanesinde canlı izleyebileceğiniz ajanlara bağlıdır. Müşteri etkileşimi ve deneyim Müşteri ve vatandaşla konuşan, konuşurken sistemde işlem yapan ajanlar: web, mobil, WhatsApp, telefon. 10 çözüm · 40 ajan Perakende & E-ticaret4 ajan ## E-ticaret müşteri desteği ve aksiyon Sipariş nerede, iade, teslimat değişikliği: ajan işlemi sipariş sisteminde tamamlar. - Yüksek hacimli talepler destek ekibi büyümeden çözülür - Sipariş güncelleme, iptal ve teslimat değişikliği konuşma içinde tamamlanır - Ürün cevapları kurumun kendi bilgi kaynaklarına bağlı, uydurma yok Devamı → Sigorta4 ajan ## Sesli AI ile hasar ve talep kaydı Çağrı merkezine gelen hasar bildirimini sesli ajan alır, kaydı sistemde açar. - Çağrı merkezi yükü düşer, bekleme süresi kısalır - Aynı anda yüzlerce görüşme, sabit kalite - Hasar kaydı ilk temasta ve tam bilgiyle açılır Devamı → Kurumsal Fonksiyonlar3 ajan ## Çağrı merkezi kalite yönetimi Görüşmelerin %100’ü objektif kriterlerle puanlanır, örnekleme biter. - Örnekleme yerine %100 değerlendirme - Tutarlı, objektif kalite ölçümü - Performans trendleri ve gelişim alanları anında görünür Devamı → Bankacılık4 ajan ## Diyalog bankacılığı Mobil bankacılık işlemlerini yazarak ya da konuşarak gerçekleştir. - Kanal fark etmeksizin konuşarak tamamlanan bankacılık - Çağrı merkezi yükünde azalma, ilk temasta çözüm oranında artış - İşlem başına operasyon maliyetinde düşüş Devamı → Perakende & E-ticaret4 ajan ## Satış ve lead yönetimi Doğru teklif, doğru zamanda: lead skorlama, kampanya ve fiyat önerisi. - Satış ekibi en yüksek dönüşüm ihtimalli lead’lere odaklanır - Kişiye özel kampanya ile dönüşüm ve sepet büyüklüğü artar - Fırsat soğumadan temas: skorlanan her lead aynı gün aranır Devamı → Sağlık4 ajan ## Sağlık operasyonları ve klinik yönlendirme Akıllı triyaj ve hasta yönlendirme; klinik karar her zaman hekimde kalır. - Hasta ilk temasta doğru bölüme ve zamana yönlenir - Gelmeyen hasta oranı düşer, kapasite kullanımı artar - Klinik ekipler idari yükten kurtulur Devamı → Telekom4 ajan ## Ses analizi ve kalite Türkçe STT ile görüşmelerin %100’ü yazıya dökülür, puanlanır; modeller kurum içinde. - Örnekleme yerine tüm görüşmelerde ölçüm - Uyum ihlali ve müşteri riski aynı gün görünür - Ses verisi kurum dışına çıkmaz Devamı → Enerji4 ajan ## Enerji müşteri operasyonları Kesinti, sayaç, bağlantı ve talep yanıtı: şebeke verisiyle konuşan müşteri ajanları. - Kesinti anında çağrı hacmi düşer, müşteri memnuniyeti artar - Fatura itirazları ve bağlantı başvuruları saatler içinde sonuçlanır - Esneklik portföyü gelire dönüşür Devamı → Kamu ve Hükümet4 ajan ## Vatandaş hizmetleri ve dosya ön inceleme Başvuru doğru birime, dosya eksiksiz, karar memurda: mevzuat maddesiyle gerekçeli. - Başvurular ilk seferde doğru birime, süre kaybı olmadan - Dosya ön inceleme süresi saatlerden dakikalara - Her karar açıklanabilir ve denetlenebilir Devamı → Telekom5 ajan ## Abone hizmetleri ve şebeke desteği Fatura, paket, taahhüt ve arıza işlemleri sesli ve yazılı kanalda, BSS ve şebeke sistemlerinde. - Çok konulu çağrılar tek oturumda, temsilcisiz kapanır - Gereksiz teknisyen ziyaretlerinde belirgin azalma - Daha kısa işlem süresi, daha yüksek memnuniyet Devamı → Operasyonel otomasyon Sinyalden iş emrine, talepten teslimata: operasyon sistemlerinin üstünde karar veren ve sevk eden ajanlar. 8 çözüm · 32 ajan Kurumsal Fonksiyonlar4 ajan ## E-posta sınıflandırma ve ek doğrulama Gelen kutusu doğru ekibe, eksik ek daha iş başlamadan tespit edilir. - Talepler ilk seferde doğru ekibe ulaşır - Standart sorulara anında cevap, memnuniyet artışı - Eksik belgeler operasyona girmeden yakalanır Devamı → Kurumsal Fonksiyonlar4 ajan ## Operasyonel otomasyon ve talep yönetimi Talepleri sınıflandıran, yönlendiren ve cevaplayan ajanlar; hızlı süreç, düşük maliyet. - Talep çözüm süresi kısalır, birikim erir - Talep başına operasyon maliyetinde düşüş - Ekipler tekrar işten kurtulur, istisnaya odaklanır Devamı → IoT ve Bağlı Cihazlar4 ajan ## Saha servis ve sevk otomasyonu Sensör alarmı ya da müşteri bildirimi → doğru teknisyen, doğru parça, doğru zaman. - Sinyalden servis emrine geçiş dakikalar içinde, elle giriş olmadan - Doğru parça ve doğru görev kartı ile ilk ziyarette çözüm - Müşteri ve kalite kaydı aynı anda güncel Devamı → Bankacılık4 ajan ## Dolandırıcılık ve itiraz operasyonları Alarm triyajı, ters ibraz dosyaları, kart bloke ve erken tahsilat tek operasyon masasında. - Alarm başına inceleme süresinde büyük düşüş - Süresi geçen itiraz dosyası kalmaz - Daha az yanlış pozitif bloke, daha az müşteri şikâyeti Devamı → Sigorta4 ajan ## Hasar operasyonları uçtan uca İhbardan ödemeye: sesli kayıt, fotoğraf ekspertizi, servis ataması ve rücu tek akışta. - Basit hasarlarda dosya kapanışı günlere iner - Eksper kapasitesi karmaşık dosyalara ayrılır - Daha fazla rücu, daha az sızıntı Devamı → Finansal Teknolojiler4 ajan ## Üye işyeri ve ödeme operasyonları İşyeri edinimi, POS ve hakediş desteği, mutabakat istisnaları ve işlem izleme tek masada. - İşyeri sorunları ilk temasta, tek kanalda çözülür - KYB saatler içinde; risk ekibi yalnızca istisnalarda - Mutabakat istisnaları ay sonunu beklemez Devamı → Lojistik4 ajan ## Lojistik kontrol kulesi Sapmayı erken yakalayan, rotayı değiştiren, müşteriyi bilgilendiren ve SLA’yı hesaplayan operasyon ajanları. - Gecikmeler saatler önce bilinir ve yönetilir - Zamanında teslimat ve müşteri bilgilendirmesinde artış - SLA finansı şeffaf ve denetlenebilir Devamı → Üretim & Otomotiv4 ajan ## Üretim ve bakım operasyonları Hat duruşundan iş emrine, bakım talebinden garanti dosyasına: MES, CMMS ve ERP üzerinde karar veren ajanlar. - Plansız duruş süresinde ölçülebilir azalma - Alarmdan iş emrine dakikalar - Garanti ve fatura süreçlerinde haftalar yerine günler Devamı → Finansal analiz ve operasyonel zekâ Belgeden kayda, kayıttan rapora: finans ve operasyon verisini doğrulayan, eşleştiren, raporlayan ajanlar. 4 çözüm · 15 ajan Kurumsal Fonksiyonlar4 ajan ## Fatura ve doküman veri çıkarımı OCR → doğrulama → ERP: elle veri girişi olmadan temiz kayıt. - Anahtar alanlar elle girilmeden çıkarılır - Finans süreçleri günlerden dakikalara iner - Doğrulanmış veri doğrudan ERP ve iş sistemlerine akar Devamı → Finansal Teknolojiler4 ajan ## Finansal analiz ve operasyonel zekâ Finansal veriyi ve belgeyi okuyan, yorumlayan, karar destekleyen ajanlar. - Mutabakat ve fatura döngüsü günlerden saatlere iner - Kararlar güncel ve kaynağı belli veriyle alınır - İstisna ve risk desenleri gerçekleşmeden görünür Devamı → Bankacılık4 ajan ## Kredi ve müşteri edinimi operasyonları Dijital hesap açılışı, limit artışı ve kredi belgeleri tek karar mantığıyla, şubeye gitmeden. - Şubeye gitmeden açılan hesap ve dakikalar içinde kredi kararı - Belge inceleme iş yükünde belirgin azalma - Denetime hazır, gerekçeli ve tutarlı kararlar Devamı → Sigorta3 ajan ## Underwriting ve yenileme operasyonları Teklif belgeleri okunur, risk değişkenleri çıkarılır, yenileme teklifi kurallar içinde hazırlanır. - Yenilemelerin büyük kısmı otomatik, zamanında teklif - Underwriter zamanı istisnai risklere - Yenileme oranında ölçülebilir artış Devamı → Bilgi yönetimi ve karar desteği Kurumun kendi bilgisine bağlı, kaynak gösteren cevaplar ve karar desteği; uydurma yok. 1 çözüm · 4 ajan Kamu ve Hükümet4 ajan ## Bilgi yönetimi ve karar desteği Mevzuat, politika ve sözleşmelere saniyeler içinde kaynaklı erişim. - Bilgiye erişim saatlerden saniyelere iner - Kararlar güncel mevzuat ve kaynağı belli belgeye dayanır - Mevzuat değişikliğinin etkisi proaktif görülür Devamı → Veri koruma ve uyum Kişisel veri, ihracat kontrolü ve kamu alımı: kuralı uygulayan, kararı uzmanda bırakan ajanlar. 3 çözüm · 11 ajan Kurumsal Fonksiyonlar4 ajan ## KVKK kişisel veri koruma Belge ve görüşmelerde kişisel veri tespiti, maskeleme ve denetime hazır veri haritası. - Kişisel veri modele ulaşmadan maskelenir - Denetime hazır, güncel veri envanteri - Belge ve çağrı arşivindeki uyum riski görünür hâle gelir Devamı → Kamu ve Hükümet3 ajan ## Kamu alımı ve ihale uyumu Teklif dosyaları şartnameyle karşılaştırılır; komisyon hazır bulgu listesi üzerinde karar verir. - Komisyon hazırlığı günlerden saatlere iner - Tutarlı, gerekçeli ve itiraza dayanıklı değerlendirme - Her bulgu kaynak belgeye bağlı, denetime hazır Devamı → Savunma Sanayii4 ajan ## Savunma ihracat kontrolü ve tedarik zinciri güvenliği Sınıflandırma, taraf tarama, lisans dosyası ve tedarikçi nitelendirme: kapalı ağda, kararı uzmanda bırakan ajanlar. - Uyum uzmanı dosya başına saatler yerine dakikalar harcar - Kontrole tabi kalem sevkiyat öncesi yakalanır - Tedarikçi riski nitelendirme öncesi görünür Devamı → İK ve çalışan deneyimi İşe alımdan onboarding’e çalışan deneyimi; İK ekibi zamanını insanlara ayırır. 2 çözüm · 7 ajan Kurumsal Fonksiyonlar3 ajan ## İK ve çalışan deneyimi İşe alımdan uyuma: veriye dayalı, hızlı ve ölçeklenebilir İK süreçleri. - Değerlendirmeden uyuma İK döngüsü kısalır - Yeni çalışan ilk gün hazır: erişim, donanım, eğitim - İK ekibi tekrar işten kurtulur, deneyime odaklanır Devamı → Kurumsal Fonksiyonlar4 ajan ## Ajan kiralama (Agent Outsourcing) Dış kaynak insan yerine dış kaynak ajan: iOS geliştirici, iş analisti, muhasebeci, testçi; iş kaydıyla. - Günler içinde işe başlayan, 7/24 çalışan roller - Ne yapıldığı, ne kadar sürdüğü ve kaça mal olduğu tek kayıtta - Bilgi workspace’te kalır; ölçekleme dakikalar Devamı → Yazılım geliştirme ve teknoloji verimliliği SDLC ajan takımı, IoT filo operasyonu, Code Aginies: yazılım ve cihaz hattının tamamı. 2 çözüm · 9 ajan Yazılım Geliştirme5 ajan ## SDLC ajan ekibi PM, analist, mimar, geliştirici ve test ajanları CI/CD hattında birlikte çalışır. - Talepten üretime geçiş süresi günlerden saatlere iner - Standart, ölçülebilir kalite: kapsam, güvenlik ve regresyon her PR’da - Mühendisler yönlendirme ve onaya odaklanır, tekrar işi ajanlar üstlenir Devamı → IoT ve Bağlı Cihazlar4 ajan ## IoT cihaz filosu operasyonu Anomali kümeleme, kök neden ve sağlık kapılı OTA dağıtımı: filo tek olay, tek karar. - Alarm gürültüsü olay sayısına iner; ekip gerçek soruna bakar - Güvenli, kanıtlı ve geri alınabilir firmware dağıtımı - Küçük on-prem modellerle düşük birim maliyet Devamı → ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Ajan kiralama (Agent Outsourcing) URL: https://www.aginies.com/tr/cozumler/agent-workforce Language: tr Summary: Dış kaynak modeli kişi-ay satar: işe başlama haftalar sürer, bilgi kişiyle gider, zaman çizelgesi ne yapıldığını göstermez. Ölçekleme yeni işe alım demektir. Kurumsal Fonksiyonlar · İK ve çalışan deneyimi # Ajan kiralama (Agent Outsourcing) Dış kaynak insan yerine dış kaynak ajan: iOS geliştirici, iş analisti, muhasebeci, testçi; iş kaydıyla. Kısaca Ajan kiralama (Agent Outsourcing), Kurumsal Fonksiyonlar sektörü için “ik ve çalışan deneyimi” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: İş tanımı ajan spesifikasyonuna çevrilir: görevler, araçlar, çıktılar, KPI’lar, onay noktaları; ajan sizin araçlarınızda hesap alır (Jira, Git, ERP, CRM); kod ve veri sizin tenant’ınızda kalır. Beklenen sonuçlar: günler içinde işe başlayan, 7/24 çalışan roller; ne yapıldığı, ne kadar sürdüğü ve kaça mal olduğu tek kayıtta; bilgi workspace’te kalır; ölçekleme dakikalar. Paket, Aginies platformunun Role specs, Work log, Shadow → sampled review, Human-in-the-loop yeteneklerini kullanır ve ajan kütüphanesindeki iOS özellik geliştirme (kiralık geliştirici ajanı), Regresyon test döngüsü (kiralık test ajanı), Ay sonu kapanış ve mutabakat (kiralık muhasebe ajanı), Backlog düzenleme ve tahminleme senaryolarıyla canlı izlenebilir. Sorun Dış kaynak modeli kişi-ay satar: işe başlama haftalar sürer, bilgi kişiyle gider, zaman çizelgesi ne yapıldığını göstermez. Ölçekleme yeni işe alım demektir. Ajanlar ne yapar - 01İş tanımı ajan spesifikasyonuna çevrilir: görevler, araçlar, çıktılar, KPI’lar, onay noktaları. - 02Ajan sizin araçlarınızda hesap alır (Jira, Git, ERP, CRM); kod ve veri sizin tenant’ınızda kalır. - 03Gölge dönemde her çıktı incelenir; kalite eşiği geçilince örneklemeli incelemeye ve eşik bazlı onaya geçilir. - 04Her görev iş kaydına yazılır: çıktı, saat karşılığı, kalite, onay, maliyet; haftalık zaman çizelgesi ve aylık fatura aynı kayıttan. Paket kapsamı - Rol kataloğundan ajan kiralama: geliştirici, analist, muhasebeci, testçi, destek - Kurumun araçlarında çalışır; iş kaydı görev, çıktı ve saat karşılığıyla - Gölge dönem, örneklemeli inceleme, otonom; eşikler ve kill-switch - Aginies tarafında adı belli ekip lideri; haftalık rapor Başlamak için gerekenler - Rol için gerekli araç erişimleri (Jira, Git, ERP, bilet sistemi) - Görev tanımı ve kabul kriterleri - Kurum tarafında inceleme sorumlusu Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler iOS Geliştirici SüpervizörüTest SüpervizörüMuhasebe SüpervizörüPlanlama Süpervizörü Uzman ajanlar Analiz AjanıGeliştirme Ajanıİnceleme AjanıTest Plan AjanıKoşum AjanıHata AjanıBelge AjanıMutabakat AjanıKapanış AjanıProje Yöneticisi AjanıAnalist AjanıMimar Ajanı Modeller (Hub kataloğu) aginies-sunkar-1aginies-talay-2aginies-tulpar-4aginies-konrul-1 Dokunulan sistemler JiraGitHubXcode CloudFigmaWorkforce Work LogCI/CDDevice / Browser GridJira / XrayERP (SAP / Logo)Bank APIe-Invoice PortalConfluenceSlack / Teams Kanallar apiappbatchemailweb Simülasyon hacimleri 1.040çalıştırma / ay %94,3başarı oranı 9,2 smedyan süre $0,0191tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Günler içinde işe başlayan, 7/24 çalışan roller - Ne yapıldığı, ne kadar sürdüğü ve kaça mal olduğu tek kayıtta - Bilgi workspace’te kalır; ölçekleme dakikalar Kullanılan yetenekler Role specsWork logShadow → sampled reviewHuman-in-the-loopKill-switchAgent-month / output billing Platformu keşfet → Ölçülen hedefler | Kabul edilen çıktı payı | %85 ve üzeri | | Rol başına maliyet | Dış kaynak kişi-ayın %30–50’si | | Bilgi kaybı (devirde) | Sıfır; bilgi workspace’te kalır | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - iOS özellik geliştirme (kiralık geliştirici ajanı)Her PR bir insan tarafından incelenir; üretim dağıtımı ekip liderinin onayında - Regresyon test döngüsü (kiralık test ajanı)Sürüm “go / no-go” kararı sürüm yöneticisinde - Ay sonu kapanış ve mutabakat (kiralık muhasebe ajanı)Deftere işlenen her kayıt ve beyanname mali müşavir onayıyla - Backlog düzenleme ve tahminlemeSprint kapsamı → ürün sahibi onayı; ajan sprint'e talep atayamaz, yalnızca önerir - paketÜretim ve finansal kayıt değişiklikleri kurum onayında - paketÇıktıların mülkiyeti kurumda; ajan-ay veya çıktı bazlı faturalama Yol haritası - 1 · Politika ve rol kataloğu · 2 haftaİK politikaları, süreç sahipleri ve sistem erişimleri tanımlanır; kişisel veri sınırları yazılır. - 2 · Pilot · 4 haftaBir departman ya da bir rolle; kalite puanı ve kullanıcı geri bildirimi haftalık izlenir. - 3 · YaygınlaştırmaTüm departmanlar; eşikler, inceleme örneklemi ve kill-switch İK yönetiminde. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Yazılım GeliştirmeHITL ### iOS özellik geliştirme (kiralık geliştirici ajanı) Jira hikâyesi alınır, SwiftUI ile geliştirilir, testler yazılır, PR açılır; inceleme ve dağıtım kararı ekipte kalır. 380 çalıştırma / ay %94,4 başarı 4 ajan Yazılım GeliştirmeHITL ### Regresyon test döngüsü (kiralık test ajanı) Her sürüm adayında regresyon koşulur, yeni hatalar kanıtla açılır, kapsama raporu yazılır; “go” kararı sürüm yöneticisinde. 260 çalıştırma / ay %95,8 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Ay sonu kapanış ve mutabakat (kiralık muhasebe ajanı) Faturalar işlenir, banka ve tedarikçi mutabakatı yapılır, kapanış listesi yürütülür; kayıtlar mali müşavir onayıyla deftere işlenir. 60 çalıştırma / ay %96,2 başarı 4 ajan Yazılım GeliştirmeHITL ### Backlog düzenleme ve tahminleme Talepler triyajlanır, netleştirilir, karmaşıklığı tahminlenir; sprint önerisi ürün sahibinin onayına gider. 340 çalıştırma / ay %92,8 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Çağrı merkezi kalite yönetimi URL: https://www.aginies.com/tr/cozumler/call-center-quality Language: tr Summary: Kalite ekipleri görüşmelerin %2’sini dinleyebiliyor. Puanlama öznel, koçluk gecikmeli, uyum ihlalleri haftalar sonra fark ediliyor. Kurumsal Fonksiyonlar · Müşteri etkileşimi ve deneyim # Çağrı merkezi kalite yönetimi Görüşmelerin %100’ü objektif kriterlerle puanlanır, örnekleme biter. Kısaca Çağrı merkezi kalite yönetimi, Kurumsal Fonksiyonlar sektörü için “müşteri etkileşimi ve deneyim” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Tüm görüşmeler yazıya dökülür, hassas veriler maskelenir; her görüşme kurumun kalite formuna göre puanlanır; gerekçe ve zaman damgası kaydedilir. Beklenen sonuçlar: örnekleme yerine %100 değerlendirme; tutarlı, objektif kalite ölçümü; performans trendleri ve gelişim alanları anında görünür. Paket, Aginies platformunun Speech-to-text, Agent decision traceability, Dashboard analytics layer, Latency tracking yeteneklerini kullanır ve ajan kütüphanesindeki Çağrı merkezi kalite izleme, Yorum yanıtlama ve kalite eskalasyonu, Fatura itirazı çözümü senaryolarıyla canlı izlenebilir. Sorun Kalite ekipleri görüşmelerin %2’sini dinleyebiliyor. Puanlama öznel, koçluk gecikmeli, uyum ihlalleri haftalar sonra fark ediliyor. Ajanlar ne yapar - 01Tüm görüşmeler yazıya dökülür, hassas veriler maskelenir. - 02Her görüşme kurumun kalite formuna göre puanlanır; gerekçe ve zaman damgası kaydedilir. - 03Uyum ihlali, tekrar eden şikâyet ve süreç darboğazı otomatik işaretlenir. - 04Temsilci ve ekip panolarında trendler; koçluk önerileri. Paket kapsamı - Tüm çağrıların transkripti ve kalite formuna göre otomatik puanlanması - Uyum ihlali, kaba dil ve zorunlu ifade kontrolleri - Temsilci bazında koçluk notu ve eğilim panosu - Örneklem yerine tam kapsam; kalite ekibi istisnalara odaklanır Başlamak için gerekenler - Çağrı kayıtlarına ve meta veriye erişim (santral veya kayıt sistemi) - Mevcut kalite formu ve puanlama kuralları - Temsilci–çağrı eşlemesi için CRM veya santral verisi Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Kalite SüpervizörüMüşteri Sesi SüpervizörüFatura Süpervizörü Uzman ajanlar Transkript AjanıPuanlama AjanıKoçluk AjanıSınıflandırma AjanıYanıt AjanıKalite AjanıNiyet AjanıFatura Analiz AjanıDüzeltme Ajanı Modeller (Hub kataloğu) aginies-voice-tr-v3aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Dokunulan sistemler Genesys SIPQM Scorecard / Case ManagementQA Rubric Vector StoreWorkdayReview Platform APIShopify / OMSJira (QA)Billing (Amdocs)Usage Mediation / CDR StoreSMS Gateway Kanallar voicebatchwebapiappwhatsapp Simülasyon hacimleri 204.000çalıştırma / ay %97,5başarı oranı 6 smedyan süre $0,0012tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Örnekleme yerine %100 değerlendirme - Tutarlı, objektif kalite ölçümü - Performans trendleri ve gelişim alanları anında görünür Kullanılan yetenekler Speech-to-textAgent decision traceabilityDashboard analytics layerLatency trackingSensitive data maskingRouting accuracy measurement Platformu keşfet → Ölçülen hedefler | Değerlendirilen çağrı payı | %3–5 örneklemden %100 kapsama | | Kalite değerlendirmesi iş yükü | %80 azalma | | Uyum ihlali tespit süresi | Haftalardan aynı güne | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Çağrı merkezi kalite izlemePuan < 60 veya uyum ihlali → takım lideri incelemesi - Yorum yanıtlama ve kalite eskalasyonu1 yıldızlı yorum yanıtları ve kusur eskalasyonları → topluluk yöneticisi onayı - Fatura itirazı çözümüİyi niyet kredisi > 75 TL → ekip lideri onayı - paketPuanlar temsilciyle paylaşılmadan önce kalite uzmanı örneklemle doğrular - paketKayıtlar Türkiye’de işlenir; kişisel veri transkriptte maskelenir Yol haritası - 1 · Keşif ve tasarım · 2–3 haftaEn sık 20 niyet ve işlem seçilir, kimlik ve onay eşikleri bilgi güvenliği ile yazılır, kanal ve sistem entegrasyonları listelenir. - 2 · Pilot · 4–6 haftaTek kanalda, sınırlı müşteri grubuyla; test korpusu koşturulur, gölge modda insan temsilci izler, eşikler gerçek trafikle ayarlanır. - 3 · YaygınlaştırmaDiğer kanallar ve işlemler eklenir; panolar, SIEM aktarımı ve kill-switch devredilir; aylık model ve kural gözden geçirmesi başlar. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Kurumsal FonksiyonlarHITL ### Çağrı merkezi kalite izleme Örneklem değil, tüm çağrılar rubriğe göre puanlanır; uyum ihlali ve koçluk fırsatı sabah liderin önünde. 90.000 çalıştırma / ay %98,8 başarı 4 ajan Perakende & E-ticaretHITL ### Yorum yanıtlama ve kalite eskalasyonu Yorumlar sınıflanır, marka tonunda yanıtlanır; tekrarlayan kusurlar iade verisiyle QA'e Jira kaydı olur. 26.000 çalıştırma / ay %97,2 başarı 4 ajan TelekomHITL ### Fatura itirazı çözümü Fatura kalemleri ve kullanım kayıtları saniyede açıklanır; iyi niyet kredisi politikaya göre uygulanır. 88.000 çalıştırma / ay %96,3 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Vatandaş hizmetleri ve dosya ön inceleme URL: https://www.aginies.com/tr/cozumler/citizen-services Language: tr Summary: Dilekçeler yanlış birime gidiyor, süreler işliyor, ruhsat ve yardım dosyalarında eksik belge kontrolü memur zamanının çoğunu alıyor. Kara kutu araçlar kamuda kullanılamıyor. Kamu ve Hükümet · Müşteri etkileşimi ve deneyim # Vatandaş hizmetleri ve dosya ön inceleme Başvuru doğru birime, dosya eksiksiz, karar memurda: mevzuat maddesiyle gerekçeli. Kısaca Vatandaş hizmetleri ve dosya ön inceleme, Kamu ve Hükümet sektörü için “müşteri etkileşimi ve deneyim” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Başvuru sınıflandırılır, mevzuattaki süre ve sorumlu birim bulunur, dosya açılır; belgeler okunur, kurum kayıtlarıyla doğrulanır, eksikler sade dille vatandaşa bildirilir. Beklenen sonuçlar: başvurular ilk seferde doğru birime, süre kaybı olmadan; dosya ön inceleme süresi saatlerden dakikalara; her karar açıklanabilir ve denetlenebilir. Paket, Aginies platformunun Intent detection, Vision extraction, Multi-source RAG, Explainable scoring yeteneklerini kullanır ve ajan kütüphanesindeki Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi, Vatandaş başvurusu yönlendirme, Ruhsat başvurusu ön inceleme, Sosyal yardım başvurusu ön değerlendirme senaryolarıyla canlı izlenebilir. Sorun Dilekçeler yanlış birime gidiyor, süreler işliyor, ruhsat ve yardım dosyalarında eksik belge kontrolü memur zamanının çoğunu alıyor. Kara kutu araçlar kamuda kullanılamıyor. Ajanlar ne yapar - 01Başvuru sınıflandırılır, mevzuattaki süre ve sorumlu birim bulunur, dosya açılır. - 02Belgeler okunur, kurum kayıtlarıyla doğrulanır, eksikler sade dille vatandaşa bildirilir. - 03Uygunluk kriterleri puanlanır; her bulgu madde numarası ve kaynak sayfayla gerekçelendirilir. - 04İdari karar yetkili memurda kalır; hassas alanlar ekranda maskelenir, iz kaydı tutulur. Paket kapsamı - Başvuru alma, belge okuma ve eksik tamamlama; e-Devlet entegrasyonu - Uygunluk değerlendirmesi ve gerekçe; memura karar paketi - Randevu, durum takibi ve bildirim; WhatsApp ve çağrı merkezi - Dilekçe yönlendirme ve süre takibi Başlamak için gerekenler - Belge yönetimi ve dosya sistemine API erişimi - e-Devlet servisleri için yetki ve entegrasyon - Hizmet bazında kriter, belge ve süre tanımları Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Vatandaş Hizmetleri SüpervizörüBaşvuru SüpervizörüRuhsat SüpervizörüSosyal Yardım Süpervizörü Uzman ajanlar Diyalog AjanıUygunluk AjanıBelge AjanıDosya AjanıSınıflandırma AjanıSLA AjanıMevzuat AjanıDoğrulama AjanıVatandaş İletişim Ajanı Modeller (Hub kataloğu) aginies-talay-2aginies-tulpar-4aginies-konrul-1aginies-sunkar-1 Dokunulan sistemler WhatsApp Businesse-Government GatewayCase ManagementDocument StoreAppointment Systeme-Devlet GatewayLegislation Vector StoreSMS GatewayGIS / Zoning PlanRegulation Vector StoreExchange / M365 MailSocial Assistance Information SystemCivil Registry / Income ServicesEligibility Criteria (vector) Kanallar whatsappvoicewebappemailbatch Simülasyon hacimleri 308.100çalıştırma / ay %94,8başarı oranı 7 smedyan süre $0,002tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Başvurular ilk seferde doğru birime, süre kaybı olmadan - Dosya ön inceleme süresi saatlerden dakikalara - Her karar açıklanabilir ve denetlenebilir Kullanılan yetenekler Intent detectionVision extractionMulti-source RAGExplainable scoringPII maskingAir-gap deployment Platformu keşfet → Ölçülen hedefler | Vatandaştan istenen tekrar belge | %70 azalma | | Başvuru ön inceleme süresi | Günlerden dakikalara | | Doğru birime ilk yönlendirme | %95 ve üzeri | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibiUygunluk sınırda, belge tutarsız veya aynı hanede başka başvuru → memur incelemesi - Vatandaş başvurusu yönlendirmeSınıflandırma güveni < 0,85 veya personel şikâyeti → evrak kayıt memuru onayı - Ruhsat başvurusu ön incelemeHer ruhsat kararı ruhsat memuru tarafından verilir; sistem yalnızca ön inceleme yapar - Sosyal yardım başvurusu ön değerlendirmeYardım kararı her zaman sosyal inceleme görevlisinde; ajan ön değerlendirme yapar - paketKarar her zaman memurda; ajan uygunluk paketini hazırlar - paketVeri kamu bulutu veya kurum veri merkezinde; erişim yetki bazlı Yol haritası - 1 · Keşif ve tasarım · 2–3 haftaEn sık 20 niyet ve işlem seçilir, kimlik ve onay eşikleri bilgi güvenliği ile yazılır, kanal ve sistem entegrasyonları listelenir. - 2 · Pilot · 4–6 haftaTek kanalda, sınırlı müşteri grubuyla; test korpusu koşturulur, gölge modda insan temsilci izler, eşikler gerçek trafikle ayarlanır. - 3 · YaygınlaştırmaDiğer kanallar ve işlemler eklenir; panolar, SIEM aktarımı ve kill-switch devredilir; aylık model ve kural gözden geçirmesi başlar. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Kamu ve HükümetHITL ### Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi Vatandaş yazar ya da arar; uygunluk kontrol edilir, belgeler okunur, başvuru açılır, randevu verilir, her adım izlenebilir. 230.000 çalıştırma / ay %93,8 başarı 5 ajan Kamu ve HükümetHITL ### Vatandaş başvurusu yönlendirme e-Devlet başvurusu saniyeler içinde sınıflandırılır, yasal süresi hesaplanır ve doğru birime SLA ile atanır. 54.000 çalıştırma / ay %98,1 başarı 4 ajan Kamu ve HükümetHITL ### Ruhsat başvurusu ön inceleme Eksik belge ve mevzuat uyumu dakikalar içinde raporlanır; kararı her zaman ruhsat memuru verir. 3.100 çalıştırma / ay %95,8 başarı 4 ajan Kamu ve HükümetHITL ### Sosyal yardım başvurusu ön değerlendirme Başvuru kurum kayıtlarıyla doğrulanır, uygunluk kriterleri puanlanır, eksikler vatandaşa bildirilir; karar sosyal inceleme görevlisinin. 21.000 çalıştırma / ay %96,6 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Hasar operasyonları uçtan uca URL: https://www.aginies.com/tr/cozumler/claims-operations-end-to-end Language: tr Summary: Hasar dosyası beş ekip ve dört sistem arasında dolaşıyor; müşteri her adımda arıyor, eksper basit dosyalara da gidiyor, rücu fırsatları kaçıyor. Sigorta · Operasyonel otomasyon # Hasar operasyonları uçtan uca İhbardan ödemeye: sesli kayıt, fotoğraf ekspertizi, servis ataması ve rücu tek akışta. Kısaca Hasar operasyonları uçtan uca, Sigorta sektörü için “operasyonel otomasyon” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: İhbar sesli veya uygulamadan alınır; poliçe eşlenir, olay formu tek görüşmede tamamlanır; fotoğraflardan parça ve işçilik tahmini; önceki hasar ve meta veri kontrolü ile sahtecilik taraması. Beklenen sonuçlar: basit hasarlarda dosya kapanışı günlere iner; eksper kapasitesi karmaşık dosyalara ayrılır; daha fazla rücu, daha az sızıntı. Paket, Aginies platformunun SIP voice, Vision damage assessment, Policy & claims tools (MCP), Human-in-the-loop yeteneklerini kullanır ve ajan kütüphanesindeki Uçtan uca kasko hasarı: ihbardan ödemeye, Sesli hasar ihbarı (FNOL), Fotoğrafla kasko hasar tespiti, Rücu ve tahsilat senaryolarıyla canlı izlenebilir. Sorun Hasar dosyası beş ekip ve dört sistem arasında dolaşıyor; müşteri her adımda arıyor, eksper basit dosyalara da gidiyor, rücu fırsatları kaçıyor. Ajanlar ne yapar - 01İhbar sesli veya uygulamadan alınır; poliçe eşlenir, olay formu tek görüşmede tamamlanır. - 02Fotoğraflardan parça ve işçilik tahmini; önceki hasar ve meta veri kontrolü ile sahtecilik taraması. - 03Teminat, muafiyet ve kusur değerlendirilir; anlaşmalı servis atanır, ödeme talimatı hazırlanır. - 04Rücu dosyası otomatik açılır; eşik üstü dosyalar eksper onay kartına düşer. Paket kapsamı - Sesli ve uygulama üzerinden ihbar; poliçe eşleme ve olay formu - Fotoğraf ekspertizi, teminat ve kusur değerlendirmesi - Anlaşmalı servis ataması, ödeme talimatı ve rücu dosyası - Müşteri bilgilendirme ve dosya takibi; eksper onay kartı Başlamak için gerekenler - Poliçe, hasar ve ödeme sistemleri API erişimi - Anlaşmalı servis ağı ve parça fiyat listesi - Hasar türü bazında eşik ve eksper kuralları Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Hasar SüpervizörüRücu Süpervizörü Uzman ajanlar İhbar AjanıFoto Ekspertiz AjanıDosya Ajanıİletişim AjanıKonuşma AjanıPoliçe AjanıHasar Kayıt AjanıGörüntü AjanıEkspertiz AjanıServis AjanıTarama AjanıKanıt AjanıTahsilat Ajanı Modeller (Hub kataloğu) aginies-voice-tr-v3aginies-konrul-1aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Dokunulan sistemler Call Centre SIPPolicy AdminClaims SystemRepair NetworkPayment HubGenesys SIPPolicy Admin SystemGuidewire ClaimCenterWhatsApp Business APIParts & Labour Catalogue (Audatex)Repair Network PortalDocument Store (S3)Insurance Registry (SBM)Email (Exchange) Kanallar voiceappwhatsappbatchemailapi Simülasyon hacimleri 73.900çalıştırma / ay %94,0başarı oranı 8,1 smedyan süre $0,0045tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Basit hasarlarda dosya kapanışı günlere iner - Eksper kapasitesi karmaşık dosyalara ayrılır - Daha fazla rücu, daha az sızıntı Kullanılan yetenekler SIP voiceVision damage assessmentPolicy & claims tools (MCP)Human-in-the-loopFull audit trailPayment hub integration Platformu keşfet → Ölçülen hedefler | Dosya kapanış süresi (basit hasar) | Haftalardan günlere | | Eksper gerektiren dosya payı | %40 azalma | | Rücu başlatılan dosya | +25% | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Uçtan uca kasko hasarı: ihbardan ödemeyeTahmin > 150.000 TL, kusur oranı belirsiz veya önceki hasarla çakışma → eksper onayı - Sesli hasar ihbarı (FNOL)Yaralanma veya rezerv > 250.000 TL → hasar uzmanına sıcak aktarım - Fotoğrafla kasko hasar tespitiTahmin > 15.000 TL veya model güveni < %80 → eksper onayı - Rücu ve tahsilatRücu talebi > 50.000 TL → rücu avukatı onayı - paketEşik üstü tutar, belirsiz kusur ve önceki hasarla çakışma eksper onayında - paketFotoğraf meta verisi ve karar gerekçesi dosyada; sahtecilik kontrolü her dosyada Yol haritası - 1 · Süreç haritası · 2 haftaSinyal kaynakları, karar noktaları ve iş emri sistemleri haritalanır; otomatik karar sınırı ve onay zinciri belirlenir. - 2 · Pilot · 4 haftaTek hat, tek bölge ya da tek süreçte; ajan öneriyi hazırlar, insan uygular; sapma ve kazanç ölçülür. - 3 · Otonom işletimEşik altı kararlar otomatik, üstü onaya; vardiya raporları ve SLA panoları devrededir. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları SigortaHITL ### Uçtan uca kasko hasarı: ihbardan ödemeye Sesli ihbar, fotoğraf ekspertizi, anlaşmalı servis ataması ve ödeme talimatı tek akışta; eksper yalnızca eşik üstünde devreye girer. 21.000 çalıştırma / ay %91,2 başarı 5 ajan SigortaHITL ### Sesli hasar ihbarı (FNOL) Müşteri arar ve anlatır; poliçe plakadan bulunur, teminat kontrol edilir, dosya açılır ve eksper atanır. 34.000 çalıştırma / ay %95,8 başarı 5 ajan SigortaHITL ### Fotoğrafla kasko hasar tespiti Fotoğraflardan hasar tespit edilir, parça/işçilikle tahmin çıkar, eksper onaylar; servis randevusu alınır. 15.500 çalıştırma / ay %93,3 başarı 4 ajan SigortaHITL ### Rücu ve tahsilat Kapanmış dosyalar rücu için taranır, kusurlu tarafın sigortacısı bulunur, talep avukat onayıyla gönderilir. 3.400 çalıştırma / ay %95,5 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Diyalog bankacılığı URL: https://www.aginies.com/tr/cozumler/conversational-banking Language: tr Summary: Müşteriler menülerde kayboluyor, çağrı merkezi basit işlemlerle doluyor. Mevcut asistanlar bilgi veriyor ama işlem yapamıyor; yapabilenler de denetlenemiyor. Bankacılık · Müşteri etkileşimi ve deneyim # Diyalog bankacılığı Mobil bankacılık işlemlerini yazarak ya da konuşarak gerçekleştir. Kısaca Diyalog bankacılığı, Bankacılık sektörü için “müşteri etkileşimi ve deneyim” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Mevcut kimlik doğrulama akışına entegre; yetkisiz işlem talebi guardrail’de durur; limit artışı, para transferi, otomatik ödeme talimatı ve ekstre tek diyalogda tamamlanır. Beklenen sonuçlar: kanal fark etmeksizin konuşarak tamamlanan bankacılık; çağrı merkezi yükünde azalma, ilk temasta çözüm oranında artış; işlem başına operasyon maliyetinde düşüş. Paket, Aginies platformunun Identity & authorisation guardrails, Supervisor · sub-agents, Human-in-the-loop, Core banking tools (MCP) yeteneklerini kullanır ve ajan kütüphanesindeki Diyalog bankacılığı: sesli ve yazılı işlem asistanı, Kredi kartı limit artışı, Sesli bankacılıkla para transferi, Kart dolandırıcılık alarmı triyajı senaryolarıyla canlı izlenebilir. Sorun Müşteriler menülerde kayboluyor, çağrı merkezi basit işlemlerle doluyor. Mevcut asistanlar bilgi veriyor ama işlem yapamıyor; yapabilenler de denetlenemiyor. Ajanlar ne yapar - 01Mevcut kimlik doğrulama akışına entegre; yetkisiz işlem talebi guardrail’de durur. - 02Limit artışı, para transferi, otomatik ödeme talimatı ve ekstre tek diyalogda tamamlanır. - 03Belirli tutar ya da risk eşiği üzerindeki işlemler insan onayına gider; her karar iz bırakır. - 04Sesli kanal SIP ile çağrı merkezine bağlanır; aynı akış yazılı ve sesli çalışır. Paket kapsamı - Mobil, web, WhatsApp ve sesli kanalda aynı işlem ajanı - Transfer, fatura, limit, kart ve itiraz işlemleri çekirdek bankacılık araçlarıyla - İşlem bazlı kimlik ve OTP eşikleri; dolandırıcılık skoru her işlemde - Çağrı merkezine SIP ile sıcak aktarım, konuşma özeti ve referanslar Başlamak için gerekenler - Çekirdek bankacılık ve kart sistemleri için işlem API’leri veya MCP sunucuları - Mevcut dijital kimlik ve OTP altyapısı - Bilgi güvenliği ile yazılmış işlem eşikleri ve onay kuralları Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Diyalog Bankacılığı SüpervizörüKart SüpervizörüSesli İşlem SüpervizörüDolandırıcılık Süpervizörü Uzman ajanlar Diyalog AjanıKimlik Ajanıİşlem AjanıRisk AjanıNiyet AjanıKonuşma AjanıTransfer AjanıTriyaj Ajanıİletişim AjanıKart Ajanı Modeller (Hub kataloğu) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-voice-tr-v3 Dokunulan sistemler Mobile App ChatCore BankingCard ManagementFraud EngineSMS GatewayKYC / Customer 360Push & SMS GatewayGenesys SIPWhatsApp Business APISalesforce CRM Kanallar appvoicewhatsappsmsapi Simülasyon hacimleri 296.400çalıştırma / ay %95,2başarı oranı 5,9 smedyan süre $0,0031tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Kanal fark etmeksizin konuşarak tamamlanan bankacılık - Çağrı merkezi yükünde azalma, ilk temasta çözüm oranında artış - İşlem başına operasyon maliyetinde düşüş Kullanılan yetenekler Identity & authorisation guardrailsSupervisor · sub-agentsHuman-in-the-loopCore banking tools (MCP)Full audit trailSIP voice Platformu keşfet → Ölçülen hedefler | Konuşarak tamamlanan işlem payı | %55–70 | | Çağrı merkezine düşen basit işlem | %40 azalma | | İşlem başına operasyon maliyeti | %50 ve üzeri düşüş | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Diyalog bankacılığı: sesli ve yazılı işlem asistanıLimit artışı > %60, itiraz tutarı > 5.000 TL veya yeni alıcıya transfer → operasyon onayı - Kredi kartı limit artışıLimit > 100.000 TL → şube yöneticisi onayı - Sesli bankacılıkla para transferiTutar > 50.000 TL veya yeni alıcı → çağrı merkezi temsilcisine sıcak aktarım - Kart dolandırıcılık alarmı triyajıTeyitli dolandırıcılık zararı > 2.500 TL → analist onayı ile kalıcı iptal - paketYeni alıcı, yüksek tutar ve itiraz her zaman ek doğrulama veya insan onayı ister - paketBDDK bilgi sistemleri mevzuatına uyumlu kurulum; tüm sistemler Türkiye’de Yol haritası - 1 · Keşif ve tasarım · 2–3 haftaEn sık 20 niyet ve işlem seçilir, kimlik ve onay eşikleri bilgi güvenliği ile yazılır, kanal ve sistem entegrasyonları listelenir. - 2 · Pilot · 4–6 haftaTek kanalda, sınırlı müşteri grubuyla; test korpusu koşturulur, gölge modda insan temsilci izler, eşikler gerçek trafikle ayarlanır. - 3 · YaygınlaştırmaDiğer kanallar ve işlemler eklenir; panolar, SIEM aktarımı ve kill-switch devredilir; aylık model ve kural gözden geçirmesi başlar. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları BankacılıkHITL ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı Müşteri konuşur ya da yazar; fatura, limit, itiraz ve transfer aynı oturumda çekirdek sistemde tamamlanır. 186.000 çalıştırma / ay %94,7 başarı 5 ajan BankacılıkHITL ### Kredi kartı limit artışı Müşteri mobilde yazar ya da çağrı merkezinde konuşur; karar saniyeler içinde verilir, eşik üstü yönetici onayına gider. 18.400 çalıştırma / ay %96,4 başarı 4 ajan BankacılıkHITL ### Sesli bankacılıkla para transferi Müşteri arar ve konuşur; ses izi, güvenlik sorusu ve OTP ile doğrulanır, transfer temsilcisiz tamamlanır. 27.000 çalıştırma / ay %93,1 başarı 5 ajan BankacılıkHITL ### Kart dolandırıcılık alarmı triyajı Kart saniyeler içinde bloklanır, müşteriye WhatsApp'tan sorulur; yanıta göre kart açılır ya da iptal edilir. 65.000 çalıştırma / ay %97,1 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Kredi ve müşteri edinimi operasyonları URL: https://www.aginies.com/tr/cozumler/credit-and-onboarding-operations Language: tr Summary: Hesap açılışı ve kredi başvurusu farklı ekiplerde, farklı araçlarla ve farklı kurallarla işleniyor. Belge incelemesi günler alıyor; müşteri aynı bilgiyi birkaç kez veriyor, karar gerekçesi dağınık kalıyor. Bankacılık · Finansal analiz ve operasyonel zekâ # Kredi ve müşteri edinimi operasyonları Dijital hesap açılışı, limit artışı ve kredi belgeleri tek karar mantığıyla, şubeye gitmeden. Kısaca Kredi ve müşteri edinimi operasyonları, Bankacılık sektörü için “finansal analiz ve operasyonel zekâ” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Kimlik, canlılık, yaptırım ve adres doğrulaması tek akışta; temiz başvurular dakikalar içinde açılır; gelir, skor ve politika kuralları otomatik onay bandını belirler; bant dışı başvuru analiste gider. Beklenen sonuçlar: şubeye gitmeden açılan hesap ve dakikalar içinde kredi kararı; belge inceleme iş yükünde belirgin azalma; denetime hazır, gerekçeli ve tutarlı kararlar. Paket, Aginies platformunun Identity & KYC tools, Policy engine, Document extraction (Vision), Human-in-the-loop yeteneklerini kullanır ve ajan kütüphanesindeki Dijital müşteri edinimi (KYC), Kredi kartı limit artışı, Konut kredisi belge incelemesi, Kredi kartı borcu ödeme senaryolarıyla canlı izlenebilir. Sorun Hesap açılışı ve kredi başvurusu farklı ekiplerde, farklı araçlarla ve farklı kurallarla işleniyor. Belge incelemesi günler alıyor; müşteri aynı bilgiyi birkaç kez veriyor, karar gerekçesi dağınık kalıyor. Ajanlar ne yapar - 01Kimlik, canlılık, yaptırım ve adres doğrulaması tek akışta; temiz başvurular dakikalar içinde açılır. - 02Gelir, skor ve politika kuralları otomatik onay bandını belirler; bant dışı başvuru analiste gider. - 03Bordro, tapu ve ekspertiz belgeleri okunur, alanlar çapraz kontrol edilir, eksikler müşteriden istenir. - 04Şube, mobil ve çağrı merkezi aynı karar motorunu kullanır; her karar gerekçesiyle saklanır. Paket kapsamı - Dijital müşteri edinimi: kimlik, canlılık, yaptırım ve adres doğrulama tek akışta - Limit artışı, kredi başvurusu ve konut kredisi belge inceleme - Gelir ve skor çekimi; politika motoru; otomatik onay bandı - Şube, mobil ve çağrı merkezi kanalında aynı karar mantığı Başlamak için gerekenler - Çekirdek bankacılık, kredi bürosu ve kimlik servisleri API erişimi - Kredi politikası ve onay matrisi - Belge tiplerine göre örnek set (bordro, tapu, ekspertiz) Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Müşteri Edinim SüpervizörüKart SüpervizörüKredi Dosya Süpervizörü Uzman ajanlar Belge AjanıCanlılık AjanıTarama AjanıKayıt AjanıNiyet AjanıRisk Ajanıİşlem AjanıÇıkarım AjanıGelir AjanıDeğerleme AjanıÖdeme AjanıSes Ajanı Modeller (Hub kataloğu) aginies-konrul-1aginies-sunkar-1aginies-tulpar-4aginies-talay-2aginies-voice-tr-v3 Dokunulan sistemler KYC / Customer 360Sanctions Screening (Dow Jones)Core BankingDocument Store (S3)SMS GatewayPush & SMS GatewayLoan Origination SystemCredit Bureau (KKB)Email (Exchange)Card ManagementKnowledge Base · kart_sss_v3OTP / SMS Gateway Kanallar appwebvoiceemailbranch Simülasyon hacimleri 118.600çalıştırma / ay %96,0başarı oranı 7,8 smedyan süre $0,0035tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Şubeye gitmeden açılan hesap ve dakikalar içinde kredi kararı - Belge inceleme iş yükünde belirgin azalma - Denetime hazır, gerekçeli ve tutarlı kararlar Kullanılan yetenekler Identity & KYC toolsPolicy engineDocument extraction (Vision)Human-in-the-loopFull audit trailCore banking tools (MCP) Platformu keşfet → Ölçülen hedefler | Şubeye gitmeden açılan hesap payı | %70–85 | | Kredi kararı süresi (otomatik bant) | Dakikalar | | Belge incelemesi iş yükü | %60 azalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Dijital müşteri edinimi (KYC)Yaptırım/PEP bulanık eşleşme skoru > 0,65 → uyum uzmanı onayı - Kredi kartı limit artışıLimit > 100.000 TL → şube yöneticisi onayı - Konut kredisi belge incelemesiKredi tutarı > 3.000.000 TL veya DTI > %45 → kredi tahsis uzmanı onayı - Kredi kartı borcu ödemeTutar > 100.000 TL → şube yöneticisi onayı - paketBant dışı başvuru, yüksek tutar ve şüpheli belge her zaman kredi analistinde - paketKarar gerekçesi ve kullanılan veri her başvuruda saklanır; BDDK mevzuatına uyumlu Yol haritası - 1 · Belge ve kural seti · 2 haftaBelge tipleri, alan sözlüğü, eşleştirme ve tolerans kuralları finans ekibiyle yazılır; örnek set etiketlenir. - 2 · Paralel koşum · 4 haftaAjan ve mevcut süreç aynı belgeleri işler; fark raporu her hafta incelenir, doğruluk eşiği karşılanınca geçiş. - 3 · Kapanış entegrasyonuAy sonu takvimine bağlanır; istisna kuyruğu, denetim izi ve mutabakat raporları iç denetime açılır. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları BankacılıkHITL ### Dijital müşteri edinimi (KYC) Kimlik, canlılık ve yaptırım kontrolü tek akışta; temiz başvurular şubeye gitmeden dakikalarda hesap açar. 42.000 çalıştırma / ay %93,8 başarı 5 ajan BankacılıkHITL ### Kredi kartı limit artışı Müşteri mobilde yazar ya da çağrı merkezinde konuşur; karar saniyeler içinde verilir, eşik üstü yönetici onayına gider. 18.400 çalıştırma / ay %96,4 başarı 4 ajan BankacılıkHITL ### Konut kredisi belge incelemesi Bordro, SGK dökümü ve ekspertiz otomatik okunur; gelir, LTV ve eksik belge kontrolü danışman beklemeden biter. 6.200 çalıştırma / ay %94,7 başarı 4 ajan BankacılıkHITL ### Kredi kartı borcu ödeme Müşteri yazar ya da konuşur; borç sorgulanır, hesap seçilir, OTP ile doğrulanır, ödeme çekirdek sistemde yapılır. 52.000 çalıştırma / ay %97,8 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Savunma ihracat kontrolü ve tedarik zinciri güvenliği URL: https://www.aginies.com/tr/cozumler/defence-export-control-compliance Language: tr Summary: Her ihracat kalemi için kontrol listesi eşlemesi, taraf taraması ve lisans dosyası elle; tedarikçi anketleri ve belgeleri nitelendirme sonrası okunuyor. Gizlilik dereceli ortamda dış model kullanılamıyor. Savunma Sanayii · Veri koruma ve uyum # Savunma ihracat kontrolü ve tedarik zinciri güvenliği Sınıflandırma, taraf tarama, lisans dosyası ve tedarikçi nitelendirme: kapalı ağda, kararı uzmanda bırakan ajanlar. Kısaca Savunma ihracat kontrolü ve tedarik zinciri güvenliği, Savunma Sanayii sektörü için “veri koruma ve uyum” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Parça parametreleri kontrol listeleriyle eşlenir; sınıflandırma gerekçesi kaynak maddelerle yazılır; alıcı, aracı ve son kullanıcı yaptırım listeleriyle taranır; sahiplik zinciri çözülür, son kullanım beyanı doğrulanır. Beklenen sonuçlar: uyum uzmanı dosya başına saatler yerine dakikalar harcar; kontrole tabi kalem sevkiyat öncesi yakalanır; tedarikçi riski nitelendirme öncesi görünür. Paket, Aginies platformunun Air-gap deployment, Control-list RAG, Party screening tools, Vision extraction yeteneklerini kullanır ve ajan kütüphanesindeki Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolü, İhracat kontrolü ve son kullanıcı taraması, Tedarikçi güvenlik ve uyum değerlendirmesi, Platform bakım (MRO) iş emri hazırlığı senaryolarıyla canlı izlenebilir. Sorun Her ihracat kalemi için kontrol listesi eşlemesi, taraf taraması ve lisans dosyası elle; tedarikçi anketleri ve belgeleri nitelendirme sonrası okunuyor. Gizlilik dereceli ortamda dış model kullanılamıyor. Ajanlar ne yapar - 01Parça parametreleri kontrol listeleriyle eşlenir; sınıflandırma gerekçesi kaynak maddelerle yazılır. - 02Alıcı, aracı ve son kullanıcı yaptırım listeleriyle taranır; sahiplik zinciri çözülür, son kullanım beyanı doğrulanır. - 03Lisans başvuru taslağı ve ek listesi hazırlanır; sipariş lisans gelene kadar bloke edilir. - 04Tedarikçi anketleri gereklilik setiyle puanlanır; eksikler ve risk sınıfı güvenlik müdürüne sunulur. Paket kapsamı - Kontrol listesi eşleme ve taraf tarama; ihracat lisansı dosyası - Tedarikçi güvenlik değerlendirmesi ve anketi - Teknik veri erişimi: yetki ve işaretleme kontrolü - Kapalı ağda, kurum veri merkezinde işletim Başlamak için gerekenler - Kontrol listeleri ve kurum sınıflandırma şeması - Tedarikçi ve parça kayıtları - Kapalı ağ kurulumu için GPU altyapısı Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Teknik Yayın Süpervizörüİhracat Uyum SüpervizörüTedarikçi Güvenlik SüpervizörüBakım Süpervizörü Uzman ajanlar Yetki AjanıYayın AjanıProsedür AjanıParça AjanıSınıflandırma AjanıTaraf Tarama AjanıLisans AjanıBelge AjanıDeğerlendirme AjanıDosya AjanıArıza AjanıTeknik Yayın AjanıPlanlama Ajanı Modeller (Hub kataloğu) aginies-sunkar-1aginies-tulpar-4aginies-talay-2aginies-konrul-1 Dokunulan sistemler Hangar Tablet AppAccess ControlTech Pub Repository (S1000D)MRO SystemAudit LogERP (SAP GTS)Control & Sanctions Lists (vector)Trade Compliance Case SystemE-mailSupplier Management (SRM)Security Requirements (vector)Document StoreMaintenance Management SystemTechnical Publications (S1000D vector)Spares InventoryTechnician App Kanallar appapiemailbatchweb Simülasyon hacimleri 17.400çalıştırma / ay %95,7başarı oranı 7,6 smedyan süre $0,0053tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Uyum uzmanı dosya başına saatler yerine dakikalar harcar - Kontrole tabi kalem sevkiyat öncesi yakalanır - Tedarikçi riski nitelendirme öncesi görünür Kullanılan yetenekler Air-gap deploymentControl-list RAGParty screening toolsVision extractionHuman-in-the-loopFull audit trail Platformu keşfet → Ölçülen hedefler | Taraf tarama süresi | Günlerden dakikalara | | Lisans dosyası eksik belge | %80 azalma | | Dış ağa çıkan veri | Sıfır | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolüProsedürden sapma önerisi → yetkili mühendis onayı; ihracat kontrollü içerik → yetki seviyesi şart - İhracat kontrolü ve son kullanıcı taramasıKontrole tabi her kalem → ihracat uyum yöneticisi onayı; ajan lisans kararı veremez - Tedarikçi güvenlik ve uyum değerlendirmesiNitelendirme kararı her zaman tesis güvenlik müdüründe; ajan yalnızca hazırlar - Platform bakım (MRO) iş emri hazırlığıUçuşa elverişliliği etkileyen her iş paketi → bakım planlayıcısı onayı - paketİhracat ve erişim kararı her zaman yetkili uyum sorumlusunda - paketTüm erişimler kullanıcı, belge ve zaman damgasıyla denetim günlüğünde Yol haritası - 1 · Kural ve kontrol tasarımı · 2–3 haftaMevzuat maddeleri kontrol setine çevrilir, karar matrisi uyum ekibiyle onaylanır, iz gereksinimleri yazılır. - 2 · Gölge dönem · 4 haftaAjan bulgu ve öneri üretir, uzman karar verir; uyumsuzluk oranı ve gerekçe kalitesi ölçülür. - 3 · Denetime hazır işletimKontrol kanıtları otomatik toplanır; denetim raporları ve istisna defteri iç denetime açılır. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Savunma SanayiiHITL ### Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolü Teknisyen hangarda sorar; doğru revizyondaki prosedür, parça numarası ve emniyet notu yetki kontrolüyle gelir, her erişim iz bırakır. 9.800 çalıştırma / ay %95,7 başarı 5 ajan Savunma SanayiiHITL ### İhracat kontrolü ve son kullanıcı taraması Parça, alıcı ve son kullanım tek akışta sınıflandırılır; lisans gereği tespit edilir, karar her zaman uyum yöneticisinde kalır. 2.600 çalıştırma / ay %96,1 başarı 4 ajan Savunma SanayiiHITL ### Tedarikçi güvenlik ve uyum değerlendirmesi Tedarikçi anketleri ve belgeleri okunur, tesis güvenlik gereklilikleriyle karşılaştırılır, eksikler listelenir; nitelendirme kararı güvenlik müdürünün. 900 çalıştırma / ay %95,2 başarı 4 ajan Savunma SanayiiHITL ### Platform bakım (MRO) iş emri hazırlığı Arıza raporu teknik yayınlarla eşlenir, görev kartı ve parça listesi çıkar, iş emri planlayıcı onayıyla açılır. 4.100 çalıştırma / ay %95,7 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # E-ticaret müşteri desteği ve aksiyon URL: https://www.aginies.com/tr/cozumler/e-commerce-support Language: tr Summary: Sezonda destek hacmi katlanıyor; ekipler aynı on soruya cevap verirken sipariş sisteminde işlem yapmak için ekran değiştiriyor. Chatbotlar cevap veriyor ama iptal edemiyor. Perakende & E-ticaret · Müşteri etkileşimi ve deneyim # E-ticaret müşteri desteği ve aksiyon Sipariş nerede, iade, teslimat değişikliği: ajan işlemi sipariş sisteminde tamamlar. Kısaca E-ticaret müşteri desteği ve aksiyon, Perakende & E-ticaret sektörü için “müşteri etkileşimi ve deneyim” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Niyet ve varlık çıkarımı: sipariş numarası, ürün, talep türü tek turda anlaşılır; katalog ve politika bilgisi RAG ile getirilir; beden, uyumluluk, iade koşulları doğru cevaplanır. Beklenen sonuçlar: yüksek hacimli talepler destek ekibi büyümeden çözülür; sipariş güncelleme, iptal ve teslimat değişikliği konuşma içinde tamamlanır; ürün cevapları kurumun kendi bilgi kaynaklarına bağlı, uydurma yok. Paket, Aginies platformunun Multi-source RAG, Supervisor · sub-agents, Session memory, REST / MCP tools yeteneklerini kullanır ve ajan kütüphanesindeki Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin, Siparişim nerede? (WISMO), İade ve para iadesi, Ürün soru-cevap asistanı senaryolarıyla canlı izlenebilir. Sorun Sezonda destek hacmi katlanıyor; ekipler aynı on soruya cevap verirken sipariş sisteminde işlem yapmak için ekran değiştiriyor. Chatbotlar cevap veriyor ama iptal edemiyor. Ajanlar ne yapar - 01Niyet ve varlık çıkarımı: sipariş numarası, ürün, talep türü tek turda anlaşılır. - 02Katalog ve politika bilgisi RAG ile getirilir; beden, uyumluluk, iade koşulları doğru cevaplanır. - 03Sipariş sistemine bağlı araçlarla teslimat adresi değişir, iptal başlatılır, iade etiketi üretilir. - 04Politika dışı talepler (yüksek tutarlı iade, kampanya istisnası) insan onayına düşer. Paket kapsamı - Web, mobil ve WhatsApp üzerinde tek diyalog ajanı; sipariş, iade, teslimat ve ürün soruları - Sipariş yönetimi, kargo ve ödeme sistemlerine bağlı işlem araçları - Katalog ve politika bilgi tabanı; iade ve garanti koşulları kaynak gösterilerek - Temsilciye sıcak aktarım ve konuşma özeti Başlamak için gerekenler - Sipariş ve kargo sistemleri için okuma/yazma API veya MCP erişimi - Güncel iade, teslimat ve kampanya politikaları - Müşteri kimliğini doğrulayan oturum veya OTP akışı Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Stil SüpervizörüSipariş Süpervizörüİade SüpervizörüAlışveriş Süpervizörü Uzman ajanlar Görsel AjanıStilist AjanıKatalog AjanıSepet AjanıNiyet AjanıTakip AjanıYanıt AjanıKabul AjanıPolitika Ajanıİşlem Ajanı Modeller (Hub kataloğu) aginies-konrul-1aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Dokunulan sistemler WhatsApp BusinessProduct Catalogue (PIM)InventoryCustomer ProfileCart & Checkout APIShopify / OMSCarrier API (Yurtiçi, DHL)WhatsApp Business APIZendeskPayment Gateway (iyzico)Email (Exchange)Vector StorePIM / CatalogueWeb Chat Widget Kanallar appwebwhatsappvoiceemail Simülasyon hacimleri 420.000çalıştırma / ay %97,4başarı oranı 5,9 smedyan süre $0,0009tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Yüksek hacimli talepler destek ekibi büyümeden çözülür - Sipariş güncelleme, iptal ve teslimat değişikliği konuşma içinde tamamlanır - Ürün cevapları kurumun kendi bilgi kaynaklarına bağlı, uydurma yok Kullanılan yetenekler Multi-source RAGSupervisor · sub-agentsSession memoryREST / MCP toolsGuardrailsIntent detectionEntity extraction Platformu keşfet → Ölçülen hedefler | İlk temasta çözüm oranı | +15–25 puan | | Temsilciye aktarılan konuşma payı | %35 altı | | Sezon tepe yükünde ek temsilci ihtiyacı | %40–60 azalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Kombin asistanı: beden, stok, bütçe ve stile göre tam kombinKampanya kuralı dışı indirim veya kişiye özel fiyat talebi → müşteri hizmetleri - İade ve para iadesiİade tutarı > 2.500 TL veya 12 ayda > 5 iade → ekip lideri onayı - paketÖdeme iadesi ve adres değişikliği kimlik doğrulaması olmadan yapılmaz - paketHer işlem müşteri, sipariş ve ajan adımıyla denetim izinde Yol haritası - 1 · Keşif ve tasarım · 2–3 haftaEn sık 20 niyet ve işlem seçilir, kimlik ve onay eşikleri bilgi güvenliği ile yazılır, kanal ve sistem entegrasyonları listelenir. - 2 · Pilot · 4–6 haftaTek kanalda, sınırlı müşteri grubuyla; test korpusu koşturulur, gölge modda insan temsilci izler, eşikler gerçek trafikle ayarlanır. - 3 · YaygınlaştırmaDiğer kanallar ve işlemler eklenir; panolar, SIEM aktarımı ve kill-switch devredilir; aylık model ve kural gözden geçirmesi başlar. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Perakende & E-ticaretHITL ### Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin Müşteri bir fotoğraf ve bir cümle gönderir; ajan stokta olan, bedenine uyan ve bütçeye sığan üç kombin kurar, seçileni sepete ekler. 142.000 çalıştırma / ay %96,1 başarı 5 ajan Perakende & E-ticaret ### Siparişim nerede? (WISMO) Müşteri sorar, sipariş ve kargo olayları anında birleşir; gecikmede yeni tarih verilir, kargo araştırılır. 110.000 çalıştırma / ay %98,1 başarı 4 ajan Perakende & E-ticaretHITL ### İade ve para iadesi Politika ve müşteri geçmişi saniyede kontrol edilir, etiket üretilir; iade tutarı kargo taramasıyla ödenir. 48.000 çalıştırma / ay %96,7 başarı 4 ajan Perakende & E-ticaret ### Ürün soru-cevap asistanı Beden, uyumluluk ve özellik soruları katalog ve iade verisiyle yanıtlanır; stok ve fiyat doğrulanır. 120.000 çalıştırma / ay %98,7 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # E-posta sınıflandırma ve ek doğrulama URL: https://www.aginies.com/tr/cozumler/email-classification Language: tr Summary: Ortak gelen kutularına günde binlerce e-posta düşüyor. Yanlış ekibe giden talepler gecikiyor; eksik ekli talepler operasyona girip geri dönüyor. Kurumsal Fonksiyonlar · Operasyonel otomasyon # E-posta sınıflandırma ve ek doğrulama Gelen kutusu doğru ekibe, eksik ek daha iş başlamadan tespit edilir. Kısaca E-posta sınıflandırma ve ek doğrulama, Kurumsal Fonksiyonlar sektörü için “operasyonel otomasyon” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: E-posta niyeti ve aciliyeti sınıflandırılır, doğru kuyruğa atanır; ekler belge tipine göre kontrol edilir: eksik, yanlış ya da okunamaz ekler işaretlenir. Beklenen sonuçlar: talepler ilk seferde doğru ekibe ulaşır; standart sorulara anında cevap, memnuniyet artışı; eksik belgeler operasyona girmeden yakalanır. Paket, Aginies platformunun Document embedding pipeline, Dynamic agent routing, Entity extraction, Email service integrations yeteneklerini kullanır ve ajan kütüphanesindeki E-posta sınıflandırma ve ek kontrolü, Borçlar hesabı (AP) posta kutusu, Vatandaş başvurusu yönlendirme, BT yardım masası L1 senaryolarıyla canlı izlenebilir. Sorun Ortak gelen kutularına günde binlerce e-posta düşüyor. Yanlış ekibe giden talepler gecikiyor; eksik ekli talepler operasyona girip geri dönüyor. Ajanlar ne yapar - 01E-posta niyeti ve aciliyeti sınıflandırılır, doğru kuyruğa atanır. - 02Ekler belge tipine göre kontrol edilir: eksik, yanlış ya da okunamaz ekler işaretlenir. - 03Standart taleplere kurum bilgi kaynağından anında cevap üretilir. - 04Her yönlendirme ve cevap denetim iziyle kaydedilir. Paket kapsamı - Ortak posta kutularının sınıflandırılması ve doğru birime yönlendirilmesi - Ek doğrulama: belge tipi, okunabilirlik, eksik ek, zararlı içerik - Taslak yanıt ve bilet oluşturma; SLA sayacı - Kişisel veri maskeleme ve saklama kuralları Başlamak için gerekenler - Microsoft 365 veya Google Workspace için posta kutusu erişimi - Bilet sistemi (ServiceNow, Jira SM) API erişimi - Birim ve konu taksonomisi, SLA tanımları Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Posta SüpervizörüAP SüpervizörüBaşvuru SüpervizörüYardım Masası Süpervizörü Uzman ajanlar Sınıflandırma AjanıEk Kontrol AjanıYönlendirme AjanıYakalama AjanıKodlama AjanıKayıt AjanıSLA AjanıDosya AjanıNiyet AjanıErişim Ajanı Modeller (Hub kataloğu) aginies-sunkar-1aginies-konrul-1aginies-talay-2aginies-tulpar-4 Dokunulan sistemler Exchange / M365 MailMailbox Rules & Checklists Vector StoreServiceNowDocument Store (S3)Oracle EBSVendor Master / Procuremente-Devlet GatewayCase ManagementLegislation Vector StoreSMS GatewayActive Directory / Entra IDMicrosoft Teams Kanallar emailbatchwebappvoice Simülasyon hacimleri 215.000çalıştırma / ay %97,8başarı oranı 5,7 smedyan süre $0,001tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Talepler ilk seferde doğru ekibe ulaşır - Standart sorulara anında cevap, memnuniyet artışı - Eksik belgeler operasyona girmeden yakalanır Kullanılan yetenekler Document embedding pipelineDynamic agent routingEntity extractionEmail service integrationsAgent templatingAudit trail logging Platformu keşfet → Ölçülen hedefler | Doğru birime ilk yönlendirme | %95 ve üzeri | | İlk yanıt süresi | Saatlerden dakikalara | | Manuel triyaj iş yükü | %70 azalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Borçlar hesabı (AP) posta kutusuSipariş (SAS) olmayan fatura > 25.000 TL → masraf yeri sahibi onayı - Vatandaş başvurusu yönlendirmeSınıflandırma güveni < 0,85 veya personel şikâyeti → evrak kayıt memuru onayı - BT yardım masası L1Finans veya üretim grubuna erişim → grup sahibi onayı - paketHukuki ve şikâyet nitelikli e-postalar otomatik yanıtlanmaz - paketSınıflandırma kararı ve gerekçesi bilette saklanır Yol haritası - 1 · Süreç haritası · 2 haftaSinyal kaynakları, karar noktaları ve iş emri sistemleri haritalanır; otomatik karar sınırı ve onay zinciri belirlenir. - 2 · Pilot · 4 haftaTek hat, tek bölge ya da tek süreçte; ajan öneriyi hazırlar, insan uygular; sapma ve kazanç ölçülür. - 3 · Otonom işletimEşik altı kararlar otomatik, üstü onaya; vardiya raporları ve SLA panoları devrededir. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Kurumsal Fonksiyonlar ### E-posta sınıflandırma ve ek kontrolü Ortak kutuya gelen her e-posta sınıflandırılır, ekler listeyle karşılaştırılır, eksikler anında sorulur. 86.000 çalıştırma / ay %97,6 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Borçlar hesabı (AP) posta kutusu ap@ kutusuna düşen her fatura yakalanır, kodlanır ve ERP'ye kaydedilir; yalnızca istisnalar analiste gider. 28.000 çalıştırma / ay %96,8 başarı 4 ajan Kamu ve HükümetHITL ### Vatandaş başvurusu yönlendirme e-Devlet başvurusu saniyeler içinde sınıflandırılır, yasal süresi hesaplanır ve doğru birime SLA ile atanır. 54.000 çalıştırma / ay %98,1 başarı 4 ajan Kurumsal FonksiyonlarHITL ### BT yardım masası L1 Şifre sıfırlama ve erişim talepleri sohbette çözülür; finans grubu erişimi grup sahibine Teams'te sorulur. 47.000 çalıştırma / ay %98,5 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Saha servis ve sevk otomasyonu URL: https://www.aginies.com/tr/cozumler/field-service-dispatch Language: tr Summary: Arıza ve sapma sinyalleri var ama servis emri, parça rezervasyonu, teknisyen ataması ve müşteri bildirimi ayrı ekranlarda elle yapılıyor. Kritik olay sıradan taleple aynı kuyrukta bekliyor. IoT ve Bağlı Cihazlar · Operasyonel otomasyon # Saha servis ve sevk otomasyonu Sensör alarmı ya da müşteri bildirimi → doğru teknisyen, doğru parça, doğru zaman. Kısaca Saha servis ve sevk otomasyonu, IoT ve Bağlı Cihazlar sektörü için “operasyonel otomasyon” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Sinyal doğrulanır ve önceliklendirilir: sensör serisi, cihaz geçmişi ve bilinen sorunlar birlikte okunur; görev kartı, parça listesi ve alet gereksinimi teknik yayınlardan ve stoktan çıkarılır. Beklenen sonuçlar: sinyalden servis emrine geçiş dakikalar içinde, elle giriş olmadan; doğru parça ve doğru görev kartı ile ilk ziyarette çözüm; müşteri ve kalite kaydı aynı anda güncel. Paket, Aginies platformunun Event ingestion, Time-series tools, Multi-source RAG, REST / MCP tools yeteneklerini kullanır ve ajan kütüphanesindeki Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA, Soğuk zincir sapması müdahalesi, Bakım talebi triyajı, Platform bakım (MRO) iş emri hazırlığı senaryolarıyla canlı izlenebilir. Sorun Arıza ve sapma sinyalleri var ama servis emri, parça rezervasyonu, teknisyen ataması ve müşteri bildirimi ayrı ekranlarda elle yapılıyor. Kritik olay sıradan taleple aynı kuyrukta bekliyor. Ajanlar ne yapar - 01Sinyal doğrulanır ve önceliklendirilir: sensör serisi, cihaz geçmişi ve bilinen sorunlar birlikte okunur. - 02Görev kartı, parça listesi ve alet gereksinimi teknik yayınlardan ve stoktan çıkarılır. - 03En yakın uygun teknisyen atanır, servis emri açılır, müşteri ETA ile bilgilendirilir. - 04Uçuşa elverişlilik, ürün imhası ya da yüksek maliyet gibi kararlar ilgili yöneticiye düşer. Paket kapsamı - Arıza ve servis taleplerinin tanısı; uzaktan çözüm ya da saha ataması - Teknisyen, parça ve rota planlama; müşteri randevusu - Saha ekibine iş emri, prosedür ve parça kartı - Müşteri bilgilendirme ve kapanış doğrulaması Başlamak için gerekenler - Saha servis yönetimi sistemi (FSM) API erişimi - Cihaz ve varlık kayıtları, parça stoğu - Teknisyen yetkinlik ve bölge tanımları Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Kontrol Kulesi SüpervizörüSoğuk Zincir SüpervizörüBakım Süpervizörü Uzman ajanlar Tahmin AjanıRota AjanıMüşteri AjanıFinans AjanıSensör AjanıKalite AjanıSaha AjanıSinyal AjanıSaha Notu AjanıTeşhis AjanıPlanlama AjanıArıza AjanıTeknik Yayın AjanıParça Ajanı Modeller (Hub kataloğu) aginies-tulpar-4aginies-talay-2aginies-sunkar-1aginies-voice-tr-v3 Dokunulan sistemler TMSTelematicsCarrier APICustomer PortalERP BillingIoT PlatformWMS / Lot TrackingStability Rules (vector)Field ServiceWhatsApp Business APISCADA / HistorianCMMS (Maximo)SAP S/4HANA MMManuals & Work Order Vector StoreMaintenance Management SystemTechnical Publications (S1000D vector)Spares InventoryTechnician App Kanallar apiemailwhatsappappvoicebatch Simülasyon hacimleri 106.200çalıştırma / ay %93,6başarı oranı 7,1 smedyan süre $0,0031tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Sinyalden servis emrine geçiş dakikalar içinde, elle giriş olmadan - Doğru parça ve doğru görev kartı ile ilk ziyarette çözüm - Müşteri ve kalite kaydı aynı anda güncel Kullanılan yetenekler Event ingestionTime-series toolsMulti-source RAGREST / MCP toolsHuman-in-the-loopFull I/O logging Platformu keşfet → Ölçülen hedefler | Uzaktan çözülen talep payı | %30–45 | | İlk ziyarette çözüm | +15 puan | | Teknisyen başına günlük iş | +20% | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLARota maliyet artışı > %15 veya tazmin > 10.000 TL → operasyon müdürü onayı - Soğuk zincir sapması müdahalesiÜrün taşıma ya da imha kararı → kalite yöneticisi onayı - Bakım talebi triyajıPlanlı duruş > 4 saat → vardiya amiri onayı - Platform bakım (MRO) iş emri hazırlığıUçuşa elverişliliği etkileyen her iş paketi → bakım planlayıcısı onayı - paketGaranti dışı değişim ve maliyetli parça insan onayında - paketUzaktan komutlar cihaz sahibinin izin kapsamıyla sınırlı Yol haritası - 1 · Süreç haritası · 2 haftaSinyal kaynakları, karar noktaları ve iş emri sistemleri haritalanır; otomatik karar sınırı ve onay zinciri belirlenir. - 2 · Pilot · 4 haftaTek hat, tek bölge ya da tek süreçte; ajan öneriyi hazırlar, insan uygular; sapma ve kazanç ölçülür. - 3 · Otonom işletimEşik altı kararlar otomatik, üstü onaya; vardiya raporları ve SLA panoları devrededir. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları LojistikHITL ### Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA Telematik sinyali gecikmeyi erken yakalar; alternatif rota ve taşıyıcı seçilir, müşteri bilgilendirilir, SLA tazmini hesaplanır. 96.000 çalıştırma / ay %93,4 başarı 5 ajan IoT ve Bağlı CihazlarHITL ### Soğuk zincir sapması müdahalesi Sıcaklık sapması anında değerlendirilir, ürün riski hesaplanır, saha ekibi sevk edilir, kalite kararı kayıt altına alınır. 2.700 çalıştırma / ay %97,4 başarı 4 ajan Üretim & OtomotivHITL ### Bakım talebi triyajı Sensör alarmı ve teknisyen sesli notu tek iş emrine dönüşür; yedek parça ve güvenli duruş penceresi hazırdır. 3.400 çalıştırma / ay %95,2 başarı 5 ajan Savunma SanayiiHITL ### Platform bakım (MRO) iş emri hazırlığı Arıza raporu teknik yayınlarla eşlenir, görev kartı ve parça listesi çıkar, iş emri planlayıcı onayıyla açılır. 4.100 çalıştırma / ay %95,7 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Finansal analiz ve operasyonel zekâ URL: https://www.aginies.com/tr/cozumler/financial-analysis-operational-intelligence Language: tr Summary: Finans ekipleri faturalar, ekstreler ve yatırım talimatları arasında veriyi elle birleştiriyor. Mutabakat istisnaları geç fark ediliyor; analiz raporu hazırlanana kadar karar penceresi kapanıyor. Finansal Teknolojiler · Finansal analiz ve operasyonel zekâ # Finansal analiz ve operasyonel zekâ Finansal veriyi ve belgeyi okuyan, yorumlayan, karar destekleyen ajanlar. Kısaca Finansal analiz ve operasyonel zekâ, Finansal Teknolojiler sektörü için “finansal analiz ve operasyonel zekâ” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Finansal asistan ekstre, defter ve raporlardan soruya doğrudan cevap verir; her rakam kaynağına bağlıdır; fatura işleme ajanı belgeyi çıkarır, sipariş ve mal kabulle eşleştirir, istisnayı işaretler. Beklenen sonuçlar: mutabakat ve fatura döngüsü günlerden saatlere iner; kararlar güncel ve kaynağı belli veriyle alınır; istisna ve risk desenleri gerçekleşmeden görünür. Paket, Aginies platformunun Entity extraction, Multi-source RAG, ERP / core tools (MCP), Anomaly detection yeteneklerini kullanır ve ajan kütüphanesindeki Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit, Ödeme mutabakatı istisnaları, Tedarikçi faturası üçlü eşleştirme, BNPL anlık taksit kararı senaryolarıyla canlı izlenebilir. Sorun Finans ekipleri faturalar, ekstreler ve yatırım talimatları arasında veriyi elle birleştiriyor. Mutabakat istisnaları geç fark ediliyor; analiz raporu hazırlanana kadar karar penceresi kapanıyor. Ajanlar ne yapar - 01Finansal asistan ekstre, defter ve raporlardan soruya doğrudan cevap verir; her rakam kaynağına bağlıdır. - 02Fatura işleme ajanı belgeyi çıkarır, sipariş ve mal kabulle eşleştirir, istisnayı işaretler. - 03Yatırım talimatları analiz edilir: tutar, enstrüman, limit ve uygunluk kuralları kontrol edilir. - 04Eşik üstü işlem ve anormal desen finans yöneticisine gerekçeli özetle sunulur. Paket kapsamı - Gelir, maliyet ve nakit verisinin çoklu kaynaktan toplanması ve doğrulanması - Varyans analizi, anomali tespiti ve kök neden açıklaması - Yönetim raporu taslağı: tablo, grafik ve yorum, kaynak gösterilerek - Soru-cevap: “bu ay pazarlama maliyeti neden arttı?” gibi sorulara veriyle yanıt Başlamak için gerekenler - Veri ambarı veya ERP raporlama katmanına okuma erişimi - Hesap planı, maliyet merkezi ve KPI tanımları - Rapor şablonları ve dağıtım listesi Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Üye İşyeri SüpervizörüMutabakat SüpervizörüFatura SüpervizörüKredi Karar Süpervizörü Uzman ajanlar Diyalog AjanıPOS AjanıOperasyon AjanıRisk AjanıEşleştirme Ajanıİnceleme AjanıDüzeltme AjanıYakalama AjanıKayıt AjanıKabul AjanıBüro AjanıKarar Ajanı Modeller (Hub kataloğu) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-konrul-1 Dokunulan sistemler WhatsApp BusinessTerminal ManagementSettlement LedgerRefund ServiceTicketingPSP / Acquirer APICard NetworkCore LedgerCase ManagementEmailSAP S/4HANAExchange / M365 MailTolerance Policy Vector StoreMerchant Checkout APICredit Bureau (KKB / Findeks)Open Banking APISMS Gateway Kanallar whatsappwebvoicebatchapiemailappsms Simülasyon hacimleri 139.620çalıştırma / ay %95,7başarı oranı 7,2 smedyan süre $0,0028tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Mutabakat ve fatura döngüsü günlerden saatlere iner - Kararlar güncel ve kaynağı belli veriyle alınır - İstisna ve risk desenleri gerçekleşmeden görünür Kullanılan yetenekler Entity extractionMulti-source RAGERP / core tools (MCP)Anomaly detectionHuman-in-the-loopFull audit trail Platformu keşfet → Ölçülen hedefler | Aylık rapor hazırlama süresi | %60–70 kısalma | | Anomali tespit süresi | Ay sonundan aynı haftaya | | Kaynak gösterilen rapor cümlesi | %100 | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Üye işyeri operasyon asistanı: POS, hakediş, iade ve limitİade > 25.000 TL veya limit artışı > %50 → risk ekibi onayı - Ödeme mutabakatı istisnalarıTek fark > 10.000 TL veya toplam > 50.000 TL → finans müdürü onayı ile ayarlama kaydı - Tedarikçi faturası üçlü eşleştirmeFiyat sapması > %2 veya > 1.000 TL → satın alma sorumlusu onayı - BNPL anlık taksit kararıSepet > 15.000 TL veya büro skoru < 1.200 → kredi uzmanı onayı - paketRapor finans yöneticisinin onayıyla dağıtılır; ajan taslak hazırlar - paketVeri erişimi rol bazlı; hassas bordro ve maaş verisi ayrı politikada Yol haritası - 1 · Belge ve kural seti · 2 haftaBelge tipleri, alan sözlüğü, eşleştirme ve tolerans kuralları finans ekibiyle yazılır; örnek set etiketlenir. - 2 · Paralel koşum · 4 haftaAjan ve mevcut süreç aynı belgeleri işler; fark raporu her hafta incelenir, doğruluk eşiği karşılanınca geçiş. - 3 · Kapanış entegrasyonuAy sonu takvimine bağlanır; istisna kuyruğu, denetim izi ve mutabakat raporları iç denetime açılır. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Finansal TeknolojilerHITL ### Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit İşyeri WhatsApp'tan yazar; POS arızası, hakediş sorgusu, iade ve limit artışı tek diyalogda çözülür. 64.000 çalıştırma / ay %93,8 başarı 5 ajan Finansal TeknolojilerHITL ### Ödeme mutabakatı istisnaları PSP mutabakat dosyası ana defterle eşlenir; küçük farklar otomatik düzeltilir, büyükleri finans onayına gider. 620 çalıştırma / ay %95,1 başarı 4 ajan Üretim & OtomotivHITL ### Tedarikçi faturası üçlü eşleştirme Fatura, sipariş ve mal kabul SAP'de otomatik eşleşir; yalnızca tolerans dışı kalemler satın almacıya gider. 21.000 çalıştırma / ay %97,1 başarı 4 ajan Finansal TeknolojilerHITL ### BNPL anlık taksit kararı Ödemede saniyeler içinde taksit kararı; büro ve açık bankacılık verisi birleşir, limit üstü uzmana gider. 54.000 çalıştırma / ay %97,3 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Dolandırıcılık ve itiraz operasyonları URL: https://www.aginies.com/tr/cozumler/fraud-and-dispute-operations Language: tr Summary: Dolandırıcılık alarmları analist kuyruğunda bekliyor, itiraz dosyaları kart şeması sürelerine yetişmiyor, müşteriyle teyit için aramalar gecikiyor. Her ekip kendi listesini tutuyor. Bankacılık · Operasyonel otomasyon # Dolandırıcılık ve itiraz operasyonları Alarm triyajı, ters ibraz dosyaları, kart bloke ve erken tahsilat tek operasyon masasında. Kısaca Dolandırıcılık ve itiraz operasyonları, Bankacılık sektörü için “operasyonel otomasyon” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Alarm skorlanır, işyeri ve davranış analizi eklenir, müşteriyle SMS veya aramayla teyit alınır; itiraz dosyası kanıt listesi, kart şeması kodu ve son tarihle açılır; eksik kanıt otomatik istenir. Beklenen sonuçlar: alarm başına inceleme süresinde büyük düşüş; süresi geçen itiraz dosyası kalmaz; daha az yanlış pozitif bloke, daha az müşteri şikâyeti. Paket, Aginies platformunun Fraud engine tools, Case management, Supervisor · sub-agents, Human-in-the-loop yeteneklerini kullanır ve ajan kütüphanesindeki Kart dolandırıcılık alarmı triyajı, Ters ibraz (chargeback) itirazı, Diyalog bankacılığı: sesli ve yazılı işlem asistanı, Erken dönem tahsilat iletişimi senaryolarıyla canlı izlenebilir. Sorun Dolandırıcılık alarmları analist kuyruğunda bekliyor, itiraz dosyaları kart şeması sürelerine yetişmiyor, müşteriyle teyit için aramalar gecikiyor. Her ekip kendi listesini tutuyor. Ajanlar ne yapar - 01Alarm skorlanır, işyeri ve davranış analizi eklenir, müşteriyle SMS veya aramayla teyit alınır. - 02İtiraz dosyası kanıt listesi, kart şeması kodu ve son tarihle açılır; eksik kanıt otomatik istenir. - 03Kart bloke, yeni kart ve geçici iade çekirdek sistemde yapılır; müşteri her adımda bilgilendirilir. - 04Erken tahsilat iletişimi ödeme planı önerisiyle; kalıcı kararlar operasyon onayına gider. Paket kapsamı - Dolandırıcılık alarmı triyajı: skor, işyeri analizi, müşteriyle teyit - Ters ibraz ve işlem itirazı dosyaları: kanıt toplama, kart şeması kuralları - Kart bloke, yeni kart ve geçici iade işlemleri - Erken tahsilat iletişimi ve ödeme planı Başlamak için gerekenler - Kart yönetimi, dolandırıcılık motoru ve itiraz sistemleri API erişimi - Kart şeması itiraz kuralları ve süreleri - Müşteri iletişim kanalları ve rıza kayıtları Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Dolandırıcılık Süpervizörüİtiraz SüpervizörüDiyalog Bankacılığı SüpervizörüTahsilat Süpervizörü Uzman ajanlar Triyaj Ajanıİletişim AjanıKart AjanıKabul AjanıKanıt AjanıDosyalama AjanıDiyalog AjanıKimlik Ajanıİşlem AjanıRisk AjanıSegment AjanıÖdeme Planı Ajanı Modeller (Hub kataloğu) aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Dokunulan sistemler Fraud EngineCard ManagementWhatsApp Business APISalesforce CRMSMS GatewayDocument Store (S3)Visa / Mastercard Dispute APIEmail (Exchange)Mobile App ChatCore BankingCollections Platform Kanallar whatsappsmsvoiceapiappemailbranchbatch Simülasyon hacimleri 356.800çalıştırma / ay %94,2başarı oranı 7,1 smedyan süre $0,0026tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Alarm başına inceleme süresinde büyük düşüş - Süresi geçen itiraz dosyası kalmaz - Daha az yanlış pozitif bloke, daha az müşteri şikâyeti Kullanılan yetenekler Fraud engine toolsCase managementSupervisor · sub-agentsHuman-in-the-loopFull audit trailSMS / voice channels Platformu keşfet → Ölçülen hedefler | Alarm başına inceleme süresi | %70 kısalma | | Süresi geçen itiraz dosyası | Sıfıra yakın | | Yanlış pozitif bloke | %30 azalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Kart dolandırıcılık alarmı triyajıTeyitli dolandırıcılık zararı > 2.500 TL → analist onayı ile kalıcı iptal - Ters ibraz (chargeback) itirazıTutar > 2.500 TL → geçici alacak öncesi itiraz ekibi lideri onayı - Diyalog bankacılığı: sesli ve yazılı işlem asistanıLimit artışı > %60, itiraz tutarı > 5.000 TL veya yeni alıcıya transfer → operasyon onayı - Erken dönem tahsilat iletişimiPlan bakiyesi > 25.000 TL veya > 3 taksit → tahsilat ekip lideri onayı - paketKalıcı bloke ve kayıp yazımı her zaman operasyon onayında - paketHer karar kanıt ve skorla birlikte dosyada; iç denetim ve kart şeması incelemesine hazır Yol haritası - 1 · Süreç haritası · 2 haftaSinyal kaynakları, karar noktaları ve iş emri sistemleri haritalanır; otomatik karar sınırı ve onay zinciri belirlenir. - 2 · Pilot · 4 haftaTek hat, tek bölge ya da tek süreçte; ajan öneriyi hazırlar, insan uygular; sapma ve kazanç ölçülür. - 3 · Otonom işletimEşik altı kararlar otomatik, üstü onaya; vardiya raporları ve SLA panoları devrededir. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları BankacılıkHITL ### Kart dolandırıcılık alarmı triyajı Kart saniyeler içinde bloklanır, müşteriye WhatsApp'tan sorulur; yanıta göre kart açılır ya da iptal edilir. 65.000 çalıştırma / ay %97,1 başarı 4 ajan BankacılıkHITL ### Ters ibraz (chargeback) itirazı İtiraz alınır, kanıt paketi toplanır ve kart şemasına dosyalanır; eşik üstü geçici alacak onaya gider. 9.800 çalıştırma / ay %95,2 başarı 4 ajan BankacılıkHITL ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı Müşteri konuşur ya da yazar; fatura, limit, itiraz ve transfer aynı oturumda çekirdek sistemde tamamlanır. 186.000 çalıştırma / ay %94,7 başarı 5 ajan BankacılıkHITL ### Erken dönem tahsilat iletişimi Gecikmiş hesaplar segmentlenir, doğru kanaldan nazikçe hatırlatılır; müşteri ödeme planını kendisi kurar. 96.000 çalıştırma / ay %91,2 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Sağlık operasyonları ve klinik yönlendirme URL: https://www.aginies.com/tr/cozumler/healthcare-operations-clinical-guidance Language: tr Summary: Hasta talepleri çağrı merkezi, portal ve mesaj kanallarında yığılıyor; yanlış polikliniğe randevu ve gelmeyen hasta kapasiteyi eritiyor. Klinik ekipler zamanını idari yönlendirmeye harcıyor. Sağlık · Müşteri etkileşimi ve deneyim # Sağlık operasyonları ve klinik yönlendirme Akıllı triyaj ve hasta yönlendirme; klinik karar her zaman hekimde kalır. Kısaca Sağlık operasyonları ve klinik yönlendirme, Sağlık sektörü için “müşteri etkileşimi ve deneyim” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Hasta talebi ve şikâyeti analiz edilir; aciliyet ve uygun bölüm kurumun triyaj protokolüne göre belirlenir; randevu ajanı hekim takvimi ve kapasiteye göre en uygun slotu önerir; hatırlatma ve yeniden planlama otomatiktir. Beklenen sonuçlar: hasta ilk temasta doğru bölüme ve zamana yönlenir; gelmeyen hasta oranı düşer, kapasite kullanımı artar; klinik ekipler idari yükten kurtulur. Paket, Aginies platformunun Intent detection, Triage protocol guardrails, Scheduling tools (MCP), Human-in-the-loop yeteneklerini kullanır ve ajan kütüphanesindeki Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuç, Hasta ön kabul ve triyaj, Randevu planlama ve no-show azaltma, Ön yetkilendirme (prior authorization) senaryolarıyla canlı izlenebilir. Sorun Hasta talepleri çağrı merkezi, portal ve mesaj kanallarında yığılıyor; yanlış polikliniğe randevu ve gelmeyen hasta kapasiteyi eritiyor. Klinik ekipler zamanını idari yönlendirmeye harcıyor. Ajanlar ne yapar - 01Hasta talebi ve şikâyeti analiz edilir; aciliyet ve uygun bölüm kurumun triyaj protokolüne göre belirlenir. - 02Randevu ajanı hekim takvimi ve kapasiteye göre en uygun slotu önerir; hatırlatma ve yeniden planlama otomatiktir. - 03Ön yetkilendirme ve sevk belgeleri hazırlanır, eksik bilgi hastadan istenir. - 04Kırmızı bayraklı semptomlar anında klinisyene yükseltilir; tanı ve tedavi kararı hiçbir zaman ajanda değildir. Paket kapsamı - Randevu, ön kayıt, hazırlık talimatı ve sonuç bildirimi tek koordinatörde - Sigorta ön onayı ve SGK/özel sigorta sorguları - Klinik protokollere bağlı hazırlık ve kontrendikasyon kontrolü - Çağrı merkezi ve WhatsApp kanalı; hekime onay kartı Başlamak için gerekenler - HBYS ve randevu sistemleri için API erişimi - Klinik protokoller ve hazırlık talimatı seti - Hasta kimlik doğrulama ve KVKK rıza kayıtları Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Hasta Yolculuğu SüpervizörüTriyaj SüpervizörüRandevu SüpervizörüÖn Yetki Süpervizörü Uzman ajanlar Diyalog AjanıRandevu AjanıKlinik Hazırlık AjanıSigorta AjanıSes AjanıSemptom AjanıAciliyet AjanıYönlendirme AjanıGelmeme Riski AjanıRezervasyon Ajanıİstem AjanıÖdeyici Kural AjanıGönderim Ajanı Modeller (Hub kataloğu) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-voice-tr-v3aginies-konrul-1 Dokunulan sistemler WhatsApp BusinessHospital Information SystemSchedulingInsurer PortalLab / ImagingEHR (HL7/FHIR)HIS SchedulingTriage Protocol Vector StoreWhatsApp Business APIHospital Information System (HIS)SMS GatewayPayer Policy Vector StorePayer PortalExchange / M365 Mail Kanallar whatsappvoiceappsmsbatchapiemailweb Simülasyon hacimleri 150.700çalıştırma / ay %96,4başarı oranı 7,3 smedyan süre $0,0023tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Hasta ilk temasta doğru bölüme ve zamana yönlenir - Gelmeyen hasta oranı düşer, kapasite kullanımı artar - Klinik ekipler idari yükten kurtulur Kullanılan yetenekler Intent detectionTriage protocol guardrailsScheduling tools (MCP)Human-in-the-loopSensitive data maskingFull audit trail Platformu keşfet → Ölçülen hedefler | Randevuya gelmeme oranı | %30–40 azalma | | Hazırlık eksikliğiyle ertelenen işlem | %50 azalma | | Çağrı merkezinde ortalama işlem süresi | %35 kısalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuçİlaç etkileşimi, kontrendikasyon şüphesi veya ek tetkik ihtiyacı → hekim onayı - Hasta ön kabul ve triyajHer vakada aciliyet seviyesini triyaj hemşiresi onaylar; "acil durum" vakaları doğrudan 112'ye yönlendirilir - Ön yetkilendirme (prior authorization)Cerrahi işlem veya maliyet > 25.000 TL → kullanım yönetimi hemşiresi onayı - paketKlinik karar hiçbir zaman ajanda değil; şüphede hekim onayı - paketSağlık verisi hastane sınırında işlenir; erişim rol ve amaç bazlı Yol haritası - 1 · Keşif ve tasarım · 2–3 haftaEn sık 20 niyet ve işlem seçilir, kimlik ve onay eşikleri bilgi güvenliği ile yazılır, kanal ve sistem entegrasyonları listelenir. - 2 · Pilot · 4–6 haftaTek kanalda, sınırlı müşteri grubuyla; test korpusu koşturulur, gölge modda insan temsilci izler, eşikler gerçek trafikle ayarlanır. - 3 · YaygınlaştırmaDiğer kanallar ve işlemler eklenir; panolar, SIEM aktarımı ve kill-switch devredilir; aylık model ve kural gözden geçirmesi başlar. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları SağlıkHITL ### Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuç Hasta yazar ya da arar; randevu, ön hazırlık talimatı, sigorta ön onayı ve sonuç bildirimi tek koordinatörde birleşir. 58.000 çalıştırma / ay %95,1 başarı 5 ajan SağlıkHITL ### Hasta ön kabul ve triyaj Hasta yazar ya da anlatır; anamnez ve aciliyet skoru saniyeler sürer, son kararı her zaman hemşire verir. 24.500 çalıştırma / ay %95,8 başarı 5 ajan Sağlık ### Randevu planlama ve no-show azaltma Gelmeme riski yüksek hastalar WhatsApp'tan önceden aranır, boşalan slotlar bekleme listesinden anında dolar. 62.000 çalıştırma / ay %98,2 başarı 4 ajan SağlıkHITL ### Ön yetkilendirme (prior authorization) İstem düşer düşmez ödeyici kuralları taranır, klinik kanıt toplanır, dosya dakikalar içinde gönderime hazır. 6.200 çalıştırma / ay %94,1 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # İK ve çalışan deneyimi URL: https://www.aginies.com/tr/cozumler/hr-employee-experience Language: tr Summary: Başvurular elle eleniyor, yeni çalışan ilk haftasını erişim ve form peşinde geçiriyor. İK ekibi aynı soruları tekrar cevaplarken stratejik işe zaman kalmıyor. Kurumsal Fonksiyonlar · İK ve çalışan deneyimi # İK ve çalışan deneyimi İşe alımdan uyuma: veriye dayalı, hızlı ve ölçeklenebilir İK süreçleri. Kısaca İK ve çalışan deneyimi, Kurumsal Fonksiyonlar sektörü için “ik ve çalışan deneyimi” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: İşe alım ajanı başvuruları ilan kriterlerine göre ön değerlendirir, mülakat planlar; nihai karar İK ve yöneticidedir; onboarding ajanı hesap, donanım ve eğitim atamalarını İK, BT ve yönetici arasında koordine eder. Beklenen sonuçlar: değerlendirmeden uyuma İK döngüsü kısalır; yeni çalışan ilk gün hazır: erişim, donanım, eğitim; İK ekibi tekrar işten kurtulur, deneyime odaklanır. Paket, Aginies platformunun HRIS / ITSM tools (MCP), Multi-source RAG, Supervisor · sub-agents, Human-in-the-loop yeteneklerini kullanır ve ajan kütüphanesindeki Çalışan işe alıştırma (onboarding), BT yardım masası L1, E-posta sınıflandırma ve ek kontrolü senaryolarıyla canlı izlenebilir. Sorun Başvurular elle eleniyor, yeni çalışan ilk haftasını erişim ve form peşinde geçiriyor. İK ekibi aynı soruları tekrar cevaplarken stratejik işe zaman kalmıyor. Ajanlar ne yapar - 01İşe alım ajanı başvuruları ilan kriterlerine göre ön değerlendirir, mülakat planlar; nihai karar İK ve yöneticidedir. - 02Onboarding ajanı hesap, donanım ve eğitim atamalarını İK, BT ve yönetici arasında koordine eder. - 03Dahili yardım masası izin, bordro ve politika sorularını kurum kaynağından anında cevaplar. - 04Kişisel veri maskelenir; her adım HRIS ve ITSM kayıtlarıyla izlenir. Paket kapsamı - İşe alım ön eleme, mülakat planlama ve aday iletişimi - Onboarding: hesap, ekipman, eğitim ve ilk hafta planı - Çalışan soruları: izin, bordro, politika; kaynak gösteren cevaplar - Ayrılış ve rol değişikliği otomasyonu Başlamak için gerekenler - İK sistemi (SAP SuccessFactors, Workday, yerel İKYS) API erişimi - Kimlik yönetimi ve BT varlık sistemlerine entegrasyon - Güncel İK politikaları ve süreç sahipleri Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler İşe Alıştırma SüpervizörüYardım Masası SüpervizörüPosta Süpervizörü Uzman ajanlar İK AjanıKimlik AjanıBT AjanıEğitim AjanıNiyet AjanıErişim AjanıKayıt AjanıSınıflandırma AjanıEk Kontrol AjanıYönlendirme Ajanı Modeller (Hub kataloğu) aginies-sunkar-1aginies-talay-2aginies-tulpar-4aginies-konrul-1 Dokunulan sistemler SAP SuccessFactorsActive Directory / Entra IDServiceNowLMSExchange / M365 MailMicrosoft TeamsMailbox Rules & Checklists Vector StoreDocument Store (S3) Kanallar apiemailappvoiceweb Simülasyon hacimleri 134.900çalıştırma / ay %97,9başarı oranı 5,5 smedyan süre $0,0007tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Değerlendirmeden uyuma İK döngüsü kısalır - Yeni çalışan ilk gün hazır: erişim, donanım, eğitim - İK ekibi tekrar işten kurtulur, deneyime odaklanır Kullanılan yetenekler HRIS / ITSM tools (MCP)Multi-source RAGSupervisor · sub-agentsHuman-in-the-loopPII maskingAudit trail Platformu keşfet → Ölçülen hedefler | Onboarding tamamlanma süresi | Haftalardan günlere | | İK’ya gelen tekrar eden soru | %60 azalma | | Aday geri dönüş süresi | 24 saat altı | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Çalışan işe alıştırma (onboarding)Şablon dışı erişim veya yönetici yetkisi → bölüm yöneticisi onayı - BT yardım masası L1Finans veya üretim grubuna erişim → grup sahibi onayı - paketİşe alım kararı ve ücret teklifi her zaman insanda; ajan eşleştirir ve planlar - paketÇalışan verisi amaç sınırlı; bordro bilgisine yalnızca yetkili rol erişir Yol haritası - 1 · Politika ve rol kataloğu · 2 haftaİK politikaları, süreç sahipleri ve sistem erişimleri tanımlanır; kişisel veri sınırları yazılır. - 2 · Pilot · 4 haftaBir departman ya da bir rolle; kalite puanı ve kullanıcı geri bildirimi haftalık izlenir. - 3 · YaygınlaştırmaTüm departmanlar; eşikler, inceleme örneklemi ve kill-switch İK yönetiminde. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Kurumsal FonksiyonlarHITL ### Çalışan işe alıştırma (onboarding) İK kaydı açıldığı anda hesaplar, ekipman ve eğitim planı hazırlanır; şablon dışı erişimler yöneticiye sorulur. 1.900 çalıştırma / ay %97,2 başarı 5 ajan Kurumsal FonksiyonlarHITL ### BT yardım masası L1 Şifre sıfırlama ve erişim talepleri sohbette çözülür; finans grubu erişimi grup sahibine Teams'te sorulur. 47.000 çalıştırma / ay %98,5 başarı 4 ajan Kurumsal Fonksiyonlar ### E-posta sınıflandırma ve ek kontrolü Ortak kutuya gelen her e-posta sınıflandırılır, ekler listeyle karşılaştırılır, eksikler anında sorulur. 86.000 çalıştırma / ay %97,6 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Bilgi yönetimi ve karar desteği URL: https://www.aginies.com/tr/cozumler/information-management-decision-support Language: tr Summary: Kurumsal bilgi mevzuat, genelge, sözleşme ve dijital varlıklar hâlinde farklı sistemlerde dağınık. Doğru belgeyi bulmak saatler alıyor; kararlar güncel olmayan ya da eksik bilgiyle veriliyor. Kamu ve Hükümet · Bilgi yönetimi ve karar desteği # Bilgi yönetimi ve karar desteği Mevzuat, politika ve sözleşmelere saniyeler içinde kaynaklı erişim. Kısaca Bilgi yönetimi ve karar desteği, Kamu ve Hükümet sektörü için “bilgi yönetimi ve karar desteği” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Mevzuat ve politika belgeleri sürümlü olarak indekslenir; RAG ile soruya madde bazında, kaynaklı cevap verilir; mevzuat analizi ajanı değişiklikleri tespit eder, etkilenen süreç ve belgeleri listeler. Beklenen sonuçlar: bilgiye erişim saatlerden saniyelere iner; kararlar güncel mevzuat ve kaynağı belli belgeye dayanır; mevzuat değişikliğinin etkisi proaktif görülür. Paket, Aginies platformunun Multi-source RAG, Document embedding pipeline, Version-aware retrieval, Citation grounding yeteneklerini kullanır ve ajan kütüphanesindeki Sözleşme madde çıkarımı, Ruhsat başvurusu ön inceleme, Ön yetkilendirme (prior authorization), Gümrük belgesi hazırlama senaryolarıyla canlı izlenebilir. Sorun Kurumsal bilgi mevzuat, genelge, sözleşme ve dijital varlıklar hâlinde farklı sistemlerde dağınık. Doğru belgeyi bulmak saatler alıyor; kararlar güncel olmayan ya da eksik bilgiyle veriliyor. Ajanlar ne yapar - 01Mevzuat ve politika belgeleri sürümlü olarak indekslenir; RAG ile soruya madde bazında, kaynaklı cevap verilir. - 02Mevzuat analizi ajanı değişiklikleri tespit eder, etkilenen süreç ve belgeleri listeler. - 03Sözleşme playbook’u: maddeler standart pozisyonla karşılaştırılır, sapmalar işaretlenir. - 04Dijital varlıklar (görsel, doküman, kayıt) otomatik etiketlenir ve aranabilir hâle gelir; erişim yetkiye göre. Paket kapsamı - Mevzuat, prosedür ve kurum belgelerine bağlı, kaynak gösteren soru-cevap - Belge karşılaştırma, özetleme ve değişiklik takibi - Erişim politikası: kim hangi belgeyi görebilir, cevapta uygulanır - Karar notu ve gerekçe taslağı; uzman onayıyla yayın Başlamak için gerekenler - Belge yönetim sistemi veya dosya paylaşımına okuma erişimi - Belge sahipleri ve güncelleme sorumluları - SSO grupları ile erişim politikası eşlemesi Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Sözleşme SüpervizörüRuhsat SüpervizörüÖn Yetki SüpervizörüGümrük Süpervizörü Uzman ajanlar Madde AjanıPlaybook AjanıCLM AjanıBelge AjanıMevzuat AjanıDosya Ajanıİstem AjanıÖdeyici Kural AjanıGönderim AjanıTarife AjanıMüşavir Ajanı Modeller (Hub kataloğu) aginies-tulpar-4aginies-talay-2aginies-konrul-1aginies-sunkar-1 Dokunulan sistemler CLM (Contract Lifecycle Management)Legal Playbook Vector StoreDocument Store (S3)Exchange / M365 Maile-Devlet GatewayCase ManagementGIS / Zoning PlanRegulation Vector StoreEHR (HL7/FHIR)Payer Policy Vector StorePayer PortalSAP S/4HANAWMSTariff & Regulation Vector StoreCustoms Broker Portal Kanallar emailwebapibatch Simülasyon hacimleri 15.000çalıştırma / ay %94,7başarı oranı 7,2 smedyan süre $0,011tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Bilgiye erişim saatlerden saniyelere iner - Kararlar güncel mevzuat ve kaynağı belli belgeye dayanır - Mevzuat değişikliğinin etkisi proaktif görülür Kullanılan yetenekler Multi-source RAGDocument embedding pipelineVersion-aware retrievalCitation groundingRole-based accessEntity extraction Platformu keşfet → Ölçülen hedefler | Kaynak gösterilen cevap payı | %100 | | Uzman tarafından doğru bulunan cevap | %90 ve üzeri | | Bilgi arama süresi | %60 azalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Sözleşme madde çıkarımıYüksek riskli sapma veya sözleşme değeri > 500.000 € → hukuk müşaviri onayı - Ruhsat başvurusu ön incelemeHer ruhsat kararı ruhsat memuru tarafından verilir; sistem yalnızca ön inceleme yapar - Ön yetkilendirme (prior authorization)Cerrahi işlem veya maliyet > 25.000 TL → kullanım yönetimi hemşiresi onayı - Gümrük belgesi hazırlamaGTİP güven skoru < 0,90 veya vergi farkı > 2.000 € → gümrük uzmanı onayı - paketKaynak bulunamayan soruya cevap verilmez; “bilmiyorum” döner - paketYasal yorum ve dış paylaşım her zaman uzman onayında Yol haritası - 1 · Kaynak envanteri · 2 haftaBilgi kaynakları, sahipleri ve erişim politikaları listelenir; chunk ve etiketleme kuralları belirlenir. - 2 · Pilot · 3–4 haftaBir birimle; kaynak gösteren cevaplar uzmanlarca puanlanır, kapsama boşlukları kapatılır. - 3 · Kurum geneliErişim politikaları SSO gruplarına bağlanır; güncelleme hattı ve kalite panosu devrededir. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Kurumsal FonksiyonlarHITL ### Sözleşme madde çıkarımı Sözleşme maddeleri çıkarılır, hukuk playbook'u ile karşılaştırılır; sapmalar redline önerisiyle avukata gider. 1.400 çalıştırma / ay %95,1 başarı 4 ajan Kamu ve HükümetHITL ### Ruhsat başvurusu ön inceleme Eksik belge ve mevzuat uyumu dakikalar içinde raporlanır; kararı her zaman ruhsat memuru verir. 3.100 çalıştırma / ay %95,8 başarı 4 ajan SağlıkHITL ### Ön yetkilendirme (prior authorization) İstem düşer düşmez ödeyici kuralları taranır, klinik kanıt toplanır, dosya dakikalar içinde gönderime hazır. 6.200 çalıştırma / ay %94,1 başarı 4 ajan LojistikHITL ### Gümrük belgesi hazırlama GTİP kodu, fatura ve çeki listesi dakikalarda hazırlanır; düşük güvenli satırlar gümrük uzmanına gider. 4.300 çalıştırma / ay %94,7 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Fatura ve doküman veri çıkarımı URL: https://www.aginies.com/tr/cozumler/invoice-document-extraction Language: tr Summary: Faturalar, sözleşmeler ve formlar e-posta ve portallardan farklı formatlarda geliyor. Veri girişi yavaş, hata payı yüksek, eksik belge işi durduruyor. Kurumsal Fonksiyonlar · Finansal analiz ve operasyonel zekâ # Fatura ve doküman veri çıkarımı OCR → doğrulama → ERP: elle veri girişi olmadan temiz kayıt. Kısaca Fatura ve doküman veri çıkarımı, Kurumsal Fonksiyonlar sektörü için “finansal analiz ve operasyonel zekâ” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Belgeler sınıflandırılır; sayfa düzeni ne olursa olsun alanlar çıkarılır; çıkarılan veri satın alma siparişi ve mal kabul kayıtlarıyla eşleştirilir, toleranslar kontrol edilir. Beklenen sonuçlar: anahtar alanlar elle girilmeden çıkarılır; finans süreçleri günlerden dakikalara iner; doğrulanmış veri doğrudan ERP ve iş sistemlerine akar. Paket, Aginies platformunun Entity extraction, Document embedding pipeline, Stateful execution, Full I/O logging yeteneklerini kullanır ve ajan kütüphanesindeki Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada, Tedarikçi faturası üçlü eşleştirme, Borçlar hesabı (AP) posta kutusu, Underwriting belge çıkarımı senaryolarıyla canlı izlenebilir. Sorun Faturalar, sözleşmeler ve formlar e-posta ve portallardan farklı formatlarda geliyor. Veri girişi yavaş, hata payı yüksek, eksik belge işi durduruyor. Ajanlar ne yapar - 01Belgeler sınıflandırılır; sayfa düzeni ne olursa olsun alanlar çıkarılır. - 02Çıkarılan veri satın alma siparişi ve mal kabul kayıtlarıyla eşleştirilir, toleranslar kontrol edilir. - 03Eksik ya da tutarsız belge anında tespit edilir, gönderene otomatik geri bildirim gider. - 04Doğrulanan kayıt ERP’ye yazılır; her alan için kaynak belge ve güven skoru saklanır. Paket kapsamı - Fatura, irsaliye, sözleşme ve form çıkarımı; alan bazında güven skoru ve kaynak belge - Sipariş ve irsaliye ile üç yönlü eşleştirme, tolerans kuralları - İstisna kuyruğu: eksik alan, tutar sapması, mükerrer fatura - ERP kayıt oluşturma ve ödeme planlama araçları Başlamak için gerekenler - ERP için kayıt ve sorgu API erişimi (SAP, Oracle, Logo, Netsis) - Belge tiplerini temsil eden 200–500 örnek belge - Finans ekibinden tolerans ve onay kuralları Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Satın Alma SüpervizörüFatura SüpervizörüAP SüpervizörüUnderwriting Süpervizörü Uzman ajanlar Talep AjanıTedarik AjanıSözleşme AjanıFatura AjanıYakalama AjanıEşleştirme AjanıKayıt AjanıKodlama AjanıBelge AjanıTıbbi Değerlendirme AjanıUnderwriting Ajanı Modeller (Hub kataloğu) aginies-sunkar-1aginies-tulpar-4aginies-talay-2aginies-konrul-1 Dokunulan sistemler Procurement InboxERPSupplier PortalContract RepositoryPayment HubSAP S/4HANAExchange / M365 MailTolerance Policy Vector StoreOracle EBSDocument Store (S3)Vendor Master / ProcurementUnderwriting WorkbenchVector StoreEmail (Exchange) Kanallar emailapibatchweb Simülasyon hacimleri 67.800çalıştırma / ay %95,9başarı oranı 7,8 smedyan süre $0,0044tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Anahtar alanlar elle girilmeden çıkarılır - Finans süreçleri günlerden dakikalara iner - Doğrulanmış veri doğrudan ERP ve iş sistemlerine akar Kullanılan yetenekler Entity extractionDocument embedding pipelineStateful executionFull I/O loggingHuman-in-the-loopREST / MCP Platformu keşfet → Ölçülen hedefler | Dokunmadan işlenen belge payı | %70–85 | | Alan doğruluğu (kritik alanlar) | %98 ve üzeri | | Fatura işleme süresi | Günlerden saatlere | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masadaTutar > 250.000 TL, tek teklif veya çerçeve dışı tedarikçi → satın alma komitesi - Tedarikçi faturası üçlü eşleştirmeFiyat sapması > %2 veya > 1.000 TL → satın alma sorumlusu onayı - Borçlar hesabı (AP) posta kutusuSipariş (SAS) olmayan fatura > 25.000 TL → masraf yeri sahibi onayı - Underwriting belge çıkarımıEk prim ≥ %50 veya teminat > 5.000.000 TL → kıdemli underwriter onayı - paketTolerans dışı sapma ve yeni tedarikçi her zaman insana - paketKaynak belge ve çıkarılan değer birlikte saklanır; denetim örneklemi hazır Yol haritası - 1 · Belge ve kural seti · 2 haftaBelge tipleri, alan sözlüğü, eşleştirme ve tolerans kuralları finans ekibiyle yazılır; örnek set etiketlenir. - 2 · Paralel koşum · 4 haftaAjan ve mevcut süreç aynı belgeleri işler; fark raporu her hafta incelenir, doğruluk eşiği karşılanınca geçiş. - 3 · Kapanış entegrasyonuAy sonu takvimine bağlanır; istisna kuyruğu, denetim izi ve mutabakat raporları iç denetime açılır. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Kurumsal FonksiyonlarHITL ### Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada Talep e-postayla gelir; teklifler karşılaştırılır, sözleşme maddeleri kontrol edilir, sipariş açılır, fatura üç yönlü eşleşir ve ödeme planlanır. 14.000 çalıştırma / ay %92,7 başarı 5 ajan Üretim & OtomotivHITL ### Tedarikçi faturası üçlü eşleştirme Fatura, sipariş ve mal kabul SAP'de otomatik eşleşir; yalnızca tolerans dışı kalemler satın almacıya gider. 21.000 çalıştırma / ay %97,1 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Borçlar hesabı (AP) posta kutusu ap@ kutusuna düşen her fatura yakalanır, kodlanır ve ERP'ye kaydedilir; yalnızca istisnalar analiste gider. 28.000 çalıştırma / ay %96,8 başarı 4 ajan SigortaHITL ### Underwriting belge çıkarımı Sağlık raporu ve tahliller okunur, kılavuza göre değerlendirilir; underwriter yalnızca eşik üstünü görür. 4.800 çalıştırma / ay %94,9 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # IoT cihaz filosu operasyonu URL: https://www.aginies.com/tr/cozumler/iot-device-fleet-operations Language: tr Summary: Binlerce cihaz aynı kök nedenden yüzlerce alarm üretiyor; firmware güncellemesi tüm filoya bir kerede çıkıyor ve geri alma planı yok. Telemetri hacmi bulut model maliyetini taşınamaz kılıyor. IoT ve Bağlı Cihazlar · Yazılım geliştirme ve teknoloji verimliliği # IoT cihaz filosu operasyonu Anomali kümeleme, kök neden ve sağlık kapılı OTA dağıtımı: filo tek olay, tek karar. Kısaca IoT cihaz filosu operasyonu, IoT ve Bağlı Cihazlar sektörü için “yazılım geliştirme ve teknoloji verimliliği” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Telemetri anomalileri model, firmware ve ağ geçidi bazında kümelenir; tek olay açılır; kök neden adayları firmware geçmişi ve bilinen sorunlarla puanlanır; güven skoru yazılır. Beklenen sonuçlar: alarm gürültüsü olay sayısına iner; ekip gerçek soruna bakar; güvenli, kanıtlı ve geri alınabilir firmware dağıtımı; küçük on-prem modellerle düşük birim maliyet. Paket, Aginies platformunun Event ingestion, Clustering, Time-series tools, Staged rollout yeteneklerini kullanır ve ajan kütüphanesindeki Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyen, Cihaz filosu anomali triyajı, Kademeli OTA güncelleme dağıtımı, Soğuk zincir sapması müdahalesi senaryolarıyla canlı izlenebilir. Sorun Binlerce cihaz aynı kök nedenden yüzlerce alarm üretiyor; firmware güncellemesi tüm filoya bir kerede çıkıyor ve geri alma planı yok. Telemetri hacmi bulut model maliyetini taşınamaz kılıyor. Ajanlar ne yapar - 01Telemetri anomalileri model, firmware ve ağ geçidi bazında kümelenir; tek olay açılır. - 02Kök neden adayları firmware geçmişi ve bilinen sorunlarla puanlanır; güven skoru yazılır. - 03OTA dağıtımı %1 → %10 → %100 kademeleriyle ilerler; her kademede sağlık kapısı, düşerse otomatik geri alma. - 04Toplu cihaz komutları ve kritik segmentler filo operasyon lideri ya da ürün sahibinin onayından geçer. Paket kapsamı - Telemetri anomalilerinin kümelenmesi ve olaya dönüştürülmesi - Aşamalı OTA dağıtımı; pilot dalga, durdurma kriterleri, geri alma - Uzaktan tanı ve komut; saha ekibiyle entegrasyon - Filo sağlık panosu ve bilgi tabanı güncellemeleri Başlamak için gerekenler - IoT platformu veya MQTT aracısına erişim; cihaz yönetimi API’si - Zaman serisi veritabanı ve filo kaydı - OTA imzalama ve dağıtım politikası Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Cihaz Destek SüpervizörüFilo Sağlık SüpervizörüDağıtım SüpervizörüSoğuk Zincir Süpervizörü Uzman ajanlar Diyalog AjanıTelemetri AjanıTanı AjanıSaha AjanıSinyal AjanıOperasyon AjanıPlan AjanıSağlık AjanıDağıtım AjanıSensör AjanıKalite Ajanı Modeller (Hub kataloğu) aginies-talay-2aginies-sunkar-1aginies-tulpar-4 Dokunulan sistemler WhatsApp BusinessDevice Cloud (MQTT)Fleet RegistryOTA ServiceField ServiceIoT Platform (MQTT / Digital Twin)Time-series DBKnown Issues (vector)ITSM / IncidentSlackCI/CD (signed artefacts)Device Management / OTAObservability (metrics)IoT PlatformWMS / Lot TrackingStability Rules (vector)WhatsApp Business API Kanallar whatsappwebapibatchapp Simülasyon hacimleri 58.020çalıştırma / ay %95,1başarı oranı 6,9 smedyan süre $0,0025tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Alarm gürültüsü olay sayısına iner; ekip gerçek soruna bakar - Güvenli, kanıtlı ve geri alınabilir firmware dağıtımı - Küçük on-prem modellerle düşük birim maliyet Kullanılan yetenekler Event ingestionClusteringTime-series toolsStaged rolloutKill-switchOn-prem models Platformu keşfet → Ölçülen hedefler | Alarmdan olaya kümelenme | %80 gürültü azalması | | OTA geri alma oranı | %1 altı | | Saha ziyareti gerektiren arıza | %35 azalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyenToplu OTA (>100 cihaz) veya garanti dışı cihaz değişimi → ürün sahibi onayı - Cihaz filosu anomali triyajıToplu cihaz komutu > 100 cihaz → filo operasyon lideri onayı - Kademeli OTA güncelleme dağıtımıKritik filo segmentine dağıtım → ürün sahibi onayı; sağlık kapısı düşerse otomatik geri alma - Soğuk zincir sapması müdahalesiÜrün taşıma ya da imha kararı → kalite yöneticisi onayı - paketToplu OTA ve kritik segmentler her zaman ürün sahibi onayında - paketCihaz komutları imzalı ve iz kayıtlı; emniyet kritik cihazlarda yalnızca okuma Yol haritası - 1 · Hat entegrasyonu · 2 haftaJira, Git, CI/CD ya da cihaz yönetimi bağlanır; izin modeli ve kapsam sınırları yazılır. - 2 · Pilot · 4–6 haftaBir takım ya da bir cihaz segmentiyle; her çıktı insan incelemesinden geçer, kabul oranı ölçülür. - 3 · ÖlçekDiğer takımlar ve segmentler; otomatik adımlar genişler, üretim ve mimari kararlar insanda kalır. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları IoT ve Bağlı CihazlarHITL ### Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyen Müşteri "cihaz çalışmıyor" der; telemetri okunur, uzaktan tanı konur, güncelleme gönderilir, gerekirse teknisyen doğru parçayla atanır. 37.000 çalıştırma / ay %94,1 başarı 5 ajan IoT ve Bağlı CihazlarHITL ### Cihaz filosu anomali triyajı Binlerce cihazdan gelen telemetri anomalileri kümelenir, kök neden aday listesi çıkar, doğru ekibe tek olay olarak düşer. 18.000 çalıştırma / ay %96,9 başarı 4 ajan IoT ve Bağlı CihazlarHITL ### Kademeli OTA güncelleme dağıtımı Firmware dağıtımı %1 → %10 → %100 kademelerinde ilerler; her kademede sağlık kapısı, kritik filolarda insan onayı. 320 çalıştırma / ay %98,1 başarı 4 ajan IoT ve Bağlı CihazlarHITL ### Soğuk zincir sapması müdahalesi Sıcaklık sapması anında değerlendirilir, ürün riski hesaplanır, saha ekibi sevk edilir, kalite kararı kayıt altına alınır. 2.700 çalıştırma / ay %97,4 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # KVKK kişisel veri koruma URL: https://www.aginies.com/tr/cozumler/kvkk-personal-data-protection Language: tr Summary: Kişisel veri sözleşmelerde, e-postalarda ve çağrı kayıtlarında dağınık; nerede ne olduğunu kimse tam bilmiyor. Modele giden her belge ve konuşma, maskelenmediğinde KVKK ve GDPR riski taşıyor. Kurumsal Fonksiyonlar · Veri koruma ve uyum # KVKK kişisel veri koruma Belge ve görüşmelerde kişisel veri tespiti, maskeleme ve denetime hazır veri haritası. Kısaca KVKK kişisel veri koruma, Kurumsal Fonksiyonlar sektörü için “veri koruma ve uyum” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: KVKK Kişisel Veri Tespiti ajanı belge ve görüşmelerde TCKN, IBAN, sağlık ve iletişim verisini bulur ve sınıflandırır; KVKK Belge Maskeleme ajanı veri modele ulaşmadan önce alanı bağlama uygun şekilde maskeler ya da anonimleştirir. Beklenen sonuçlar: kişisel veri modele ulaşmadan maskelenir; denetime hazır, güncel veri envanteri; belge ve çağrı arşivindeki uyum riski görünür hâle gelir. Paket, Aginies platformunun PII detection, PII masking, Entity extraction, On-prem deployment yeteneklerini kullanır ve ajan kütüphanesindeki Underwriting belge çıkarımı, Konut kredisi belge incelemesi, Çağrı merkezi kalite izleme, E-posta sınıflandırma ve ek kontrolü senaryolarıyla canlı izlenebilir. Sorun Kişisel veri sözleşmelerde, e-postalarda ve çağrı kayıtlarında dağınık; nerede ne olduğunu kimse tam bilmiyor. Modele giden her belge ve konuşma, maskelenmediğinde KVKK ve GDPR riski taşıyor. Ajanlar ne yapar - 01KVKK Kişisel Veri Tespiti ajanı belge ve görüşmelerde TCKN, IBAN, sağlık ve iletişim verisini bulur ve sınıflandırır. - 02KVKK Belge Maskeleme ajanı veri modele ulaşmadan önce alanı bağlama uygun şekilde maskeler ya da anonimleştirir. - 03Tespitler veri haritasına işlenir: hangi veri, hangi sistemde, hangi amaçla, ne kadar süre. - 04Politika dışı paylaşım ve silme talepleri uyum ekibine düşer; her karar denetim iziyle saklanır. Paket kapsamı - Belge ve konuşmalarda kişisel veri tespiti ve sınıflandırması - Maskeleme, anonimleştirme ve saklama süresi kuralları - Veri sahibi başvurularının (erişim, silme) kayıt ve yanıt süreci - Veri envanteri ve işleme kaydı güncellemeleri Başlamak için gerekenler - Belge depoları ve iletişim kanallarına okuma erişimi - Veri envanteri ve saklama politikası - Veri sorumlusu ve KVKK ekibinden onay kuralları Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Underwriting SüpervizörüKredi Dosya SüpervizörüKalite SüpervizörüPosta Süpervizörü Uzman ajanlar Belge AjanıTıbbi Değerlendirme AjanıUnderwriting AjanıÇıkarım AjanıGelir AjanıDeğerleme AjanıTranskript AjanıPuanlama AjanıKoçluk AjanıSınıflandırma AjanıEk Kontrol AjanıYönlendirme Ajanı Modeller (Hub kataloğu) aginies-konrul-1aginies-tulpar-4aginies-talay-2aginies-voice-tr-v3aginies-sunkar-1 Dokunulan sistemler Underwriting WorkbenchDocument Store (S3)Vector StoreEmail (Exchange)Loan Origination SystemCredit Bureau (KKB)Genesys SIPQM Scorecard / Case ManagementQA Rubric Vector StoreWorkdayExchange / M365 MailMailbox Rules & Checklists Vector StoreServiceNow Kanallar emailwebbatchbranchvoice Simülasyon hacimleri 187.000çalıştırma / ay %98,0başarı oranı 7,8 smedyan süre $0,0017tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Kişisel veri modele ulaşmadan maskelenir - Denetime hazır, güncel veri envanteri - Belge ve çağrı arşivindeki uyum riski görünür hâle gelir Kullanılan yetenekler PII detectionPII maskingEntity extractionOn-prem deploymentData map & lineageHuman-in-the-loopAudit trail Platformu keşfet → Ölçülen hedefler | Kişisel veri tespit doğruluğu | %97 ve üzeri | | Veri sahibi başvurusuna yanıt süresi | Yasal sürenin yarısı | | Manuel maskeleme iş yükü | %80 azalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Underwriting belge çıkarımıEk prim ≥ %50 veya teminat > 5.000.000 TL → kıdemli underwriter onayı - Konut kredisi belge incelemesiKredi tutarı > 3.000.000 TL veya DTI > %45 → kredi tahsis uzmanı onayı - Çağrı merkezi kalite izlemePuan < 60 veya uyum ihlali → takım lideri incelemesi - paketSilme işlemi ve dış paylaşım her zaman veri sorumlusu onayında - paketTespit modeli kurum içinde çalışır; veri Türkiye dışına çıkmaz Yol haritası - 1 · Kural ve kontrol tasarımı · 2–3 haftaMevzuat maddeleri kontrol setine çevrilir, karar matrisi uyum ekibiyle onaylanır, iz gereksinimleri yazılır. - 2 · Gölge dönem · 4 haftaAjan bulgu ve öneri üretir, uzman karar verir; uyumsuzluk oranı ve gerekçe kalitesi ölçülür. - 3 · Denetime hazır işletimKontrol kanıtları otomatik toplanır; denetim raporları ve istisna defteri iç denetime açılır. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları SigortaHITL ### Underwriting belge çıkarımı Sağlık raporu ve tahliller okunur, kılavuza göre değerlendirilir; underwriter yalnızca eşik üstünü görür. 4.800 çalıştırma / ay %94,9 başarı 4 ajan BankacılıkHITL ### Konut kredisi belge incelemesi Bordro, SGK dökümü ve ekspertiz otomatik okunur; gelir, LTV ve eksik belge kontrolü danışman beklemeden biter. 6.200 çalıştırma / ay %94,7 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Çağrı merkezi kalite izleme Örneklem değil, tüm çağrılar rubriğe göre puanlanır; uyum ihlali ve koçluk fırsatı sabah liderin önünde. 90.000 çalıştırma / ay %98,8 başarı 4 ajan Kurumsal Fonksiyonlar ### E-posta sınıflandırma ve ek kontrolü Ortak kutuya gelen her e-posta sınıflandırılır, ekler listeyle karşılaştırılır, eksikler anında sorulur. 86.000 çalıştırma / ay %97,6 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Lojistik kontrol kulesi URL: https://www.aginies.com/tr/cozumler/logistics-control-tower Language: tr Summary: Gecikme müşteri aradığında öğreniliyor; alternatif rota ve taşıyıcı seçimi elle, SLA tazmini ay sonunda tartışılıyor, gümrük belgeleri son dakikada hazırlanıyor. Lojistik · Operasyonel otomasyon # Lojistik kontrol kulesi Sapmayı erken yakalayan, rotayı değiştiren, müşteriyi bilgilendiren ve SLA’yı hesaplayan operasyon ajanları. Kısaca Lojistik kontrol kulesi, Lojistik sektörü için “operasyonel otomasyon” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Telematik ve taşıyıcı verisinden sapma tahmin edilir; SLA etkisi sevkiyat bazında hesaplanır; rota, aktarma ve taşıyıcı seçenekleri maliyetle karşılaştırılır; eşik içindeyse yeniden rezervasyon yapılır. Beklenen sonuçlar: gecikmeler saatler önce bilinir ve yönetilir; zamanında teslimat ve müşteri bilgilendirmesinde artış; SLA finansı şeffaf ve denetlenebilir. Paket, Aginies platformunun TMS & telematics tools, Carrier APIs, Supervisor · sub-agents, Human-in-the-loop yeteneklerini kullanır ve ajan kütüphanesindeki Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA, Sevkiyat istisna yönetimi, Gümrük belgesi hazırlama, Soğuk zincir sapması müdahalesi senaryolarıyla canlı izlenebilir. Sorun Gecikme müşteri aradığında öğreniliyor; alternatif rota ve taşıyıcı seçimi elle, SLA tazmini ay sonunda tartışılıyor, gümrük belgeleri son dakikada hazırlanıyor. Ajanlar ne yapar - 01Telematik ve taşıyıcı verisinden sapma tahmin edilir; SLA etkisi sevkiyat bazında hesaplanır. - 02Rota, aktarma ve taşıyıcı seçenekleri maliyetle karşılaştırılır; eşik içindeyse yeniden rezervasyon yapılır. - 03Alıcı ve gönderen yeni ETA ve seçeneklerle bilgilendirilir; soğuk zincir sapmaları ayrı protokolle yönetilir. - 04Gümrük belgeleri sevkiyat verisinden hazırlanır; SLA kaydı, tazmin ve taşıyıcı rücusu otomatik açılır. Paket kapsamı - Sevkiyat sapmalarının erken tahmini; telematik ve taşıyıcı verisiyle - Rota ve taşıyıcı alternatifleri, yeniden rezervasyon - Gümrük belgesi hazırlığı ve soğuk zincir sapma müdahalesi - Müşteri bilgilendirme, SLA kaydı ve tazmin hesabı Başlamak için gerekenler - TMS, telematik ve taşıyıcı API erişimi - SLA sözleşmeleri ve tazmin kuralları - Gümrük ve ürün gereksinimleri (soğuk zincir toleransları) Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Kontrol Kulesi SüpervizörüSevkiyat SüpervizörüGümrük SüpervizörüSoğuk Zincir Süpervizörü Uzman ajanlar Tahmin AjanıRota AjanıMüşteri AjanıFinans Ajanıİstisna AjanıEtki AjanıAksiyon AjanıTarife AjanıBelge AjanıMüşavir AjanıSensör AjanıKalite AjanıSaha Ajanı Modeller (Hub kataloğu) aginies-tulpar-4aginies-talay-2aginies-sunkar-1 Dokunulan sistemler TMSTelematicsCarrier APICustomer PortalERP BillingTMS (Transport Management)Salesforce CRMWhatsApp Business APISAP S/4HANAWMSTariff & Regulation Vector StoreCustoms Broker PortalDocument Store (S3)IoT PlatformWMS / Lot TrackingStability Rules (vector)Field Service Kanallar apiemailwhatsappbatchapp Simülasyon hacimleri 115.800çalıştırma / ay %93,8başarı oranı 6,9 smedyan süre $0,0032tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Gecikmeler saatler önce bilinir ve yönetilir - Zamanında teslimat ve müşteri bilgilendirmesinde artış - SLA finansı şeffaf ve denetlenebilir Kullanılan yetenekler TMS & telematics toolsCarrier APIsSupervisor · sub-agentsHuman-in-the-loopCustomer portal integrationFull audit trail Platformu keşfet → Ölçülen hedefler | Gecikme tespit öncülü | Saatler önce | | Zamanında teslimat | +5–10 puan | | Müşteri arama hacmi (nerede kargom) | %50 azalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLARota maliyet artışı > %15 veya tazmin > 10.000 TL → operasyon müdürü onayı - Sevkiyat istisna yönetimiEk maliyet > 1.000 € → sevk şefi onayı - Gümrük belgesi hazırlamaGTİP güven skoru < 0,90 veya vergi farkı > 2.000 € → gümrük uzmanı onayı - Soğuk zincir sapması müdahalesiÜrün taşıma ya da imha kararı → kalite yöneticisi onayı - paketMaliyet artışı ve tazmin eşik üstü operasyon müdürü onayında - paketHer rota kararı maliyet ve SLA etkisiyle kayıt altında Yol haritası - 1 · Süreç haritası · 2 haftaSinyal kaynakları, karar noktaları ve iş emri sistemleri haritalanır; otomatik karar sınırı ve onay zinciri belirlenir. - 2 · Pilot · 4 haftaTek hat, tek bölge ya da tek süreçte; ajan öneriyi hazırlar, insan uygular; sapma ve kazanç ölçülür. - 3 · Otonom işletimEşik altı kararlar otomatik, üstü onaya; vardiya raporları ve SLA panoları devrededir. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları LojistikHITL ### Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA Telematik sinyali gecikmeyi erken yakalar; alternatif rota ve taşıyıcı seçilir, müşteri bilgilendirilir, SLA tazmini hesaplanır. 96.000 çalıştırma / ay %93,4 başarı 5 ajan LojistikHITL ### Sevkiyat istisna yönetimi Gecikme ya da hasar geldiğinde etki hesaplanır, taşıyıcı yeniden planlanır, müşteri daha sormadan bilgilenir. 12.800 çalıştırma / ay %95,5 başarı 4 ajan LojistikHITL ### Gümrük belgesi hazırlama GTİP kodu, fatura ve çeki listesi dakikalarda hazırlanır; düşük güvenli satırlar gümrük uzmanına gider. 4.300 çalıştırma / ay %94,7 başarı 4 ajan IoT ve Bağlı CihazlarHITL ### Soğuk zincir sapması müdahalesi Sıcaklık sapması anında değerlendirilir, ürün riski hesaplanır, saha ekibi sevk edilir, kalite kararı kayıt altına alınır. 2.700 çalıştırma / ay %97,4 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Üretim ve bakım operasyonları URL: https://www.aginies.com/tr/cozumler/manufacturing-operations Language: tr Summary: Duruş alarmı düştüğünde kök neden tahmini tecrübeye kalıyor, parça stoğu ayrı sistemde, plan değişikliği telefonla yapılıyor; garanti ve tedarikçi faturaları haftalarca bekliyor. Üretim & Otomotiv · Operasyonel otomasyon # Üretim ve bakım operasyonları Hat duruşundan iş emrine, bakım talebinden garanti dosyasına: MES, CMMS ve ERP üzerinde karar veren ajanlar. Kısaca Üretim ve bakım operasyonları, Üretim & Otomotiv sektörü için “operasyonel otomasyon” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Sensör geçmişi ve bakım kayıtlarından kök neden sıralanır; parça stoğu kontrol edilip rezerve edilir; CMMS iş emri prosedür ve tahmini süreyle açılır; üretim planı simüle edilip vardiya amirine sunulur. Beklenen sonuçlar: plansız duruş süresinde ölçülebilir azalma; alarmdan iş emrine dakikalar; garanti ve fatura süreçlerinde haftalar yerine günler. Paket, Aginies platformunun MES / SCADA tools, CMMS & ERP integration, Supervisor · sub-agents, Human-in-the-loop yeteneklerini kullanır ve ajan kütüphanesindeki Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna, Bakım talebi triyajı, Garanti talebi işleme, Tedarikçi faturası üçlü eşleştirme senaryolarıyla canlı izlenebilir. Sorun Duruş alarmı düştüğünde kök neden tahmini tecrübeye kalıyor, parça stoğu ayrı sistemde, plan değişikliği telefonla yapılıyor; garanti ve tedarikçi faturaları haftalarca bekliyor. Ajanlar ne yapar - 01Sensör geçmişi ve bakım kayıtlarından kök neden sıralanır; parça stoğu kontrol edilip rezerve edilir. - 02CMMS iş emri prosedür ve tahmini süreyle açılır; üretim planı simüle edilip vardiya amirine sunulur. - 03Bakım talepleri triyajlanır, kestirimci uyarılar planlı bakıma dönüştürülür. - 04Garanti talepleri ve tedarikçi faturaları belgeden okunur, eşleştirilir, istisnalar insana gider. Paket kapsamı - Hat duruşu müdahalesi: kök neden, parça, iş emri, plan güncellemesi - Bakım talebi triyajı ve kestirimci bakım uyarıları - Garanti talebi işleme ve tedarikçi fatura eşleştirme - Vardiya raporları ve OEE panosu Başlamak için gerekenler - MES, SCADA historian ve CMMS API erişimi - Yedek parça stoğu ve tedarikçi kayıtları (ERP) - Bakım prosedürleri ve emniyet kuralları Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Üretim SüpervizörüBakım SüpervizörüGaranti SüpervizörüFatura Süpervizörü Uzman ajanlar Tanı AjanıParça AjanıPlan AjanıRapor AjanıSinyal AjanıSaha Notu AjanıTeşhis AjanıPlanlama AjanıTalep AjanıDoğrulama AjanıÖdeme AjanıYakalama AjanıEşleştirme AjanıKayıt Ajanı Modeller (Hub kataloğu) aginies-tulpar-4aginies-sunkar-1aginies-talay-2aginies-voice-tr-v3aginies-konrul-1 Dokunulan sistemler MESSCADA HistorianCMMSERP Spare PartsShift Channel (Teams)SCADA / HistorianCMMS (Maximo)SAP S/4HANA MMManuals & Work Order Vector StoreDealer PortalSAP S/4HANAWarranty Policy Vector StoreDocument Store (S3)Exchange / M365 MailTolerance Policy Vector Store Kanallar apiappvoicewebemailbatch Simülasyon hacimleri 36.100çalıştırma / ay %96,3başarı oranı 7,1 smedyan süre $0,005tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Plansız duruş süresinde ölçülebilir azalma - Alarmdan iş emrine dakikalar - Garanti ve fatura süreçlerinde haftalar yerine günler Kullanılan yetenekler MES / SCADA toolsCMMS & ERP integrationSupervisor · sub-agentsHuman-in-the-loopDocument extraction (Vision)Full audit trail Platformu keşfet → Ölçülen hedefler | Plansız duruş süresi | %20–30 azalma | | Alarmdan iş emrine süre | Dakikalar | | Garanti talebi işleme süresi | %60 kısalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporunaPlan değişikliği > 2 saat veya 50.000 TL üzeri acil satın alma → vardiya amiri onayı - Bakım talebi triyajıPlanlı duruş > 4 saat → vardiya amiri onayı - Garanti talebi işlemeGaranti talebi > 5.000 € → bölge garanti müdürü onayı - Tedarikçi faturası üçlü eşleştirmeFiyat sapması > %2 veya > 1.000 TL → satın alma sorumlusu onayı - paketÜretim planı değişikliği ve acil satın alma vardiya amiri onayında - paketAjan ekipmana komut vermez; öneri üretir, bakım ekibi uygular Yol haritası - 1 · Süreç haritası · 2 haftaSinyal kaynakları, karar noktaları ve iş emri sistemleri haritalanır; otomatik karar sınırı ve onay zinciri belirlenir. - 2 · Pilot · 4 haftaTek hat, tek bölge ya da tek süreçte; ajan öneriyi hazırlar, insan uygular; sapma ve kazanç ölçülür. - 3 · Otonom işletimEşik altı kararlar otomatik, üstü onaya; vardiya raporları ve SLA panoları devrededir. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Üretim & OtomotivHITL ### Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna MES alarmı düşer; kök neden tarihsel veriyle bulunur, yedek parça kontrol edilir, iş emri ve plan güncellemesi dakikalar içinde hazırdır. 3.100 çalıştırma / ay %92,2 başarı 5 ajan Üretim & OtomotivHITL ### Bakım talebi triyajı Sensör alarmı ve teknisyen sesli notu tek iş emrine dönüşür; yedek parça ve güvenli duruş penceresi hazırdır. 3.400 çalıştırma / ay %95,2 başarı 5 ajan Üretim & OtomotivHITL ### Garanti talebi işleme Bayi talebi dakikalarda kapsam, parça ve işçilik yönünden doğrulanır; eşik üstü tutar bölge müdürüne gider. 8.600 çalıştırma / ay %96,3 başarı 4 ajan Üretim & OtomotivHITL ### Tedarikçi faturası üçlü eşleştirme Fatura, sipariş ve mal kabul SAP'de otomatik eşleşir; yalnızca tolerans dışı kalemler satın almacıya gider. 21.000 çalıştırma / ay %97,1 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Üye işyeri ve ödeme operasyonları URL: https://www.aginies.com/tr/cozumler/merchant-and-payment-operations Language: tr Summary: Ödeme kuruluşunda işyeri destek hattı, risk ekibi ve mutabakat ekibi aynı işyeri için ayrı ayrı çalışıyor; KYB günler sürüyor, hakediş soruları çağrı merkezini dolduruyor, istisnalar ay sonuna kalıyor. Finansal Teknolojiler · Operasyonel otomasyon # Üye işyeri ve ödeme operasyonları İşyeri edinimi, POS ve hakediş desteği, mutabakat istisnaları ve işlem izleme tek masada. Kısaca Üye işyeri ve ödeme operasyonları, Finansal Teknolojiler sektörü için “operasyonel otomasyon” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: KYB: ticaret sicili, yetkili ve yaptırım kontrolü, risk sınıfı ve sözleşme tek akışta; operasyon asistanı WhatsApp ve portalda POS tanısı, hakediş dökümü, iade ve limit taleplerini çözer. Beklenen sonuçlar: işyeri sorunları ilk temasta, tek kanalda çözülür; KYB saatler içinde; risk ekibi yalnızca istisnalarda; mutabakat istisnaları ay sonunu beklemez. Paket, Aginies platformunun KYB & sanctions tools, Terminal & settlement tools, Policy engine, Human-in-the-loop yeteneklerini kullanır ve ajan kütüphanesindeki Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit, Üye işyeri KYB onboarding, Ödeme mutabakatı istisnaları, AML işlem izleme alarm triyajı senaryolarıyla canlı izlenebilir. Sorun Ödeme kuruluşunda işyeri destek hattı, risk ekibi ve mutabakat ekibi aynı işyeri için ayrı ayrı çalışıyor; KYB günler sürüyor, hakediş soruları çağrı merkezini dolduruyor, istisnalar ay sonuna kalıyor. Ajanlar ne yapar - 01KYB: ticaret sicili, yetkili ve yaptırım kontrolü, risk sınıfı ve sözleşme tek akışta. - 02Operasyon asistanı WhatsApp ve portalda POS tanısı, hakediş dökümü, iade ve limit taleplerini çözer. - 03Mutabakat istisnaları eşleştirilir ve açıklanır; AML uyarıları kanıtla risk ekibine sunulur. - 04BNPL ve taksit kararları politika motoruyla; sınır dışı talep her zaman insana. Paket kapsamı - Üye işyeri edinimi (KYB), risk sınıfı ve sözleşme - Operasyon asistanı: POS, hakediş, iade ve limit tek diyalogda - Mutabakat istisnaları ve işlem izleme (AML) uyarıları - BNPL ve taksit kararları politika motoruyla Başlamak için gerekenler - Terminal yönetimi, hakediş defteri ve iade servisi API erişimi - Risk politikası ve TCMB/MASAK yükümlülüklerine göre kurallar - WhatsApp Business ve işyeri portalı kanalları Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Üye İşyeri SüpervizörüÜye İşyeri Edinim SüpervizörüMutabakat SüpervizörüAML Süpervizörü Uzman ajanlar Diyalog AjanıPOS AjanıOperasyon AjanıRisk AjanıKimlik AjanıOrtaklık AjanıSözleşme AjanıEşleştirme Ajanıİnceleme AjanıDüzeltme AjanıZenginleştirme AjanıTarama AjanıAnlatı AjanıDosyalama Ajanı Modeller (Hub kataloğu) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-konrul-1 Dokunulan sistemler WhatsApp BusinessTerminal ManagementSettlement LedgerRefund ServiceTicketingMerchant PortalTrade Registry (MERSİS)Sanctions Screeninge-SignEmailPSP / Acquirer APICard NetworkCore LedgerCase ManagementMASAK Reporting Portal Kanallar whatsappwebvoiceemailbatchapi Simülasyon hacimleri 74.520çalıştırma / ay %94,0başarı oranı 8,1 smedyan süre $0,0043tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - İşyeri sorunları ilk temasta, tek kanalda çözülür - KYB saatler içinde; risk ekibi yalnızca istisnalarda - Mutabakat istisnaları ay sonunu beklemez Kullanılan yetenekler KYB & sanctions toolsTerminal & settlement toolsPolicy engineHuman-in-the-loopWhatsApp BusinessFull audit trail Platformu keşfet → Ölçülen hedefler | İşyeri destek talebi ilk temasta çözüm | %70 ve üzeri | | KYB tamamlanma süresi | Günlerden saatlere | | Mutabakat istisnası kapanış süresi | %60 kısalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Üye işyeri operasyon asistanı: POS, hakediş, iade ve limitİade > 25.000 TL veya limit artışı > %50 → risk ekibi onayı - Üye işyeri KYB onboardingRisk sınıfı yüksek veya UBO bulanık eşleşme > 0,65 → uyum yöneticisi onayı - Ödeme mutabakatı istisnalarıTek fark > 10.000 TL veya toplam > 50.000 TL → finans müdürü onayı ile ayarlama kaydı - AML işlem izleme alarm triyajıHer alarm → uyum analisti kararı zorunlu; ajan tek başına STR dosyalamaz, vaka kapatmaz - paketYüksek iade, limit artışı ve AML uyarısı her zaman risk ekibinde - paketTopluluk bulutu veya kurum veri merkezi; ödeme mevzuatına uyumlu kurulum Yol haritası - 1 · Süreç haritası · 2 haftaSinyal kaynakları, karar noktaları ve iş emri sistemleri haritalanır; otomatik karar sınırı ve onay zinciri belirlenir. - 2 · Pilot · 4 haftaTek hat, tek bölge ya da tek süreçte; ajan öneriyi hazırlar, insan uygular; sapma ve kazanç ölçülür. - 3 · Otonom işletimEşik altı kararlar otomatik, üstü onaya; vardiya raporları ve SLA panoları devrededir. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Finansal TeknolojilerHITL ### Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit İşyeri WhatsApp'tan yazar; POS arızası, hakediş sorgusu, iade ve limit artışı tek diyalogda çözülür. 64.000 çalıştırma / ay %93,8 başarı 5 ajan Finansal TeknolojilerHITL ### Üye işyeri KYB onboarding Sicil, gerçek faydalanıcı ve yaptırım kontrolü tek akışta; düşük riskli işyeri aynı gün ödeme almaya başlar. 1.400 çalıştırma / ay %93,4 başarı 5 ajan Finansal TeknolojilerHITL ### Ödeme mutabakatı istisnaları PSP mutabakat dosyası ana defterle eşlenir; küçük farklar otomatik düzeltilir, büyükleri finans onayına gider. 620 çalıştırma / ay %95,1 başarı 4 ajan Finansal TeknolojilerHITL ### AML işlem izleme alarm triyajı Alarm zenginleştirilir, anlatı taslağı hazır gelir; kararı analist verir, bildirim dakikalarda dosyalanır. 8.500 çalıştırma / ay %95,8 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Operasyonel otomasyon ve talep yönetimi URL: https://www.aginies.com/tr/cozumler/operational-automation-demand Language: tr Summary: Talepler e-posta, portal ve çağrı üzerinden dağınık geliyor; her biri elle okunup ekibe atanıyor. Yönlendirme hatası ve tekrar iş, süreç süresini ve maliyeti büyütüyor. Kurumsal Fonksiyonlar · Operasyonel otomasyon # Operasyonel otomasyon ve talep yönetimi Talepleri sınıflandıran, yönlendiren ve cevaplayan ajanlar; hızlı süreç, düşük maliyet. Kısaca Operasyonel otomasyon ve talep yönetimi, Kurumsal Fonksiyonlar sektörü için “operasyonel otomasyon” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Gelen talep kanaldan bağımsız sınıflandırılır: konu, aciliyet ve sorumlu ekip belirlenir; dinamik yönlendirme talebi doğru kuyruğa ya da ajana atar; eksik bilgi talep sahibinden istenir. Beklenen sonuçlar: talep çözüm süresi kısalır, birikim erir; talep başına operasyon maliyetinde düşüş; ekipler tekrar işten kurtulur, istisnaya odaklanır. Paket, Aginies platformunun Dynamic agent routing, Intent detection, Email service integrations, Ticketing tools (MCP) yeteneklerini kullanır ve ajan kütüphanesindeki Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna, Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA, BT yardım masası L1, Bakım talebi triyajı senaryolarıyla canlı izlenebilir. Sorun Talepler e-posta, portal ve çağrı üzerinden dağınık geliyor; her biri elle okunup ekibe atanıyor. Yönlendirme hatası ve tekrar iş, süreç süresini ve maliyeti büyütüyor. Ajanlar ne yapar - 01Gelen talep kanaldan bağımsız sınıflandırılır: konu, aciliyet ve sorumlu ekip belirlenir. - 02Dinamik yönlendirme talebi doğru kuyruğa ya da ajana atar; eksik bilgi talep sahibinden istenir. - 03Standart talepler (şifre, durum sorgusu, form) kurum bilgi kaynağı ve araçlarla anında çözülür. - 04SLA riski ve tekrar eden talep desenleri panoda görünür; kritik kararlar insana kalır. Paket kapsamı - Talep ve bilet akışının sınıflandırılması, önceliklendirilmesi ve atanması - Standart taleplerin (erişim, şifre, form, bilgi) uçtan uca çözümü - Onay zincirleri ve SLA takibi; gecikme uyarıları - Bilgi tabanı ve otomasyon araçlarıyla ilk seviye destek Başlamak için gerekenler - ITSM / bilet sistemi API erişimi - Hizmet kataloğu, SLA ve onay kuralları - Kimlik yönetimi ve otomasyon araçlarına yetkili erişim Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Üretim SüpervizörüKontrol Kulesi SüpervizörüYardım Masası SüpervizörüBakım Süpervizörü Uzman ajanlar Tanı AjanıParça AjanıPlan AjanıRapor AjanıTahmin AjanıRota AjanıMüşteri AjanıFinans AjanıNiyet AjanıErişim AjanıKayıt AjanıSinyal AjanıSaha Notu AjanıTeşhis AjanıPlanlama Ajanı Modeller (Hub kataloğu) aginies-tulpar-4aginies-sunkar-1aginies-talay-2aginies-voice-tr-v3 Dokunulan sistemler MESSCADA HistorianCMMSERP Spare PartsShift Channel (Teams)TMSTelematicsCarrier APICustomer PortalERP BillingActive Directory / Entra IDServiceNowMicrosoft TeamsSCADA / HistorianCMMS (Maximo)SAP S/4HANA MMManuals & Work Order Vector Store Kanallar apiappemailwhatsappvoiceweb Simülasyon hacimleri 149.500çalıştırma / ay %95,0başarı oranı 7,9 smedyan süre $0,0023tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Talep çözüm süresi kısalır, birikim erir - Talep başına operasyon maliyetinde düşüş - Ekipler tekrar işten kurtulur, istisnaya odaklanır Kullanılan yetenekler Dynamic agent routingIntent detectionEmail service integrationsTicketing tools (MCP)Agent templatingAudit trail logging Platformu keşfet → Ölçülen hedefler | Otomatik kapatılan talep payı | %45–60 | | Ortalama çözüm süresi | %50 kısalma | | SLA ihlali | %60 azalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporunaPlan değişikliği > 2 saat veya 50.000 TL üzeri acil satın alma → vardiya amiri onayı - Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLARota maliyet artışı > %15 veya tazmin > 10.000 TL → operasyon müdürü onayı - BT yardım masası L1Finans veya üretim grubuna erişim → grup sahibi onayı - Bakım talebi triyajıPlanlı duruş > 4 saat → vardiya amiri onayı - paketYetki yükseltme ve üretim sistemlerine erişim her zaman onaylı - paketHer otomatik işlem bilet üzerinde kim/ne/ne zaman ile kayıtlı Yol haritası - 1 · Süreç haritası · 2 haftaSinyal kaynakları, karar noktaları ve iş emri sistemleri haritalanır; otomatik karar sınırı ve onay zinciri belirlenir. - 2 · Pilot · 4 haftaTek hat, tek bölge ya da tek süreçte; ajan öneriyi hazırlar, insan uygular; sapma ve kazanç ölçülür. - 3 · Otonom işletimEşik altı kararlar otomatik, üstü onaya; vardiya raporları ve SLA panoları devrededir. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Üretim & OtomotivHITL ### Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna MES alarmı düşer; kök neden tarihsel veriyle bulunur, yedek parça kontrol edilir, iş emri ve plan güncellemesi dakikalar içinde hazırdır. 3.100 çalıştırma / ay %92,2 başarı 5 ajan LojistikHITL ### Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA Telematik sinyali gecikmeyi erken yakalar; alternatif rota ve taşıyıcı seçilir, müşteri bilgilendirilir, SLA tazmini hesaplanır. 96.000 çalıştırma / ay %93,4 başarı 5 ajan Kurumsal FonksiyonlarHITL ### BT yardım masası L1 Şifre sıfırlama ve erişim talepleri sohbette çözülür; finans grubu erişimi grup sahibine Teams'te sorulur. 47.000 çalıştırma / ay %98,5 başarı 4 ajan Üretim & OtomotivHITL ### Bakım talebi triyajı Sensör alarmı ve teknisyen sesli notu tek iş emrine dönüşür; yedek parça ve güvenli duruş penceresi hazırdır. 3.400 çalıştırma / ay %95,2 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Kamu alımı ve ihale uyumu URL: https://www.aginies.com/tr/cozumler/public-procurement-compliance Language: tr Summary: İhale dosyalarında yüzlerce belge ve onlarca şartname maddesi elle karşılaştırılıyor; eksikler komisyon toplantısında fark ediliyor, itiraz riski artıyor. Kamu ve Hükümet · Veri koruma ve uyum # Kamu alımı ve ihale uyumu Teklif dosyaları şartnameyle karşılaştırılır; komisyon hazır bulgu listesi üzerinde karar verir. Kısaca Kamu alımı ve ihale uyumu, Kamu ve Hükümet sektörü için “veri koruma ve uyum” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Teklif belgeleri sınıflandırılır ve okunur: teminat, yeterlik, iş deneyimi, teknik ekler, e-imza; her teklif idari ve teknik şartname maddeleriyle eşlenir; uygunsuzluk adayları işaretlenir. Beklenen sonuçlar: komisyon hazırlığı günlerden saatlere iner; tutarlı, gerekçeli ve itiraza dayanıklı değerlendirme; her bulgu kaynak belgeye bağlı, denetime hazır. Paket, Aginies platformunun Vision extraction, Document classification, Clause matching, Explainable findings yeteneklerini kullanır ve ajan kütüphanesindeki Kamu ihalesi teklif dosyası ön kontrolü, Sözleşme madde çıkarımı, Tedarikçi güvenlik ve uyum değerlendirmesi senaryolarıyla canlı izlenebilir. Sorun İhale dosyalarında yüzlerce belge ve onlarca şartname maddesi elle karşılaştırılıyor; eksikler komisyon toplantısında fark ediliyor, itiraz riski artıyor. Ajanlar ne yapar - 01Teklif belgeleri sınıflandırılır ve okunur: teminat, yeterlik, iş deneyimi, teknik ekler, e-imza. - 02Her teklif idari ve teknik şartname maddeleriyle eşlenir; uygunsuzluk adayları işaretlenir. - 03Kontrol listesi ve tutanak taslağı madde numaralarıyla oluşturulur. - 04Uygunluk ve eleme kararı ihale komisyonunda kalır; belirsiz maddeler komisyona işaretlenir. Paket kapsamı - Teklif dosyalarının şartname ve mevzuatla karşılaştırılması - Eksik belge, uygunsuz teklif ve çıkar çatışması bulguları - Komisyon için bulgu listesi ve gerekçe taslağı - İhale takvimi ve süre takibi Başlamak için gerekenler - EKAP ve kurum ihale dosyalarına okuma erişimi - Şartname şablonları ve değerlendirme kriterleri - Komisyon üyeleri ve onay akışı Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler İhale SüpervizörüSözleşme SüpervizörüTedarikçi Güvenlik Süpervizörü Uzman ajanlar Belge AjanıŞartname AjanıTutanak AjanıMadde AjanıPlaybook AjanıCLM AjanıDeğerlendirme AjanıDosya Ajanı Modeller (Hub kataloğu) aginies-konrul-1aginies-tulpar-4aginies-talay-2 Dokunulan sistemler e-Procurement PlatformSpecification Vector StoreDocument StoreCLM (Contract Lifecycle Management)Legal Playbook Vector StoreDocument Store (S3)Exchange / M365 MailSupplier Management (SRM)Security Requirements (vector)E-mail Kanallar batchwebemailapi Simülasyon hacimleri 2.680çalıştırma / ay %95,1başarı oranı 8 smedyan süre $0,0177tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Komisyon hazırlığı günlerden saatlere iner - Tutarlı, gerekçeli ve itiraza dayanıklı değerlendirme - Her bulgu kaynak belgeye bağlı, denetime hazır Kullanılan yetenekler Vision extractionDocument classificationClause matchingExplainable findingsHuman-in-the-loopOn-prem models Platformu keşfet → Ölçülen hedefler | Dosya ön inceleme süresi | %70 kısalma | | Gözden kaçan eksik belge | Sıfıra yakın | | Komisyon kararına itiraz | Azalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Kamu ihalesi teklif dosyası ön kontrolüUygunluk ve eleme kararı her zaman ihale komisyonunda; ajan yalnızca hazırlar - Sözleşme madde çıkarımıYüksek riskli sapma veya sözleşme değeri > 500.000 € → hukuk müşaviri onayı - Tedarikçi güvenlik ve uyum değerlendirmesiNitelendirme kararı her zaman tesis güvenlik müdüründe; ajan yalnızca hazırlar - paketElenme ve ihale kararı yalnızca komisyonda; ajan bulgu üretir - paketHer bulgu belge ve madde referansıyla; denetime hazır Yol haritası - 1 · Kural ve kontrol tasarımı · 2–3 haftaMevzuat maddeleri kontrol setine çevrilir, karar matrisi uyum ekibiyle onaylanır, iz gereksinimleri yazılır. - 2 · Gölge dönem · 4 haftaAjan bulgu ve öneri üretir, uzman karar verir; uyumsuzluk oranı ve gerekçe kalitesi ölçülür. - 3 · Denetime hazır işletimKontrol kanıtları otomatik toplanır; denetim raporları ve istisna defteri iç denetime açılır. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Kamu ve HükümetHITL ### Kamu ihalesi teklif dosyası ön kontrolü Teklif dosyaları şartnameyle karşılaştırılır, eksik ve uygunsuz belgeler listelenir; değerlendirme kararı komisyonun. 380 çalıştırma / ay %94,8 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Sözleşme madde çıkarımı Sözleşme maddeleri çıkarılır, hukuk playbook'u ile karşılaştırılır; sapmalar redline önerisiyle avukata gider. 1.400 çalıştırma / ay %95,1 başarı 4 ajan Savunma SanayiiHITL ### Tedarikçi güvenlik ve uyum değerlendirmesi Tedarikçi anketleri ve belgeleri okunur, tesis güvenlik gereklilikleriyle karşılaştırılır, eksikler listelenir; nitelendirme kararı güvenlik müdürünün. 900 çalıştırma / ay %95,2 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Satış ve lead yönetimi URL: https://www.aginies.com/tr/cozumler/sales-lead-management Language: tr Summary: Lead’ler CRM’de birikiyor ama satış ekibi hangisinin sıcak olduğunu bilmiyor; kampanyalar herkese aynı gidiyor. Fiyat ve hediye önerileri sezgiyle veriliyor, fırsatlar soğuyor. Perakende & E-ticaret · Müşteri etkileşimi ve deneyim # Satış ve lead yönetimi Doğru teklif, doğru zamanda: lead skorlama, kampanya ve fiyat önerisi. Kısaca Satış ve lead yönetimi, Perakende & E-ticaret sektörü için “müşteri etkileşimi ve deneyim” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Lead AI web, form ve CRM sinyallerini birleştirir; her lead’i niyet ve uyum skoruyla önceliklendirir; kampanya ajanı segment ve geçmiş davranışa göre kişiye özel teklif ve hediye önerisi üretir. Beklenen sonuçlar: satış ekibi en yüksek dönüşüm ihtimalli lead’lere odaklanır; kişiye özel kampanya ile dönüşüm ve sepet büyüklüğü artar; fırsat soğumadan temas: skorlanan her lead aynı gün aranır. Paket, Aginies platformunun Lead scoring, Recommendation engine, CRM tools (MCP), Multi-source RAG yeteneklerini kullanır ve ajan kütüphanesindeki Satış fırsatı nitelendirme, Tarife yükseltme danışmanı, Yorum yanıtlama ve kalite eskalasyonu, Ürün soru-cevap asistanı senaryolarıyla canlı izlenebilir. Sorun Lead’ler CRM’de birikiyor ama satış ekibi hangisinin sıcak olduğunu bilmiyor; kampanyalar herkese aynı gidiyor. Fiyat ve hediye önerileri sezgiyle veriliyor, fırsatlar soğuyor. Ajanlar ne yapar - 01Lead AI web, form ve CRM sinyallerini birleştirir; her lead’i niyet ve uyum skoruyla önceliklendirir. - 02Kampanya ajanı segment ve geçmiş davranışa göre kişiye özel teklif ve hediye önerisi üretir. - 03Fiyat analizi ajanı rakip ve stok verisini okur; marj kuralları içinde fiyat önerir. - 04Sıcak lead’ler temsilciye konuşma özetiyle devredilir; indirim eşiği üzeri teklifler onaya gider. Paket kapsamı - Web, e-posta ve WhatsApp’tan gelen lead’lerin skorlanması ve zenginleştirilmesi - CRM’e kayıt, temsilci ataması ve ilk temas taslağı - Teklif ve paket önerisi; ürün bilgi tabanından kaynaklı - Takip hatırlatmaları ve kaybolan fırsat analizi Başlamak için gerekenler - CRM (Salesforce, HubSpot, Dynamics) API erişimi - Skorlama kriterleri ve bölge/temsilci atama kuralları - Güncel ürün, fiyat ve kampanya bilgisi Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Satış SüpervizörüTarife SüpervizörüMüşteri Sesi SüpervizörüAlışveriş Süpervizörü Uzman ajanlar Zenginleştirme AjanıSkorlama AjanıCRM AjanıKullanım AjanıTeklif AjanıProvizyon AjanıSınıflandırma AjanıYanıt AjanıKalite AjanıNiyet AjanıKatalog Ajanı Modeller (Hub kataloğu) aginies-tulpar-4aginies-talay-2aginies-sunkar-1 Dokunulan sistemler HubSpotEnrichment APIMicrosoft TeamsBilling (Amdocs)Product Catalogue / Offer EngineHLR / ProvisioningWhatsApp Business APISMS GatewayReview Platform APIShopify / OMSJira (QA)Vector StorePIM / CatalogueWeb Chat Widget Kanallar webapiemailappwhatsappvoice Simülasyon hacimleri 213.600çalıştırma / ay %97,8başarı oranı 5,4 smedyan süre $0,0007tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Satış ekibi en yüksek dönüşüm ihtimalli lead’lere odaklanır - Kişiye özel kampanya ile dönüşüm ve sepet büyüklüğü artar - Fırsat soğumadan temas: skorlanan her lead aynı gün aranır Kullanılan yetenekler Lead scoringRecommendation engineCRM tools (MCP)Multi-source RAGHuman-in-the-loopSession memory Platformu keşfet → Ölçülen hedefler | Lead’e ilk temas süresi | Saatlerden dakikalara | | Nitelikli lead dönüşümü | +10–20 puan | | CRM veri eksiksizliği | %95 ve üzeri | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Satış fırsatı nitelendirmeKurumsal (> 1.000 çalışan) veya rakip şüphesi → SDR lideri kabulü - Tarife yükseltme danışmanıTaahhütlü veya indirimli özel teklif → kampanya yöneticisi onayı - Yorum yanıtlama ve kalite eskalasyonu1 yıldızlı yorum yanıtları ve kusur eskalasyonları → topluluk yöneticisi onayı - paketFiyat ve indirim teklifi yetki matrisine göre; sınır dışı teklif insana - paketİzin dışı pazarlama teması yapılmaz; KVKK izin kaydı her temasta kontrol edilir Yol haritası - 1 · Keşif ve tasarım · 2–3 haftaEn sık 20 niyet ve işlem seçilir, kimlik ve onay eşikleri bilgi güvenliği ile yazılır, kanal ve sistem entegrasyonları listelenir. - 2 · Pilot · 4–6 haftaTek kanalda, sınırlı müşteri grubuyla; test korpusu koşturulur, gölge modda insan temsilci izler, eşikler gerçek trafikle ayarlanır. - 3 · YaygınlaştırmaDiğer kanallar ve işlemler eklenir; panolar, SIEM aktarımı ve kill-switch devredilir; aylık model ve kural gözden geçirmesi başlar. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Kurumsal FonksiyonlarHITL ### Satış fırsatı nitelendirme Gelen talep saniyeler içinde zenginleştirilir, skorlanır ve CRM'e işlenir; SDR aramaya bağlam notuyla başlar. 15.600 çalıştırma / ay %97,9 başarı 4 ajan TelekomHITL ### Tarife yükseltme danışmanı Kullanım analiz edilir, doğru tarife karşılaştırmalı sunulur; onayla değişiklik anında provizyonlanır. 52.000 çalıştırma / ay %95,8 başarı 4 ajan Perakende & E-ticaretHITL ### Yorum yanıtlama ve kalite eskalasyonu Yorumlar sınıflanır, marka tonunda yanıtlanır; tekrarlayan kusurlar iade verisiyle QA'e Jira kaydı olur. 26.000 çalıştırma / ay %97,2 başarı 4 ajan Perakende & E-ticaret ### Ürün soru-cevap asistanı Beden, uyumluluk ve özellik soruları katalog ve iade verisiyle yanıtlanır; stok ve fiyat doğrulanır. 120.000 çalıştırma / ay %98,7 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # SDLC ajan ekibi URL: https://www.aginies.com/tr/cozumler/sdlc-agent-team Language: tr Summary: Yazılım ekipleri talep, analiz, tasarım, kodlama, test ve dağıtım arasında el değiştirirken zaman kaybediyor. Kod üreten asistanlar var; ama iş kalemi açılışından üretime kadar süreci sahiplenen, onay noktaları belli bir ekip yok. Yazılım Geliştirme · Yazılım geliştirme ve teknoloji verimliliği # SDLC ajan ekibi PM, analist, mimar, geliştirici ve test ajanları CI/CD hattında birlikte çalışır. Kısaca SDLC ajan ekibi, Yazılım Geliştirme sektörü için “yazılım geliştirme ve teknoloji verimliliği” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Proje Yöneticisi Ajanı talepleri önceliklendirir, Analist Ajanı kabul kriterlerini yazar, Mimar Ajanı tasarımı ve etkiyi çıkarır; geliştirici Ajanı dalda kodu yazar; Test Ajanı birim ve entegrasyon testlerini üretip çalıştırır, kapsamı raporlar. Beklenen sonuçlar: talepten üretime geçiş süresi günlerden saatlere iner; standart, ölçülebilir kalite: kapsam, güvenlik ve regresyon her PR’da; mühendisler yönlendirme ve onaya odaklanır, tekrar işi ajanlar üstlenir. Paket, Aginies platformunun Supervisor · sub-agents, Tools / MCP (Git, CI, Jira), Human-in-the-loop, Stateful execution yeteneklerini kullanır ve ajan kütüphanesindeki Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR, Özellik teslimat hattı, CI/CD sürüm hattı, Olay müdahalesi ve acil düzeltme, Kod incelemesi ve güvenlik kapısı senaryolarıyla canlı izlenebilir. Sorun Yazılım ekipleri talep, analiz, tasarım, kodlama, test ve dağıtım arasında el değiştirirken zaman kaybediyor. Kod üreten asistanlar var; ama iş kalemi açılışından üretime kadar süreci sahiplenen, onay noktaları belli bir ekip yok. Ajanlar ne yapar - 01Proje Yöneticisi Ajanı talepleri önceliklendirir, Analist Ajanı kabul kriterlerini yazar, Mimar Ajanı tasarımı ve etkiyi çıkarır. - 02Geliştirici Ajanı dalda kodu yazar; Test Ajanı birim ve entegrasyon testlerini üretip çalıştırır, kapsamı raporlar. - 03CI/CD hattında SAST, bağımlılık taraması ve staging duman testleri otomatik koşar; üretim dağıtımı yayın yöneticisi onayına gider. - 04Her adım Jira, Git ve CI kayıtlarıyla izlenir; kod incelemesi ve üretim kararı her zaman insanda kalır. Paket kapsamı - PM, analist, mimar, geliştirici ve test ajanları; Jira, Git ve CI/CD üzerinde - Gereksinimden PR’a: analiz notu, tasarım kararı, kod, test, inceleme - Code Aginies ile IDE içi tamamlama; kod kurum dışına çıkmaz - Olay müdahalesi ve hotfix hattı; sürüm notları Başlamak için gerekenler - Jira, Git ve CI/CD için servis hesapları ve izin modeli - Kodlama standartları, inceleme kuralları ve tanım-of-done - Kapsam sınırları: hangi depolar, hangi ortamlar Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Modernizasyon SüpervizörüTeslimat SüpervizörüSürüm SüpervizörüOlay Süpervizörüİnceleme Süpervizörü Uzman ajanlar Analiz AjanıTest AjanıGeliştirici Ajanıİnceleme AjanıAnalist AjanıMimar AjanıDerleme AjanıGüvenlik AjanıSürüm AjanıPost-mortem Ajanıİnceleyici Ajanı Modeller (Hub kataloğu) aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Dokunulan sistemler JiraGit (self-hosted)CI PipelineStatic AnalysisGitHubCI Runner (GitHub Actions)ConfluenceSonarQube + SnykKubernetes / ArgoCDSlackObservability (Grafana / Datadog)SonarQubeDependency Scanner (Snyk) Kanallar apibatchweb Simülasyon hacimleri 8.570çalıştırma / ay %92,6başarı oranı 16 smedyan süre $0,0144tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Talepten üretime geçiş süresi günlerden saatlere iner - Standart, ölçülebilir kalite: kapsam, güvenlik ve regresyon her PR’da - Mühendisler yönlendirme ve onaya odaklanır, tekrar işi ajanlar üstlenir Kullanılan yetenekler Supervisor · sub-agentsTools / MCP (Git, CI, Jira)Human-in-the-loopStateful executionFull audit trailPrompt versioning Platformu keşfet → Ölçülen hedefler | İlk incelemede kabul edilen PR payı | %60–75 | | Gereksinimden PR’a geçen süre | %40–50 kısalma | | Test kapsamı | +15–25 puan | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Legacy modül modernizasyonu: analiz, test, yeniden yazım, PRÜretim şeması değişikliği veya kapsam dışı dosya → mimar onayı; PR birleştirme her zaman insanda - Özellik teslimat hattıHer PR → en az 1 insan kod incelemesi onayı; ajan main'e doğrudan yazamaz - CI/CD sürüm hattıHer üretim dağıtımı → sürüm yöneticisi onayı zorunlu; kritik CVE varsa kapı otomatik kapanır - Olay müdahalesi ve acil düzeltmeHer acil üretim dağıtımı → nöbetçi mühendis onayı; yama PR'ı da insan incelemesinden geçer - Kod incelemesi ve güvenlik kapısıHer PR → CODEOWNERS insan onayı; yüksek SAST bulgusu veya kapsam < %80 kapıyı kapatır - paketPR birleştirme ve üretim dağıtımı her zaman insan onayında - paketAjan yalnızca atanan depo ve dallarda yazar; sır tarayıcı her PR’da Yol haritası - 1 · Hat entegrasyonu · 2 haftaJira, Git, CI/CD ya da cihaz yönetimi bağlanır; izin modeli ve kapsam sınırları yazılır. - 2 · Pilot · 4–6 haftaBir takım ya da bir cihaz segmentiyle; her çıktı insan incelemesinden geçer, kabul oranı ölçülür. - 3 · ÖlçekDiğer takımlar ve segmentler; otomatik adımlar genişler, üretim ve mimari kararlar insanda kalır. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Yazılım GeliştirmeHITL ### Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR Jira'daki bir epic'ten başlar; eski modül okunur, davranışı testlerle sabitlenir, yeniden yazılır ve kanıtlı bir PR açılır. 1.900 çalıştırma / ay %90,4 başarı 5 ajan Yazılım GeliştirmeHITL ### Özellik teslimat hattı Bir Jira epic'i ajanlarla kriter, tasarım, kod ve teste dönüşür; insan yalnızca PR'ı inceler ve onaylar. 1.200 çalıştırma / ay %91,2 başarı 5 ajan Yazılım GeliştirmeHITL ### CI/CD sürüm hattı Derleme, test, tarama ve staging otomatik; üretime yalnızca sürüm yöneticisi onayıyla ArgoCD ile çıkılır. 950 çalıştırma / ay %94,7 başarı 5 ajan Yazılım GeliştirmeHITL ### Olay müdahalesi ve acil düzeltme Alarm gelir, hata yeniden üretilir, yama ve regresyon hazır; nöbetçi onaylar, post-mortem taslağı bekler. 320 çalıştırma / ay %90,5 başarı 5 ajan Yazılım GeliştirmeHITL ### Kod incelemesi ve güvenlik kapısı Her PR ajanlarca incelenir, taranır, kapsamı ölçülür; geliştirici düzeltir, insan yalnızca temiz PR'ı onaylar. 4.200 çalıştırma / ay %93,6 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Abone hizmetleri ve şebeke desteği URL: https://www.aginies.com/tr/cozumler/telecom-subscriber-care Language: tr Summary: Abone aynı çağrıda fatura, paket ve arıza konuşuyor; IVR ağacı ve temsilci ekranları bunu tek oturumda çözemiyor, teknisyen çoğu zaman gereksiz gidiyor. Telekom · Müşteri etkileşimi ve deneyim # Abone hizmetleri ve şebeke desteği Fatura, paket, taahhüt ve arıza işlemleri sesli ve yazılı kanalda, BSS ve şebeke sistemlerinde. Kısaca Abone hizmetleri ve şebeke desteği, Telekom sektörü için “müşteri etkileşimi ve deneyim” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Niyetler ayrıştırılır; fatura kalemleri açıklanır, düzeltme ve paket değişikliği BSS’te uygulanır; hat ve modem tanısı uzaktan yapılır; firmware ve yeniden başlatma ile çözülmezse teknisyen atanır. Beklenen sonuçlar: çok konulu çağrılar tek oturumda, temsilcisiz kapanır; gereksiz teknisyen ziyaretlerinde belirgin azalma; daha kısa işlem süresi, daha yüksek memnuniyet. Paket, Aginies platformunun BSS / OSS tools (MCP), SIP voice, Supervisor · sub-agents, Policy engine yeteneklerini kullanır ve ajan kütüphanesindeki Abone hizmet asistanı: fatura, paket, arıza ve taahhüt, Fatura itirazı çözümü, Tarife yükseltme danışmanı, SIM değişimi dolandırıcılık kontrolü, Şebeke kesintisi iletişimi senaryolarıyla canlı izlenebilir. Sorun Abone aynı çağrıda fatura, paket ve arıza konuşuyor; IVR ağacı ve temsilci ekranları bunu tek oturumda çözemiyor, teknisyen çoğu zaman gereksiz gidiyor. Ajanlar ne yapar - 01Niyetler ayrıştırılır; fatura kalemleri açıklanır, düzeltme ve paket değişikliği BSS’te uygulanır. - 02Hat ve modem tanısı uzaktan yapılır; firmware ve yeniden başlatma ile çözülmezse teknisyen atanır. - 03Taahhüt ve cayma koşulları hesaplanır, uygun teklif sunulur; iptal eğilimi elde tutma ekibine iletilir. - 04SIM değişimi ve şüpheli talepler dolandırıcılık kontrolünden geçer; kesinti bilgisi proaktif gönderilir. Paket kapsamı - Fatura itirazı, paket değişikliği ve taahhüt işlemleri BSS üzerinde - Hat ve modem tanısı, uzaktan onarım, teknisyen ataması - SIM değişimi dolandırıcılık kontrolü ve kesinti iletişimi - IVR, uygulama ve WhatsApp’ta aynı asistan; temsilciye sıcak aktarım Başlamak için gerekenler - BSS/faturalama ve OSS/şebeke sistemleri API erişimi - Tarife, taahhüt ve cayma kuralları - IVR/SIP ve dijital kanallar entegrasyonu Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Abone Hizmet SüpervizörüFatura SüpervizörüTarife SüpervizörüSIM Güvenlik SüpervizörüKesinti İletişim Süpervizörü Uzman ajanlar Diyalog AjanıHesap AjanıŞebeke AjanıTaahhüt AjanıNiyet AjanıFatura Analiz AjanıDüzeltme AjanıKullanım AjanıTeklif AjanıProvizyon AjanıRisk AjanıDoğrulama AjanıTespit AjanıEtki Ajanıİletişim Ajanı Modeller (Hub kataloğu) aginies-talay-2aginies-sunkar-1aginies-tulpar-4 Dokunulan sistemler IVR / SIPBSS / BillingOSS NetworkCRMSMS GatewayBilling (Amdocs)Usage Mediation / CDR StoreProduct Catalogue / Offer EngineHLR / ProvisioningWhatsApp Business APISalesforce CRMBank SIM-Swap Signal APIServiceNowNetwork OSS Kanallar voiceappwhatsappbranchapismsbatch Simülasyon hacimleri 592.400çalıştırma / ay %95,0başarı oranı 7,3 smedyan süre $0,0016tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Çok konulu çağrılar tek oturumda, temsilcisiz kapanır - Gereksiz teknisyen ziyaretlerinde belirgin azalma - Daha kısa işlem süresi, daha yüksek memnuniyet Kullanılan yetenekler BSS / OSS tools (MCP)SIP voiceSupervisor · sub-agentsPolicy engineHuman-in-the-loopFull audit trail Platformu keşfet → Ölçülen hedefler | Temsilcisiz çözülen talep payı | %60–70 | | Teknisyen gerekmeden kapanan arıza | %35–45 | | Ortalama işlem süresi | %40 kısalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Abone hizmet asistanı: fatura, paket, arıza ve taahhütTaahhüt iptali, 500 TL üzeri fatura düzeltmesi veya teknisyen önceliği → temsilci onayı - Fatura itirazı çözümüİyi niyet kredisi > 75 TL → ekip lideri onayı - Tarife yükseltme danışmanıTaahhütlü veya indirimli özel teklif → kampanya yöneticisi onayı - SIM değişimi dolandırıcılık kontrolüRisk skoru ≥ 70 veya 72 saat içinde kimlik bilgisi değişikliği → dolandırıcılık masası onayı - Şebeke kesintisi iletişimiEtkilenen abone > 10.000 veya P1 kesinti → şebeke operasyon yöneticisi onayı - paketTaahhüt iptali, yüksek fatura düzeltmesi ve SIM değişimi insan onayında - paketAbone verisi BTK yükümlülüklerine göre işlenir; her işlem iz bırakır Yol haritası - 1 · Keşif ve tasarım · 2–3 haftaEn sık 20 niyet ve işlem seçilir, kimlik ve onay eşikleri bilgi güvenliği ile yazılır, kanal ve sistem entegrasyonları listelenir. - 2 · Pilot · 4–6 haftaTek kanalda, sınırlı müşteri grubuyla; test korpusu koşturulur, gölge modda insan temsilci izler, eşikler gerçek trafikle ayarlanır. - 3 · YaygınlaştırmaDiğer kanallar ve işlemler eklenir; panolar, SIEM aktarımı ve kill-switch devredilir; aylık model ve kural gözden geçirmesi başlar. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları TelekomHITL ### Abone hizmet asistanı: fatura, paket, arıza ve taahhüt Abone arar ya da yazar; fatura itirazı, paket değişikliği ve hat arızası aynı oturumda BSS ve şebeke sistemlerinde çözülür. 410.000 çalıştırma / ay %94,3 başarı 5 ajan TelekomHITL ### Fatura itirazı çözümü Fatura kalemleri ve kullanım kayıtları saniyede açıklanır; iyi niyet kredisi politikaya göre uygulanır. 88.000 çalıştırma / ay %96,3 başarı 4 ajan TelekomHITL ### Tarife yükseltme danışmanı Kullanım analiz edilir, doğru tarife karşılaştırmalı sunulur; onayla değişiklik anında provizyonlanır. 52.000 çalıştırma / ay %95,8 başarı 4 ajan TelekomHITL ### SIM değişimi dolandırıcılık kontrolü Her SIM değişimi provizyondan önce risk skorlanır; şüpheli talepler bekletilir ve bankalara sinyal iletilir. 41.000 çalıştırma / ay %97,7 başarı 4 ajan TelekomHITL ### Şebeke kesintisi iletişimi Kesinti kaydı açılınca etkilenen aboneler bulunur, mesaj yazılıp onaylanır; müşteri aramadan bilgilendirilir. 1.400 çalıştırma / ay %98,5 başarı 4 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Underwriting ve yenileme operasyonları URL: https://www.aginies.com/tr/cozumler/underwriting-and-renewal Language: tr Summary: Yenileme dönemi geldiğinde teklifler elle hazırlanıyor, belgeler eksik geliyor, underwriter basit dosyalarla vakit kaybediyor; müşteri geç aranınca poliçe rakibe gidiyor. Sigorta · Finansal analiz ve operasyonel zekâ # Underwriting ve yenileme operasyonları Teklif belgeleri okunur, risk değişkenleri çıkarılır, yenileme teklifi kurallar içinde hazırlanır. Kısaca Underwriting ve yenileme operasyonları, Sigorta sektörü için “finansal analiz ve operasyonel zekâ” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Teklif ve yenileme belgelerinden risk değişkenleri çıkarılır, eksikler müşteriden istenir; fiyatlama motoru ve underwriting kurallarıyla teklif hazırlanır; yetki limiti içindeyse gönderilir. Beklenen sonuçlar: yenilemelerin büyük kısmı otomatik, zamanında teklif; underwriter zamanı istisnai risklere; yenileme oranında ölçülebilir artış. Paket, Aginies platformunun Document extraction (Vision), Pricing engine tools, Policy engine, Human-in-the-loop yeteneklerini kullanır ve ajan kütüphanesindeki Poliçe yenileme teklifi, Underwriting belge çıkarımı, Sesli hasar ihbarı (FNOL) senaryolarıyla canlı izlenebilir. Sorun Yenileme dönemi geldiğinde teklifler elle hazırlanıyor, belgeler eksik geliyor, underwriter basit dosyalarla vakit kaybediyor; müşteri geç aranınca poliçe rakibe gidiyor. Ajanlar ne yapar - 01Teklif ve yenileme belgelerinden risk değişkenleri çıkarılır, eksikler müşteriden istenir. - 02Fiyatlama motoru ve underwriting kurallarıyla teklif hazırlanır; yetki limiti içindeyse gönderilir. - 03Yenileme hatırlatması, teklif ve eksik belge iletişimi müşteriyle tercih ettiği kanalda. - 04Limit üstü ve istisnai riskler karar paketiyle underwriter kuyruğuna düşer. Paket kapsamı - Teklif ve yenileme belgelerinin çıkarımı; risk değişkenleri - Yenileme teklifi ve fiyat önerisi; underwriting kurallarıyla - Müşteri iletişimi: yenileme hatırlatma, teklif, eksik belge - Underwriter için karar paketi ve istisna kuyruğu Başlamak için gerekenler - Poliçe yönetimi ve fiyatlama motoru API erişimi - Underwriting kuralları ve yetki limitleri - Örnek teklif ve yenileme belgeleri Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Yenileme SüpervizörüUnderwriting SüpervizörüHasar Süpervizörü Uzman ajanlar Fiyatlama AjanıTeklif AjanıTanzim AjanıBelge AjanıTıbbi Değerlendirme AjanıUnderwriting AjanıKonuşma Ajanıİhbar AjanıPoliçe AjanıHasar Kayıt Ajanı Modeller (Hub kataloğu) aginies-talay-2aginies-tulpar-4aginies-sunkar-1aginies-konrul-1aginies-voice-tr-v3 Dokunulan sistemler Policy Admin SystemRating EngineWhatsApp Business APIEmail (Exchange)Underwriting WorkbenchDocument Store (S3)Vector StoreGenesys SIPGuidewire ClaimCenter Kanallar batchwhatsappemailappwebvoice Simülasyon hacimleri 60.800çalıştırma / ay %95,2başarı oranı 6,2 smedyan süre $0,003tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Yenilemelerin büyük kısmı otomatik, zamanında teklif - Underwriter zamanı istisnai risklere - Yenileme oranında ölçülebilir artış Kullanılan yetenekler Document extraction (Vision)Pricing engine toolsPolicy engineHuman-in-the-loopMulti-channel messagingFull audit trail Platformu keşfet → Ölçülen hedefler | Otomatik yenilenen poliçe payı | %60–75 | | Teklif hazırlama süresi | %60 kısalma | | Yenileme oranı | +5–8 puan | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Poliçe yenileme teklifiPrim artışı > %20 veya indirim > %10 → yenileme ekibi lideri onayı - Underwriting belge çıkarımıEk prim ≥ %50 veya teminat > 5.000.000 TL → kıdemli underwriter onayı - Sesli hasar ihbarı (FNOL)Yaralanma veya rezerv > 250.000 TL → hasar uzmanına sıcak aktarım - paketYetki limiti üstü fiyat ve istisnai risk her zaman underwriter onayında - paketFiyat önerisi gerekçesi ve kullanılan değişkenler poliçe dosyasında Yol haritası - 1 · Belge ve kural seti · 2 haftaBelge tipleri, alan sözlüğü, eşleştirme ve tolerans kuralları finans ekibiyle yazılır; örnek set etiketlenir. - 2 · Paralel koşum · 4 haftaAjan ve mevcut süreç aynı belgeleri işler; fark raporu her hafta incelenir, doğruluk eşiği karşılanınca geçiş. - 3 · Kapanış entegrasyonuAy sonu takvimine bağlanır; istisna kuyruğu, denetim izi ve mutabakat raporları iç denetime açılır. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları SigortaHITL ### Poliçe yenileme teklifi Poliçeler yeniden fiyatlanır, sert artışlar onaya gider; müşteri tek dokunuşla onaylar, poliçe anında kesilir. 22.000 çalıştırma / ay %94,4 başarı 4 ajan SigortaHITL ### Underwriting belge çıkarımı Sağlık raporu ve tahliller okunur, kılavuza göre değerlendirilir; underwriter yalnızca eşik üstünü görür. 4.800 çalıştırma / ay %94,9 başarı 4 ajan SigortaHITL ### Sesli hasar ihbarı (FNOL) Müşteri arar ve anlatır; poliçe plakadan bulunur, teminat kontrol edilir, dosya açılır ve eksper atanır. 34.000 çalıştırma / ay %95,8 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Enerji müşteri operasyonları URL: https://www.aginies.com/tr/cozumler/utility-customer-operations Language: tr Summary: Müşteri kanalları ile şebeke sistemleri birbirinden kopuk: kesintide yüzlerce arama, tahmini faturada itiraz yığını, çatı GES başvurusunda mühendis zamanı belge kontrolüne gidiyor. Enerji · Müşteri etkileşimi ve deneyim # Enerji müşteri operasyonları Kesinti, sayaç, bağlantı ve talep yanıtı: şebeke verisiyle konuşan müşteri ajanları. Kısaca Enerji müşteri operasyonları, Enerji sektörü için “müşteri etkileşimi ve deneyim” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Kesinti bildirimleri SCADA ve fider topolojisiyle kümelenir; etkilenen herkes ekip yola çıkmadan bilgilenir; sayaç fotoğrafı görüntü modeliyle okunur, mevzuat kontrol edilir, fatura düzeltilir. Beklenen sonuçlar: kesinti anında çağrı hacmi düşer, müşteri memnuniyeti artar; fatura itirazları ve bağlantı başvuruları saatler içinde sonuçlanır; esneklik portföyü gelire dönüşür. Paket, Aginies platformunun Voice (STT/TTS), Vision extraction, Multi-source RAG, REST / MCP tools yeteneklerini kullanır ve ajan kütüphanesindeki Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda, Kesinti bildirimi yönetimi, Sayaç okuma itirazı, Çatı GES bağlantı başvurusu senaryolarıyla canlı izlenebilir. Sorun Müşteri kanalları ile şebeke sistemleri birbirinden kopuk: kesintide yüzlerce arama, tahmini faturada itiraz yığını, çatı GES başvurusunda mühendis zamanı belge kontrolüne gidiyor. Ajanlar ne yapar - 01Kesinti bildirimleri SCADA ve fider topolojisiyle kümelenir; etkilenen herkes ekip yola çıkmadan bilgilenir. - 02Sayaç fotoğrafı görüntü modeliyle okunur, mevzuat kontrol edilir, fatura düzeltilir. - 03Bağlantı başvuruları belge, kapasite ve mevzuat açısından ön değerlendirilir; görüş mühendisin imzasıyla çıkar. - 04Talep yanıtı olaylarında katılımcılar seçilir, teklif gönderilir, azaltım ölçülür, ödeme hesaplanır. Paket kapsamı - Fatura açıklaması, itiraz ve taksit; akıllı sayaç verisiyle - Kesinti bilgisi ve tahmini süre; OMS entegrasyonu - Bağlantı, abonelik ve çatı GES başvuruları - IVR, WhatsApp ve uygulama kanalı; temsilciye sıcak aktarım Başlamak için gerekenler - Faturalama (CIS) ve kesinti yönetimi (OMS) API erişimi - Akıllı sayaç veri erişimi - Tarife, taksit ve itiraz politikaları Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Abone Hizmet SüpervizörüKesinti SüpervizörüFatura İtiraz SüpervizörüBağlantı Süpervizörü Uzman ajanlar Diyalog AjanıFatura AjanıŞebeke AjanıTahsilat AjanıSes AjanıBildirim AjanıKümeleme Ajanıİletişim AjanıGörüntü Ajanıİşlem AjanıBelge AjanıMevzuat AjanıDosya Ajanı Modeller (Hub kataloğu) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-voice-tr-v3aginies-konrul-1 Dokunulan sistemler IVR / SIPBilling (CIS)Outage ManagementSmart Meter DataSMS GatewayCustomer CRM / BillingSCADA / HistorianNetwork OSSWhatsApp Business APIBilling / CIS (SAP IS-U)Meter Data ManagementRegulation Vector Store (EPDK)Connection PortalGIS / Network ModelE-mail Kanallar voicewhatsappappwebemailapi Simülasyon hacimleri 572.600çalıştırma / ay %94,8başarı oranı 6,5 smedyan süre $0,0016tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Kesinti anında çağrı hacmi düşer, müşteri memnuniyeti artar - Fatura itirazları ve bağlantı başvuruları saatler içinde sonuçlanır - Esneklik portföyü gelire dönüşür Kullanılan yetenekler Voice (STT/TTS)Vision extractionMulti-source RAGREST / MCP toolsHuman-in-the-loopBatch events Platformu keşfet → Ölçülen hedefler | Temsilcisiz çözülen çağrı payı | %55–65 | | Fatura itirazı çözüm süresi | Günlerden dakikalara | | Kesinti dönemi çağrı yükü | %50 azalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogdaTaksit > 6 ay, fatura iptali veya sayaç şüphesi → müşteri hizmetleri onayı - Kesinti bildirimi yönetimiToplu bilgilendirme > 1.000 müşteri → kontrol odası dispeçeri onayı - Sayaç okuma itirazıİade / düzeltme > 1.000 TL → faturalama ekip lideri onayı - Çatı GES bağlantı başvurusuBağlantı görüşü her zaman şebeke planlama mühendisinin imzasıyla - paketFatura iptali ve uzun taksit her zaman müşteri hizmetleri onayında - paketSayaç verisi yalnızca abonenin kendi hesabı için kullanılır Yol haritası - 1 · Keşif ve tasarım · 2–3 haftaEn sık 20 niyet ve işlem seçilir, kimlik ve onay eşikleri bilgi güvenliği ile yazılır, kanal ve sistem entegrasyonları listelenir. - 2 · Pilot · 4–6 haftaTek kanalda, sınırlı müşteri grubuyla; test korpusu koşturulur, gölge modda insan temsilci izler, eşikler gerçek trafikle ayarlanır. - 3 · YaygınlaştırmaDiğer kanallar ve işlemler eklenir; panolar, SIEM aktarımı ve kill-switch devredilir; aylık model ve kural gözden geçirmesi başlar. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları EnerjiHITL ### Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda Abone arar ya da yazar; yüksek fatura açıklanır, kesinti süresi verilir, sayaç verisi gösterilir, taksit planı kurulur. 520.000 çalıştırma / ay %94,6 başarı 5 ajan EnerjiHITL ### Kesinti bildirimi yönetimi Bildirimler fider bazında kümelenir, tek OSS kaydı açılır, etkilenen herkes ekip yola çıkmadan bilgilenir. 38.000 çalıştırma / ay %97,7 başarı 5 ajan EnerjiHITL ### Sayaç okuma itirazı Müşteri sayaç fotoğrafı yollar; endeks görüntüden okunur, tahmini fatura düzeltilir, büyük iade lidere gider. 9.200 çalıştırma / ay %94,9 başarı 4 ajan EnerjiHITL ### Çatı GES bağlantı başvurusu Başvuru belgeleri okunur, trafo kapasitesi ve mevzuat kontrol edilir, eksikler aynı gün bildirilir; bağlantı görüşü mühendisin. 5.400 çalıştırma / ay %95,3 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Ses analizi ve kalite URL: https://www.aginies.com/tr/cozumler/voice-analysis-quality Language: tr Summary: Milyonlarca dakikalık görüşme dinlenmeden arşive gidiyor; müşteri duygusu, niyet ve uyum ihlali görünmez kalıyor. Bulut STT servisleri Türkçede yetersiz, üstelik ses verisini kurum dışına çıkarıyor. Telekom · Müşteri etkileşimi ve deneyim # Ses analizi ve kalite Türkçe STT ile görüşmelerin %100’ü yazıya dökülür, puanlanır; modeller kurum içinde. Kısaca Ses analizi ve kalite, Telekom sektörü için “müşteri etkileşimi ve deneyim” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: Voice Aginies modelleri görüşmelerin %100’ünü kurum içinde Türkçe yazıya döker; kişisel veri maskelenir; her görüşmede duygu, niyet ve konu çıkarılır; müşteri kırılganlığı ve ayrılma sinyali işaretlenir. Beklenen sonuçlar: örnekleme yerine tüm görüşmelerde ölçüm; uyum ihlali ve müşteri riski aynı gün görünür; ses verisi kurum dışına çıkmaz. Paket, Aginies platformunun Speech-to-text (Turkish), On-prem models, Sentiment & intent, Compliance phrase checks yeteneklerini kullanır ve ajan kütüphanesindeki Çağrı merkezi kalite izleme, Abone hizmet asistanı: fatura, paket, arıza ve taahhüt, Diyalog bankacılığı: sesli ve yazılı işlem asistanı, Sesli hasar ihbarı (FNOL) senaryolarıyla canlı izlenebilir. Sorun Milyonlarca dakikalık görüşme dinlenmeden arşive gidiyor; müşteri duygusu, niyet ve uyum ihlali görünmez kalıyor. Bulut STT servisleri Türkçede yetersiz, üstelik ses verisini kurum dışına çıkarıyor. Ajanlar ne yapar - 01Voice Aginies modelleri görüşmelerin %100’ünü kurum içinde Türkçe yazıya döker; kişisel veri maskelenir. - 02Her görüşmede duygu, niyet ve konu çıkarılır; müşteri kırılganlığı ve ayrılma sinyali işaretlenir. - 03Kalite formu ve zorunlu uyum ifadeleri (aydınlatma, kayıt bildirimi) otomatik puanlanır, gerekçe kaydedilir. - 04Temsilci bazında koçluk önerileri üretilir; eğitim kararı ve yaptırım takım liderinde kalır. Paket kapsamı - Çağrı transkripti, duygu ve konu analizi; abone bazında geçmiş - Kampanya ve taahhüt iletişiminde uyum kontrolü - Şikâyet ve iptal sinyallerinin erken tespiti, elde tutma aksiyonu - Temsilci koçluğu ve kalite panosu Başlamak için gerekenler - Santral ve kayıt sistemine erişim; SIP veya kayıt API’si - Kampanya, taahhüt ve iptal politikaları - CRM ve faturalama ile abone eşlemesi Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Kalite SüpervizörüAbone Hizmet SüpervizörüDiyalog Bankacılığı SüpervizörüHasar Süpervizörü Uzman ajanlar Transkript AjanıPuanlama AjanıKoçluk AjanıDiyalog AjanıHesap AjanıŞebeke AjanıTaahhüt AjanıKimlik Ajanıİşlem AjanıRisk AjanıKonuşma Ajanıİhbar AjanıPoliçe AjanıHasar Kayıt Ajanı Modeller (Hub kataloğu) aginies-voice-tr-v3aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Dokunulan sistemler Genesys SIPQM Scorecard / Case ManagementQA Rubric Vector StoreWorkdayIVR / SIPBSS / BillingOSS NetworkCRMSMS GatewayMobile App ChatCore BankingCard ManagementFraud EnginePolicy Admin SystemGuidewire ClaimCenterWhatsApp Business API Kanallar voicebatchappwhatsapp Simülasyon hacimleri 720.000çalıştırma / ay %95,0başarı oranı 7,4 smedyan süre $0,0024tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Örnekleme yerine tüm görüşmelerde ölçüm - Uyum ihlali ve müşteri riski aynı gün görünür - Ses verisi kurum dışına çıkmaz Kullanılan yetenekler Speech-to-text (Turkish)On-prem modelsSentiment & intentCompliance phrase checksSensitive data maskingDashboard analytics layerCoaching insights Platformu keşfet → Ölçülen hedefler | Analiz edilen çağrı payı | %100 | | İptal sinyali tespit süresi | Aynı gün | | Kalite değerlendirme iş yükü | %75 azalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Çağrı merkezi kalite izlemePuan < 60 veya uyum ihlali → takım lideri incelemesi - Abone hizmet asistanı: fatura, paket, arıza ve taahhütTaahhüt iptali, 500 TL üzeri fatura düzeltmesi veya teknisyen önceliği → temsilci onayı - Diyalog bankacılığı: sesli ve yazılı işlem asistanıLimit artışı > %60, itiraz tutarı > 5.000 TL veya yeni alıcıya transfer → operasyon onayı - Sesli hasar ihbarı (FNOL)Yaralanma veya rezerv > 250.000 TL → hasar uzmanına sıcak aktarım - paketAbone bazlı aksiyon (teklif, arama) elde tutma ekibinin onayıyla - paketSes kayıtları ve transkriptler kişisel veri kurallarıyla saklanır Yol haritası - 1 · Keşif ve tasarım · 2–3 haftaEn sık 20 niyet ve işlem seçilir, kimlik ve onay eşikleri bilgi güvenliği ile yazılır, kanal ve sistem entegrasyonları listelenir. - 2 · Pilot · 4–6 haftaTek kanalda, sınırlı müşteri grubuyla; test korpusu koşturulur, gölge modda insan temsilci izler, eşikler gerçek trafikle ayarlanır. - 3 · YaygınlaştırmaDiğer kanallar ve işlemler eklenir; panolar, SIEM aktarımı ve kill-switch devredilir; aylık model ve kural gözden geçirmesi başlar. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları Kurumsal FonksiyonlarHITL ### Çağrı merkezi kalite izleme Örneklem değil, tüm çağrılar rubriğe göre puanlanır; uyum ihlali ve koçluk fırsatı sabah liderin önünde. 90.000 çalıştırma / ay %98,8 başarı 4 ajan TelekomHITL ### Abone hizmet asistanı: fatura, paket, arıza ve taahhüt Abone arar ya da yazar; fatura itirazı, paket değişikliği ve hat arızası aynı oturumda BSS ve şebeke sistemlerinde çözülür. 410.000 çalıştırma / ay %94,3 başarı 5 ajan BankacılıkHITL ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı Müşteri konuşur ya da yazar; fatura, limit, itiraz ve transfer aynı oturumda çekirdek sistemde tamamlanır. 186.000 çalıştırma / ay %94,7 başarı 5 ajan SigortaHITL ### Sesli hasar ihbarı (FNOL) Müşteri arar ve anlatır; poliçe plakadan bulunur, teminat kontrol edilir, dosya açılır ve eksper atanır. 34.000 çalıştırma / ay %95,8 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Sesli AI ile hasar ve talep kaydı URL: https://www.aginies.com/tr/cozumler/voice-claims-intake Language: tr Summary: Hasar bildirimleri çağrı merkezinde yığılıyor; bekleme süresi uzuyor, temsilciler formu doldururken müşteri hattı meşgul ediyor. Fırtına günlerinde kapasite yetmiyor. Sigorta · Müşteri etkileşimi ve deneyim # Sesli AI ile hasar ve talep kaydı Çağrı merkezine gelen hasar bildirimini sesli ajan alır, kaydı sistemde açar. Kısaca Sesli AI ile hasar ve talep kaydı, Sigorta sektörü için “müşteri etkileşimi ve deneyim” alanında bir Aginies çözüm paketidir. Ajanlar ne yapar: SIP entegrasyonuyla çağrı doğrudan sesli ajana düşer; akışlı konuşma tanıma gecikmesiz çalışır; poliçe, olay yeri, tarih, taraflar ve hasar tipi konuşma içinde çıkarılır ve teyit ettirilir. Beklenen sonuçlar: çağrı merkezi yükü düşer, bekleme süresi kısalır; aynı anda yüzlerce görüşme, sabit kalite; hasar kaydı ilk temasta ve tam bilgiyle açılır. Paket, Aginies platformunun Speech-to-text (streaming), Call-centre SIP integration, Intent detection, Stateful execution yeteneklerini kullanır ve ajan kütüphanesindeki Uçtan uca kasko hasarı: ihbardan ödemeye, Sesli hasar ihbarı (FNOL), Diyalog bankacılığı: sesli ve yazılı işlem asistanı, Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda senaryolarıyla canlı izlenebilir. Sorun Hasar bildirimleri çağrı merkezinde yığılıyor; bekleme süresi uzuyor, temsilciler formu doldururken müşteri hattı meşgul ediyor. Fırtına günlerinde kapasite yetmiyor. Ajanlar ne yapar - 01SIP entegrasyonuyla çağrı doğrudan sesli ajana düşer; akışlı konuşma tanıma gecikmesiz çalışır. - 02Poliçe, olay yeri, tarih, taraflar ve hasar tipi konuşma içinde çıkarılır ve teyit ettirilir. - 03Hasar kaydı poliçe sisteminde açılır, eksper ataması kurallara göre tetiklenir. - 04Hassas veri maskelenir; şüpheli desenler dolandırıcılık ekibine işaretlenir. Paket kapsamı - Çağrı merkezine gelen hasar ve talep çağrılarının sesli ajanla alınması - Poliçe eşleme, olay formu ve eksik bilgi tamamlama tek görüşmede - Hasar sistemine dosya açma ve ilk aksiyon (ikame araç, servis, eksper) - Kayıt, transkript ve özet; temsilciye sıcak aktarım Başlamak için gerekenler - SIP ile çağrı merkezi santrali entegrasyonu - Poliçe ve hasar sistemleri için API erişimi - Hasar türüne göre zorunlu alan listesi Mimari Bağlı senaryolardan türetilmiştir: bu paketi işleten süpervizörler, uzman ajanlar, modeller ve dokunulan sistemler. Süpervizörler Hasar SüpervizörüDiyalog Bankacılığı SüpervizörüAbone Hizmet Süpervizörü Uzman ajanlar İhbar AjanıFoto Ekspertiz AjanıDosya Ajanıİletişim AjanıKonuşma AjanıPoliçe AjanıHasar Kayıt AjanıDiyalog AjanıKimlik Ajanıİşlem AjanıRisk AjanıFatura AjanıŞebeke AjanıTahsilat Ajanı Modeller (Hub kataloğu) aginies-voice-tr-v3aginies-konrul-1aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Dokunulan sistemler Call Centre SIPPolicy AdminClaims SystemRepair NetworkPayment HubGenesys SIPPolicy Admin SystemGuidewire ClaimCenterWhatsApp Business APIMobile App ChatCore BankingCard ManagementFraud EngineSMS GatewayIVR / SIPBilling (CIS)Outage ManagementSmart Meter Data Kanallar voiceappwhatsapp Simülasyon hacimleri 761.000çalıştırma / ay %94,6başarı oranı 7,8 smedyan süre $0,0022tahmini maliyet / çalıştırma Bağlı senaryoların simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Sonuçlar - Çağrı merkezi yükü düşer, bekleme süresi kısalır - Aynı anda yüzlerce görüşme, sabit kalite - Hasar kaydı ilk temasta ve tam bilgiyle açılır Kullanılan yetenekler Speech-to-text (streaming)Call-centre SIP integrationIntent detectionStateful executionSensitive data masking Platformu keşfet → Ölçülen hedefler | Temsilcisiz tamamlanan ihbar payı | %60–75 | | İhbar başına ortalama süre | %40 kısalma | | Eksik bilgiyle açılan dosya | %80 azalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef, pilotun ilk haftasında birlikte sabitlenir. Yönetişim ve onay kuralları Senaryo bazlı onay kuralları ve paketin genel kontrolleri. - Uçtan uca kasko hasarı: ihbardan ödemeyeTahmin > 150.000 TL, kusur oranı belirsiz veya önceki hasarla çakışma → eksper onayı - Sesli hasar ihbarı (FNOL)Yaralanma veya rezerv > 250.000 TL → hasar uzmanına sıcak aktarım - Diyalog bankacılığı: sesli ve yazılı işlem asistanıLimit artışı > %60, itiraz tutarı > 5.000 TL veya yeni alıcıya transfer → operasyon onayı - Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogdaTaksit > 6 ay, fatura iptali veya sayaç şüphesi → müşteri hizmetleri onayı - paketYaralanma, büyük hasar ve dolandırıcılık şüphesi anında insana aktarılır - paketSes kayıtları saklama süresi ve erişimi sigorta mevzuatına göre Yol haritası - 1 · Keşif ve tasarım · 2–3 haftaEn sık 20 niyet ve işlem seçilir, kimlik ve onay eşikleri bilgi güvenliği ile yazılır, kanal ve sistem entegrasyonları listelenir. - 2 · Pilot · 4–6 haftaTek kanalda, sınırlı müşteri grubuyla; test korpusu koşturulur, gölge modda insan temsilci izler, eşikler gerçek trafikle ayarlanır. - 3 · YaygınlaştırmaDiğer kanallar ve işlemler eklenir; panolar, SIEM aktarımı ve kill-switch devredilir; aylık model ve kural gözden geçirmesi başlar. İlgili ajanlar ## Bu paketin ajan kütüphanesindeki senaryoları SigortaHITL ### Uçtan uca kasko hasarı: ihbardan ödemeye Sesli ihbar, fotoğraf ekspertizi, anlaşmalı servis ataması ve ödeme talimatı tek akışta; eksper yalnızca eşik üstünde devreye girer. 21.000 çalıştırma / ay %91,2 başarı 5 ajan SigortaHITL ### Sesli hasar ihbarı (FNOL) Müşteri arar ve anlatır; poliçe plakadan bulunur, teminat kontrol edilir, dosya açılır ve eksper atanır. 34.000 çalıştırma / ay %95,8 başarı 5 ajan BankacılıkHITL ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı Müşteri konuşur ya da yazar; fatura, limit, itiraz ve transfer aynı oturumda çekirdek sistemde tamamlanır. 186.000 çalıştırma / ay %94,7 başarı 5 ajan EnerjiHITL ### Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda Abone arar ya da yazar; yüksek fatura açıklanır, kesinti süresi verilir, sayaç verisi gösterilir, taksit planı kurulur. 520.000 çalıştırma / ay %94,6 başarı 5 ajan ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Aginies’e genel bakış · Dokümanlar URL: https://www.aginies.com/tr/dokumanlar Language: tr Summary: Kurumsal AI dönüşümü: pilottan üretime, uçtan uca. Aginies nedir, hangi sorunu çözer, nasıl çalışır ve kimin için tasarlandı. Dokümanlar Bu sayfada Aginies nedir?Sorun: üretime geçişTeklif: tek sorumlu muhatapNasıl çalışır: dört adımÜrün ailesiÜretim kanıtıKimin içinÇözüm paketleri Genel bakış # Aginies’e genel bakış Kurumsal AI dönüşümü: pilottan üretime, uçtan uca. Aginies nedir, hangi sorunu çözer, nasıl çalışır ve kimin için tasarlandı. Güncelleme: Eylül 2026 ## Aginies nedir? Aginies, kurumsal agentic AI orkestrasyon platformudur. Süpervizör ajanlar hedefi planlar, uzman ajanlar çekirdek sistemlerde aksiyon alır; her adım izlenir, gerekirse insan onayından geçer ve maliyetlendirilir. Modelden bağımsızdır; on-prem, hibrit ya da bulutta çalışır. Kurumun tüm AI senaryolarının koştuğu tek yürütme katmanıdır. Aginies iki biçimde alınır: platform olarak (kurumun kendi ajanlarını kurduğu, işlettiği yürütme katmanı) ve ajan iş gücü olarak (Workforce Aginies: iOS geliştirici, iş analisti, muhasebeci, testçi gibi rollerin ajan olarak kiralanması). İkisi aynı platformda, aynı yönetişimle çalışır; ikincisi dış kaynak insan yerine dış kaynak ajan sunar. Tezimiz basit: kurumsal AI dönüşümü pilottan üretime, uçtan uca ele alınmalıdır. Kurumların büyük çoğunluğu AI ile bir şeyler deniyor; çok azı bunu üretimde, ölçekte ve kontrollü biçimde işletebiliyor. Aginies bu boşluğu kapatmak için tasarlandı. ## Sorun: üretime geçiş Yapay zekâda asıl sorun üretime geçiştir. Pilotlar etkileyicidir; üretime geçiş yolu ise çoğunlukla tanımsızdır. Sahada dört başarısızlık kalıbı tekrar eder: - PoC çukuru: pilot çalışır, üretime geçiş yolu tanımsız kalır. - Dağınık girişimler: her birim kendi aracını dener; kurumsal bütünlük oluşmaz. - Ölçülemeyen değer: “çalışıyor” ile “işe yarıyor” arasındaki fark raporlanamaz. - Yönetişim boşluğu: güvenlik, uyum ve denetlenebilirlik sonradan akla gelir. Sonuç: bütçe harcanır, dönüşüm gerçekleşmez. ## Teklif: tek sorumlu muhatap Kurumsal AI dönüşümünü sahiplenen ve uçtan uca işleten bir çözüm ortağı. Üç bileşen tek çatı altında toplanır; hesap verebilirlik tek yerde kalır. Danışman gelip gitmez; ürün birlikte işletilir. | Bileşen | Ne sağlar | Neden önemli | | --- | --- | --- | | Yetkin ekip | Strateji, veri, mühendislik ve değişim yönetimi bir arada | Senaryo seçimi, kurulum ve işletim aynı ekipte kalır | | Aginies platformu | Kurumsal sınıf agentic orkestrasyon katmanı | Tüm AI iş yükleri tek yürütme katmanında; izlenir ve yönetilir | | Egemen altyapı | Ülke sınırları içinde GPU-as-a-Service | Veri ve model kontrolü kurumda; sermaye yatırımı olmadan ölçeklenir | ## Nasıl çalışır: dört adım Agentic AI gerçek iş süreçlerini dört adımda yürütür. Her adım aynı orkestrasyon katmanında gerçekleşir ve aynı denetim iziyle kaydedilir. - Sinyalleri anla: uygulama, API, doküman, ses ve olaylardan gelen girdiler yürütmeye hazır görevlere dönüşür. - Süpervizör planlar: karmaşık hedef yapılandırılmış adımlara bölünür; sorumluluklar uzman ajanlara dağıtılır. - Ajanlar aksiyon alır: bilgi getirilir, belgeler işlenir, araçlar çağrılır, kurumsal sistemlerde gerçek işlemler tetiklenir. - Her şey izlenir: ajan kararları, gecikme, maliyet ve araç kullanımı tam görünürlükle izlenir ve iyileştirilir. ## Ürün ailesi Platform Aginies çatı üründür; diğerleri onun modülleridir. Tüm ajanlar tek kontrol düzleminde kurulur, izlenir ve yönetilir. | Ürün | Kapsam | Öne çıkanlar | | --- | --- | --- | | Platform Aginies | Orkestrasyon çekirdeği ve kontrol düzlemi | Görsel ajan tasarımcısı, Otonom Ajan, bilgi tabanı, loglar, panolar, workspace ayrımı | | Voice Aginies | STT · TTS · ASR | Türkçe için eğitilmiş ses modelleri; çağrı merkezi SIP entegrasyonu; on-prem servis | | Code Aginies | Geliştirici ekipleri | IDE registry, IDE eklentileri, kurum içi kod tamamlama, SDLC ajan ekibi | | Hub Aginies | Model Hub · MCP Hub · proxy | İnce ayarlı kurumsal modeller, on-prem oss modeller, kendi anahtarlarınızla proxy, onaylı MCP sunucuları | | Vision Aginies | Doküman · görüntü · maskeleme | Fatura, sözleşme ve form çıkarımı; hasar fotoğrafı; kişisel veri tespiti ve maskeleme | | Workforce Aginies | Kiralık ajan iş gücü | iOS geliştirici, iş analisti, muhasebeci, testçi gibi roller ajan olarak kiralanır; iş kaydı, gölge dönem, ajan-ay faturalama | | Dialog Aginies | Diyalog arayüzü | Web widget, mobil, WhatsApp; oturum hafızası; insan devralma; mesaj bazında denetim izi | ## Üretim kanıtı Aginies üzerinde bugün canlıda koşan iş yüklerinin son 30 günlük üretim verileri: - 13,1M+ çalıştırma (son 30 gün) - %97,4 başarı oranı - 1.920+ ajan üretimde aktif - 13.100+ ajan tüm dağıtımlarda Bu iş yükleri Türkiye’de perakende, sigorta, otomotiv ve bankacılık kurumlarında çalışmaktadır. Ürün verisinin yanında, sitedeki ajan kütüphanesinden türetilen gerçekler: | Gösterge | Değer | Kaynak | | --- | --- | --- | | Canlı senaryo | 77 | Ajan kütüphanesi | | Sektör | 14 | Senaryo tanımları | | Uzman ajan | 261 | Senaryo tanımları | | Farklı araç tanımı | 657 | Ajan araç listeleri | | Bağlı sistem | 200 | Senaryo sistem listeleri | | İnsan onayı kuralı olan senaryo | 73 | Onay kuralları | | Kurumsal sistemde işlem adımı | 127 | Adım türleri | | Simüle çalıştırma / ay (toplam) | 3.320.970 | Senaryo metrikleri (tahmini) | Bu tablo derleme sırasında ajan kütüphanesinin senaryo tanımlarından türetilir; senaryo eklendikçe değişir. Hacim ve maliyet değerleri simülasyon tanımlarıdır, tahminidir. ## Kimin için Aginies, AI’ı pilot aşamasından çıkarıp kontrollü biçimde üretime almak isteyen büyük ve orta ölçekli kurumlar için tasarlandı. Özellikle regüle sektörlerin gereksinimleri mimarinin parçasıdır. - CIO, CDO ve dönüşüm liderleri: dağınık girişimleri tek platformda birleştirmek - İş birimi yöneticileri: müşteri, operasyon ve finans süreçlerinde ölçülebilir sonuç - Risk, uyum ve iç denetim: her karar için izlenebilirlik ve insan onayı - Mühendislik ekipleri: ajan kütüphanesi, SDLC ajan ekibi ve kurum içi kod tamamlama Ajan kütüphanesi on dört sektörü kapsar: bankacılık, finansal teknolojiler, sigorta, perakende ve e-ticaret, telekom, sağlık, üretim ve otomotiv, lojistik, enerji, kamu ve hükümet, savunma sanayii, IoT ve bağlı cihazlar, kurumsal fonksiyonlar ve yazılım geliştirme. ## Çözüm paketleri 30 çözüm paketi yedi alanda toplanır. Her paket kapsam, başlangıç gereksinimleri, ölçülen hedefler, yönetişim kuralları ve bir yol haritasıyla tanımlıdır; mimarisi ve hacimleri bağlı senaryolardan türetilir. | Paket | Sektör | Alan | Bağlı senaryo | İlk ölçülen hedef | | --- | --- | --- | --- | --- | | E-ticaret müşteri desteği ve aksiyon | Perakende & E-ticaret | Müşteri etkileşimi ve deneyim | 4 | İlk temasta çözüm oranı: +15–25 puan | | Fatura ve doküman veri çıkarımı | Kurumsal Fonksiyonlar | Finansal analiz ve operasyonel zekâ | 4 | Dokunmadan işlenen belge payı: %70–85 | | Sesli AI ile hasar ve talep kaydı | Sigorta | Müşteri etkileşimi ve deneyim | 4 | Temsilcisiz tamamlanan ihbar payı: %60–75 | | E-posta sınıflandırma ve ek doğrulama | Kurumsal Fonksiyonlar | Operasyonel otomasyon | 4 | Doğru birime ilk yönlendirme: %95 ve üzeri | | Çağrı merkezi kalite yönetimi | Kurumsal Fonksiyonlar | Müşteri etkileşimi ve deneyim | 3 | Değerlendirilen çağrı payı: %3–5 örneklemden %100 kapsama | | Diyalog bankacılığı | Bankacılık | Müşteri etkileşimi ve deneyim | 4 | Konuşarak tamamlanan işlem payı: %55–70 | | SDLC ajan ekibi | Yazılım Geliştirme | Yazılım geliştirme ve teknoloji verimliliği | 5 | İlk incelemede kabul edilen PR payı: %60–75 | | Satış ve lead yönetimi | Perakende & E-ticaret | Müşteri etkileşimi ve deneyim | 4 | Lead’e ilk temas süresi: Saatlerden dakikalara | | Finansal analiz ve operasyonel zekâ | Finansal Teknolojiler | Finansal analiz ve operasyonel zekâ | 4 | Aylık rapor hazırlama süresi: %60–70 kısalma | | Operasyonel otomasyon ve talep yönetimi | Kurumsal Fonksiyonlar | Operasyonel otomasyon | 4 | Otomatik kapatılan talep payı: %45–60 | | Bilgi yönetimi ve karar desteği | Kamu ve Hükümet | Bilgi yönetimi ve karar desteği | 4 | Kaynak gösterilen cevap payı: %100 | | Sağlık operasyonları ve klinik yönlendirme | Sağlık | Müşteri etkileşimi ve deneyim | 4 | Randevuya gelmeme oranı: %30–40 azalma | | İK ve çalışan deneyimi | Kurumsal Fonksiyonlar | İK ve çalışan deneyimi | 3 | Onboarding tamamlanma süresi: Haftalardan günlere | | KVKK kişisel veri koruma | Kurumsal Fonksiyonlar | Veri koruma ve uyum | 4 | Kişisel veri tespit doğruluğu: %97 ve üzeri | | Ses analizi ve kalite | Telekom | Müşteri etkileşimi ve deneyim | 4 | Analiz edilen çağrı payı: %100 | | Saha servis ve sevk otomasyonu | IoT ve Bağlı Cihazlar | Operasyonel otomasyon | 4 | Uzaktan çözülen talep payı: %30–45 | | IoT cihaz filosu operasyonu | IoT ve Bağlı Cihazlar | Yazılım geliştirme ve teknoloji verimliliği | 4 | Alarmdan olaya kümelenme: %80 gürültü azalması | | Enerji müşteri operasyonları | Enerji | Müşteri etkileşimi ve deneyim | 4 | Temsilcisiz çözülen çağrı payı: %55–65 | | Vatandaş hizmetleri ve dosya ön inceleme | Kamu ve Hükümet | Müşteri etkileşimi ve deneyim | 4 | Vatandaştan istenen tekrar belge: %70 azalma | | Kamu alımı ve ihale uyumu | Kamu ve Hükümet | Veri koruma ve uyum | 3 | Dosya ön inceleme süresi: %70 kısalma | | Savunma ihracat kontrolü ve tedarik zinciri güvenliği | Savunma Sanayii | Veri koruma ve uyum | 4 | Taraf tarama süresi: Günlerden dakikalara | | Ajan kiralama (Agent Outsourcing) | Kurumsal Fonksiyonlar | İK ve çalışan deneyimi | 4 | Kabul edilen çıktı payı: %85 ve üzeri | | Kredi ve müşteri edinimi operasyonları | Bankacılık | Finansal analiz ve operasyonel zekâ | 4 | Şubeye gitmeden açılan hesap payı: %70–85 | | Dolandırıcılık ve itiraz operasyonları | Bankacılık | Operasyonel otomasyon | 4 | Alarm başına inceleme süresi: %70 kısalma | | Hasar operasyonları uçtan uca | Sigorta | Operasyonel otomasyon | 4 | Dosya kapanış süresi (basit hasar): Haftalardan günlere | | Underwriting ve yenileme operasyonları | Sigorta | Finansal analiz ve operasyonel zekâ | 3 | Otomatik yenilenen poliçe payı: %60–75 | | Üye işyeri ve ödeme operasyonları | Finansal Teknolojiler | Operasyonel otomasyon | 4 | İşyeri destek talebi ilk temasta çözüm: %70 ve üzeri | | Abone hizmetleri ve şebeke desteği | Telekom | Müşteri etkileşimi ve deneyim | 5 | Temsilcisiz çözülen talep payı: %60–70 | | Lojistik kontrol kulesi | Lojistik | Operasyonel otomasyon | 4 | Gecikme tespit öncülü: Saatler önce | | Üretim ve bakım operasyonları | Üretim & Otomotiv | Operasyonel otomasyon | 4 | Plansız duruş süresi: %20–30 azalma | Hedefler benzer kurulumlardan aralık olarak verilir; başlangıç değeri ve hedef pilotun ilk haftasında kurumla birlikte sabitlenir. Değerler keşif oturumunda kurumun kendi verisiyle yeniden hesaplanır. --- # Ajan kiralama: dış kaynak insan yerine dış kaynak ajan · Dokümanlar URL: https://www.aginies.com/tr/dokumanlar/agent-outsourcing Language: tr Summary: Workforce Aginies ile iOS geliştirici, iş analisti, muhasebeci ve test mühendisi gibi roller ajan olarak kiralanır. Bu sayfa modeli, rolleri, iş kaydını, çalışma modelini, fiyatlamayı ve yönetişimi anlatır. Dokümanlar Bu sayfada Model nedir?Rol kataloğuİş kaydı: neyi ne kadar yaptığıÇalışma modeli: beş adımTicari modelYönetişim ve güvenlikÖlçülen hedefler ve yol haritası Ajan kiralama # Ajan kiralama: dış kaynak insan yerine dış kaynak ajan Workforce Aginies ile iOS geliştirici, iş analisti, muhasebeci ve test mühendisi gibi roller ajan olarak kiralanır. Bu sayfa modeli, rolleri, iş kaydını, çalışma modelini, fiyatlamayı ve yönetişimi anlatır. Güncelleme: Eylül 2026 ## Model nedir? Klasik dış kaynak modeli kişi-ay satar: bir tedarikçi size geliştirici, analist ya da muhasebeci verir; siz saat başına ya da ay başına ödersiniz. Ajan kiralamada aynı rolü bir ajan üstlenir. Ajan sizin araçlarınızda (Jira, Git, ERP, CRM) hesap alır, sizin ekibinize rapor verir, görevleri iş kaydına yazar. Kararlar sizin onaylayanınızda kalır; Aginies tarafında adı belli bir ekip lideri ajanı yönetir. Konumlandırma cümlesi: “Rolü kiralayın, ajan üstlensin.” Ajan bir rolün sahibidir; ancak yetkisi araç bazında sınırlı, her adımı kayıtlı ve her zaman durdurulabilir. - İşe başlama günler sürer: rol tanımı, erişim, iki haftalık gölge dönem. - Bilgi workspace’te kalır: bir sonraki ajan aynı test korpusu ve bağlantılarla başlar. - İzlenebilirlik iş kaydıyla sağlanır: görev, çıktı, saat karşılığı, kalite, onay, maliyet. - Ölçekleme dakikalar: aynı rolden ikinci ajan; talep düşünce kapasite ve fatura düşer. ## Rol kataloğu Katalogda bugün on iki rol var; her rol bir ajan spesifikasyonudur: ne yapar, nerede çalışır, hangi çıktıları üretir, hangi KPI’larla izlenir, hangi adımda insan onayı gerekir. Kataloğa sığmayan roller keşif oturumunda Otonom Ajan ile günler içinde tanımlanır. | Rol | Yerine geçtiği | Örnek çıktı | İnsan kontrolü | | --- | --- | --- | --- | | iOS Geliştirici Ajanı | Dış kaynak iOS geliştirici | Testli PR, kapanan hikâye puanı | Her PR insan incelemeli; dağıtım onaylı | | Backend Geliştirici Ajanı | Dış kaynak backend geliştirici | Uç nokta, migrasyon, entegrasyon testi | Şema değişikliği ve üretim dağıtımı onaylı | | Test Mühendisi Ajanı | Dış kaynak testçi | Regresyon koşumu, kanıtlı hata, kapsama raporu | Sürüm go / no-go sürüm yöneticisinde | | DevOps / SRE Ajanı | Dış kaynak DevOps mühendisi | Pipeline değişikliği, kapanan olay, maliyet raporu | Üretim altyapı değişikliği onaylı | | İş Analisti Ajanı | Dış kaynak iş analisti | Hazır hikâye, kabul kriteri, süreç haritası | Ürün sahibi hikâyeleri onaylar | | Veri Analisti Ajanı | Dış kaynak veri analisti | Rapor, pano, veri kalitesi bulgusu | Dışa paylaşılan rapor onaylı | | Muhasebe Ajanı | Dış kaynak muhasebeci | Kayıt, mutabakat, kapanış görevi | Kayıtlar mali müşavir onayıyla; beyanname insanda | | Satın Alma Uzmanı Ajanı | Dış kaynak satın alma uzmanı | RFQ, eşleşen fatura, tasarruf raporu | Sipariş onayı ve tedarikçi seçimi yöneticide | | Müşteri Destek Temsilcisi Ajanı | Dış kaynak çağrı merkezi temsilcisi | Çözülen bilet, ilk temasta çözüm | İade ve tutar üstü işlemler onaylı; %10 örneklem | | Satış Geliştirme Ajanı | Dış kaynak SDR | Nitelendirilen fırsat, planlanan toplantı | Dış iletişim her zaman insan onaylı | | İşe Alım Uzmanı Ajanı | Dış kaynak işe alım uzmanı | Ön elenen aday, planlanan mülakat | Eleme ve teklif kararı İK’da | | Sözleşme Analisti Ajanı | Dış kaynak hukuk asistanı | İncelenen sözleşme, redline, özet | Her redline avukat onayıyla | ## İş kaydı: neyi ne kadar yaptığı Her görev iş kaydında bir satırdır. Satır şunları taşır: bilet ya da hikâye numarası, başlangıç ve bitiş, somut çıktı (PR, rapor, kayıt, bilet), saat karşılığı (aynı işin insan süresi tahmini), kalite (inceleme sonucu, yeniden çalışma, kontrol ailesi puanları), onay (kim, ne zaman, gerekçe) ve maliyet (token, araç çağrısı, model). - Günlük: iş kaydı ve onay kuyruğu; ekip lideriniz onay bekleyenleri görür. - Haftalık: zaman çizelgesi, KPI’lar ve kalite; klasik dış kaynak raporuyla aynı biçimde, ama görev bazında. - Aylık: fatura, gözden geçirme ve ölçek kararı; fatura kalemleri iş kaydından üretilir. Saat karşılığı bir karşılaştırma ölçüsüdür: aynı işin insanla kaç saat süreceğini gösterir. Fatura ajan-ay ya da çıktı bazlıdır. ## Çalışma modeli: beş adım - Rol tanımı: iş tanımı ajan spesifikasyonuna çevrilir (görevler, araçlar, çıktılar, KPI’lar, onay noktaları). - Erişim ve onboarding: ajan sizin araçlarınızda hesap alır; SSO, roller, okuma / yazma izinleri, veri sınırları. - Gölge dönem (iki hafta): her çıktı bir insan tarafından incelenir; hatalar test korpusuna eklenir. - Örneklemeli inceleme: kalite eşiği geçilince %10 örnekleme ve eşik bazlı onay; kill-switch her zaman açık. - Aylık gözden geçirme ve ölçek: iş kaydı, KPI’lar ve maliyet birlikte okunur; rol genişletilir, eklenir ya da sonlandırılır. Her adımda sorumluluk nettir: rol tanımı ve gözden geçirme sizinle Aginies ekip liderinin ortak işi; erişim BT ve güvenliğin; inceleme ve onay sizin ekibinizin. ## Ticari model | Model | Nasıl çalışır | Uygun olduğu durum | | --- | --- | --- | | Ajan-ay | Rol başına aylık sabit ücret; sınırsız görev, tanımlı KPI ve SLA. | Sürekli iş yükü olan roller | | Çıktı bazlı | Birleştirilen PR, çözülen bilet, işlenen fatura başına ücret; iş kaydı fatura kalemidir. | Dalgalı ya da proje bazlı işler | | Ajan pod’u | Birden çok rol + Aginies ekip lideri; ürün ekibi gibi çalışır. | Uçtan uca teslimat isteyen ekipler | Sitede liste fiyatı yoktur; rol, hacim ve onay yoğunluğuna göre keşif oturumunda birlikte belirlenir. Çıktıların ve üretilen bilginin mülkiyeti müşterinindir; sözleşme aylık sonlandırılabilir ve ayrılırken test korpusu müşteride kalır. ## Yönetişim ve güvenlik - Ajan sizin tenant’ınızda ve araçlarınızda çalışır; kod ve veri dışarı çıkmaz. - Araç bazında okuma / yazma izni; eşik aşan her adım onaya düşer (onay matrisi Güvenlik sayfasında). - Her çıktı için onaylayan ve gerekçe kayıtlı; denetim dışa aktarımı CSV / JSON. - Kill-switch: rol ya da görev bazında anında durdurma; geri alma planı her yazma işleminde. - Aginies tarafında adı belli ekip lideri: haftalık planlama, aylık gözden geçirme, ilk müdahale. Nasıl anlatılır: “Ajan çalışan gibi yönetilir, sistem gibi denetlenir.” Müşteriye önce iş kaydını gösterin, sonra rol kataloğunu; fiyat konuşması keşif oturumuna kalır. ## Ölçülen hedefler ve yol haritası | Gösterge | Hedef aralığı | | --- | --- | | Kabul edilen çıktı payı | %85 ve üzeri | | Rol başına maliyet | Dış kaynak kişi-ayın %30–50’si | | Bilgi kaybı (devirde) | Sıfır; bilgi workspace’te kalır | Başlamak için gerekenler: - Rol için gerekli araç erişimleri (Jira, Git, ERP, bilet sistemi) - Görev tanımı ve kabul kriterleri - Kurum tarafında inceleme sorumlusu Yol haritası: - 1 · Politika ve rol kataloğu · 2 hafta: İK politikaları, süreç sahipleri ve sistem erişimleri tanımlanır; kişisel veri sınırları yazılır. - 2 · Pilot · 4 hafta: Bir departman ya da bir rolle; kalite puanı ve kullanıcı geri bildirimi haftalık izlenir. - 3 · Yaygınlaştırma: Tüm departmanlar; eşikler, inceleme örneklemi ve kill-switch İK yönetiminde. Hedefler benzer kurulumlardan aralık olarak verilir; başlangıç değeri ve hedef ilk haftada birlikte sabitlenir. --- # Ajanlar nasıl kurulur ve çalışır · Dokümanlar URL: https://www.aginies.com/tr/dokumanlar/agents Language: tr Summary: Yapı taşları, süpervizör → alt ajan deseni, araçlar ve MCP politikaları, onaylar, çalıştırma yaşam döngüsü, ajan kütüphanesi ve SDLC ajan ekibi. Dokümanlar Bu sayfada Yapı taşlarıSüpervizör → alt ajan → süpervizör deseniAraçlar ve MCP politikalarıOnaylar ve eşiklerÇalıştırma yaşam döngüsüAjan kütüphanesiSDLC ajan ekibi Ajanlar # Ajanlar nasıl kurulur ve çalışır Yapı taşları, süpervizör → alt ajan deseni, araçlar ve MCP politikaları, onaylar, çalıştırma yaşam döngüsü, ajan kütüphanesi ve SDLC ajan ekibi. Güncelleme: Eylül 2026 ## Yapı taşları Görsel ajan tasarımcısında her ajan sekiz yapı taşından kurulur. Aynı yapı taşları Otonom Ajan tarafından da kullanılır; diyalogla üretilen bir ajan, elle tasarlanan ajanla aynı biçimde çalışır ve izlenir. | Yapı taşı | Görevi | Tipik kullanım | | --- | --- | --- | | Ajan | Kendi modeli, araçları ve hafızasıyla bir uzman | Kredi ajanı, hasar ajanı, fatura ajanı | | Süpervizör | Hedefi adımlara böler, ajanlara dağıtır, sonucu birleştirir | Her çok ajanlı senaryonun giriş noktası | | Yönlendirici | Niyete, kurala ya da skora göre süreci dallandırır | Niyet sınıflandırma, risk skoruna göre yol seçimi | | Hafıza | Oturum ve uzun dönem bağlamı ajanlar arasında paylaşır | Müşteri bağlamı, önceki turlar | | Guardrail | Girdi ve çıktıyı politikaya göre denetler; durdurur ya da maskeler | PII maskeleme, yetkisiz talep, prompt injection | | İnsan onayı | Kritik kararda durur, yetkiliye sunar, kararı kaydeder | Eşik üstü tutar, politika istisnası | | Araç / MCP | Kurumsal sisteme okuma ya da yazma; araç başına politika | Core, CRM, ERP, ticketing çağrıları | | Paralel & döngü | Bağımsız adımları aynı anda, tekrarlı işi koşula kadar çalıştırır | Çoklu kaynak sorgusu, toplu belge işleme | ## Süpervizör → alt ajan → süpervizör deseni Aginies’in temel yürütme deseni budur. Süpervizör planlar ve sonuçları birleştirir; uzman ajanlar dar ve iyi tanımlı görevleri üstlenir. Kontrol her adımda süpervizöre geri döner; böylece bir alt ajanın hatası tüm süreci bozmaz. - Süpervizör talebi alır ve hedefi yapılandırılmış adımlara böler. - Her adımı uygun uzman ajana devreder; bağımsız adımlar paralel koşar. - Alt ajan kendi modeli ve araçlarıyla çalışır, sonucu ve gerekçesini döndürür. - Süpervizör sonucu doğrular, birleştirir ve bir sonraki adıma karar verir. - Onay gerektiren adımda süreç durur; karar kaydedilir, sonra devam eder. - Tamamlanınca sonuç kanala iletilir; tüm iz denetim kaydına yazılır. Yürütme durumludur: adımlar, ara sonuçlar ve araç yanıtları saklanır. Bir bağımlılık başarısız olduğunda süpervizör yeniden dener, yedek modele geçer ya da insana yükseltir; süreç kaldığı yerden devam eder. ## Araçlar ve MCP politikaları Ajanlar kurumsal sistemlere araçlar üzerinden ulaşır: REST API, veritabanı ve Model Context Protocol (MCP) sunucuları. Her araç okuma ya da yazma olarak sınıflandırılır ve kendi çağrı politikasını taşır. MCP Hub yalnızca onaylı sunucuları kataloglar. - Okuma / yazma sınıfı: yazma araçları varsayılan olarak onay ya da eşik kuralına bağlıdır - Araç başına çağrı politikası: izinli parametreler, hız sınırı, hangi ajan çağırabilir - Girdi ve çıktı maskeleme: araca giden ve araçtan dönen hassas veri redakte edilir - Her çağrı izde görünür: parametre, süre, yanıt kodu ve sonuç - MCP Hub: onaylı sunucular, sürüm ve erişim yetkisi merkezi yönetimde - Ajanlar arası iletişim (A2A) aynı politika çerçevesinde çalışır Çekirdek sistemlere yetki kademeli açılır: önce okuma, sonra kontrollü yazma, en sonda otonom yazma. Her kademe ölçülür ve onaylanır. ## Onaylar ve eşikler İnsan onayı ajan tanımının yapı taşıdır. Kritik kararda süreç durur, karar yetkiliye bağlam ve gerekçeyle sunulur, verilen karar denetim izine yazılır. Eşikler ve yetkililer ajan tanımında belirlenir. - Tutar eşiği: belirli bir değerin üzerindeki işlem yetkiliye gider - Risk eşiği: skor ya da guardrail sinyali onay tetikler - Politika istisnası: kural dışı talep her zaman insana düşer - Geri alınamaz işlemler: silme, ödeme ve sözleşme değişikliği için zorunlu onay - Yetkili rolü: şube yöneticisi, uyum ekibi, yayın yöneticisi gibi tanımlı roller - Kill-switch: ajan ya da senaryo bazında anında durdurma ve otomasyon sınırı Örnek: kart borcu ödeme ajanında “tutar > 100.000 → şube yöneticisi onayı”. Kural ajan tanımında yaşar; her karar iz bırakır. ## Çalıştırma yaşam döngüsü Her çalıştırma dokuz adım türünden oluşur. Kütüphanedeki her ajan sayfası, adımları bu türlerle gösterir; her adımın süresi, token kullanımı ve dokunduğu sistem izde kayıtlıdır. - Alım (ingest): sinyal gelir; ses, e-posta, API, mesaj ya da olay - Plan (plan): süpervizör hedefi böler ve yönlendirir - Getirme (retrieve): bilgi tabanı ya da veritabanında arama (RAG) - Akıl yürütme (reason): sınıflandırma, çıkarım, skorlama - Araç (tool): kurumsal sisteme okuma çağrısı ya da MCP aracı - Doğrulama (verify): guardrail, politika ve tutarlılık kontrolü - Onay (approval): insan kontrol noktası - Yazma (write): kurumsal sistemde durum değiştiren işlem - Bildirim (notify): müşteri ya da çalışan iletişimi Bir çalıştırma alım adımıyla başlar, yazma ya da bildirim adımıyla biter. Uçtan uca gecikme, adım sürelerinin toplamından türetilir; çalıştırma başına maliyet, model çağrılarının token kullanımından hesaplanır ve senaryo, birim ve kanal bazında atfedilir. ## Ajan kütüphanesi Kütüphane 14 sektörde 77 ajan içerir. Her ajan gerçek ajan tanımlarından türetilmiş bir simülasyondur: süpervizör, uzman ajanlar, sistemler, adımlar ve onay kuralı tanımlıdır. Her biri Otonom Ajan ile aynı biçimde üretilebilir ve kurumun sistemlerine uyarlanabilir. Bir ajan sayfası şöyle okunur: - Başlık ve slogan: ajanın amacı ve sonuç odaklı özeti - Tetikleyici ve sonuç: çalıştırmayı ne başlatır, bittiğinde ne doğru olur - Kanallar: ses, e-posta, uygulama, API, WhatsApp, web ya da toplu iş - Süpervizör ve ajanlar: her ajanın rolü, modeli ve araçları - Sistemler: dokunulan core, CRM, ERP, ticketing, ödeme ve diğer sistemler - Adımlar: yürütme sırasıyla adım türü, hedef sistem, süre ve token - Onay kuralı: varsa insan kontrol noktasının eşiği ve yetkilisi - Metrikler: aylık çalıştırma, başarı oranı, medyan süre, çalıştırma başına maliyet - Esinlenilen desen: ajanın yansıttığı, dünyada kanıtlanmış uygulama Metrikler bir kurumsal dağıtım için gerçekçi göstergelerdir; kurumun hacmi ve sistemlerine göre keşif oturumunda yeniden hesaplanır. Kütüphanenin tam listesi; her satır bir senaryo sayfasıdır. | Senaryo | Sektör | Kanallar | Ajan | Sistem | Onay kuralı | | --- | --- | --- | --- | --- | --- | | Diyalog bankacılığı: sesli ve yazılı işlem asistanı | Bankacılık | app, voice, whatsapp | 4 | 5 | Limit artışı > %60, itiraz tutarı > 5.000 TL veya yeni alıcıya transfer → operasyon onayı | | Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit | Finansal Teknolojiler | whatsapp, web, voice | 4 | 5 | İade > 25.000 TL veya limit artışı > %50 → risk ekibi onayı | | Uçtan uca kasko hasarı: ihbardan ödemeye | Sigorta | voice, app | 4 | 5 | Tahmin > 150.000 TL, kusur oranı belirsiz veya önceki hasarla çakışma → eksper onayı | | Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin | Perakende & E-ticaret | app, web, whatsapp | 4 | 5 | Kampanya kuralı dışı indirim veya kişiye özel fiyat talebi → müşteri hizmetleri | | Abone hizmet asistanı: fatura, paket, arıza ve taahhüt | Telekom | voice, app, whatsapp | 4 | 5 | Taahhüt iptali, 500 TL üzeri fatura düzeltmesi veya teknisyen önceliği → temsilci onayı | | Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR | Yazılım Geliştirme | api, batch | 4 | 4 | Üretim şeması değişikliği veya kapsam dışı dosya → mimar onayı; PR birleştirme her zaman insanda | | Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuç | Sağlık | whatsapp, voice, app | 4 | 5 | İlaç etkileşimi, kontrendikasyon şüphesi veya ek tetkik ihtiyacı → hekim onayı | | Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna | Üretim & Otomotiv | api, app | 4 | 5 | Plan değişikliği > 2 saat veya 50.000 TL üzeri acil satın alma → vardiya amiri onayı | | Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA | Lojistik | api, email, whatsapp | 4 | 5 | Rota maliyet artışı > %15 veya tazmin > 10.000 TL → operasyon müdürü onayı | | Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda | Enerji | voice, whatsapp, app | 4 | 5 | Taksit > 6 ay, fatura iptali veya sayaç şüphesi → müşteri hizmetleri onayı | | Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi | Kamu ve Hükümet | whatsapp, voice, web | 4 | 5 | Uygunluk sınırda, belge tutarsız veya aynı hanede başka başvuru → memur incelemesi | | Satın almadan ödemeye: talep, teklif, sözleşme ve fatura tek masada | Kurumsal Fonksiyonlar | email, api | 4 | 5 | Tutar > 250.000 TL, tek teklif veya çerçeve dışı tedarikçi → satın alma komitesi | | Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolü | Savunma Sanayii | app | 4 | 5 | Prosedürden sapma önerisi → yetkili mühendis onayı; ihracat kontrollü içerik → yetki seviyesi şart | | Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyen | IoT ve Bağlı Cihazlar | whatsapp, web, api | 4 | 5 | Toplu OTA (>100 cihaz) veya garanti dışı cihaz değişimi → ürün sahibi onayı | | Kredi kartı limit artışı | Bankacılık | app, voice | 3 | 3 | Limit > 100.000 TL → şube yöneticisi onayı | | Dijital müşteri edinimi (KYC) | Bankacılık | app, web | 4 | 5 | Yaptırım/PEP bulanık eşleşme skoru > 0,65 → uyum uzmanı onayı | | Kart dolandırıcılık alarmı triyajı | Bankacılık | whatsapp, sms, voice, api | 3 | 5 | Teyitli dolandırıcılık zararı > 2.500 TL → analist onayı ile kalıcı iptal | | Ters ibraz (chargeback) itirazı | Bankacılık | app, email, branch | 3 | 4 | Tutar > 2.500 TL → geçici alacak öncesi itiraz ekibi lideri onayı | | Erken dönem tahsilat iletişimi | Bankacılık | batch, whatsapp, sms, voice | 3 | 4 | Plan bakiyesi > 25.000 TL veya > 3 taksit → tahsilat ekip lideri onayı | | Konut kredisi belge incelemesi | Bankacılık | web, email, branch | 3 | 4 | Kredi tutarı > 3.000.000 TL veya DTI > %45 → kredi tahsis uzmanı onayı | | Sesli bankacılıkla para transferi | Bankacılık | voice | 4 | 4 | Tutar > 50.000 TL veya yeni alıcı → çağrı merkezi temsilcisine sıcak aktarım | | Kredi kartı borcu ödeme | Bankacılık | app, voice | 3 | 4 | Tutar > 100.000 TL → şube yöneticisi onayı | | Üye işyeri KYB onboarding | Finansal Teknolojiler | web, email | 4 | 5 | Risk sınıfı yüksek veya UBO bulanık eşleşme > 0,65 → uyum yöneticisi onayı | | AML işlem izleme alarm triyajı | Finansal Teknolojiler | api, batch | 4 | 4 | Her alarm → uyum analisti kararı zorunlu; ajan tek başına STR dosyalamaz, vaka kapatmaz | | BNPL anlık taksit kararı | Finansal Teknolojiler | api, app, sms | 4 | 5 | Sepet > 15.000 TL veya büro skoru 10.000 TL veya toplam > 50.000 TL → finans müdürü onayı ile ayarlama kaydı | | Sesli hasar ihbarı (FNOL) | Sigorta | voice, whatsapp | 4 | 4 | Yaralanma veya rezerv > 250.000 TL → hasar uzmanına sıcak aktarım | | Poliçe yenileme teklifi | Sigorta | batch, whatsapp, email, app | 3 | 4 | Prim artışı > %20 veya indirim > %10 → yenileme ekibi lideri onayı | | Underwriting belge çıkarımı | Sigorta | email, web, batch | 3 | 4 | Ek prim ≥ %50 veya teminat > 5.000.000 TL → kıdemli underwriter onayı | | Fotoğrafla kasko hasar tespiti | Sigorta | app, whatsapp | 3 | 4 | Tahmin > 15.000 TL veya model güveni 50.000 TL → rücu avukatı onayı | | Siparişim nerede? (WISMO) | Perakende & E-ticaret | whatsapp, web, app, voice | 3 | 4 | — | | İade ve para iadesi | Perakende & E-ticaret | app, web, whatsapp, email | 3 | 4 | İade tutarı > 2.500 TL veya 12 ayda > 5 iade → ekip lideri onayı | | Ürün soru-cevap asistanı | Perakende & E-ticaret | web, app, whatsapp | 3 | 4 | — | | Katalog zenginleştirme (toplu) | Perakende & E-ticaret | batch, email | 3 | 4 | Yeni kategori veya tedarikçi fiyatı sapması > %15 → kategori yöneticisi onayı | | Yorum yanıtlama ve kalite eskalasyonu | Perakende & E-ticaret | web, api | 3 | 3 | 1 yıldızlı yorum yanıtları ve kusur eskalasyonları → topluluk yöneticisi onayı | | Fatura itirazı çözümü | Telekom | app, voice, whatsapp | 3 | 3 | İyi niyet kredisi > 75 TL → ekip lideri onayı | | SIM değişimi dolandırıcılık kontrolü | Telekom | branch, app, api | 3 | 4 | Risk skoru ≥ 70 veya 72 saat içinde kimlik bilgisi değişikliği → dolandırıcılık masası onayı | | Şebeke kesintisi iletişimi | Telekom | sms, whatsapp, api, batch | 3 | 4 | Etkilenen abone > 10.000 veya P1 kesinti → şebeke operasyon yöneticisi onayı | | Tarife yükseltme danışmanı | Telekom | app, whatsapp, voice | 3 | 5 | Taahhütlü veya indirimli özel teklif → kampanya yöneticisi onayı | | Özellik teslimat hattı | Yazılım Geliştirme | api, web | 4 | 4 | Her PR → en az 1 insan kod incelemesi onayı; ajan main'e doğrudan yazamaz | | CI/CD sürüm hattı | Yazılım Geliştirme | api, batch | 4 | 4 | Her üretim dağıtımı → sürüm yöneticisi onayı zorunlu; kritik CVE varsa kapı otomatik kapanır | | Olay müdahalesi ve acil düzeltme | Yazılım Geliştirme | api, web | 4 | 5 | Her acil üretim dağıtımı → nöbetçi mühendis onayı; yama PR'ı da insan incelemesinden geçer | | Kod incelemesi ve güvenlik kapısı | Yazılım Geliştirme | api, web | 4 | 4 | Her PR → CODEOWNERS insan onayı; yüksek SAST bulgusu veya kapsam 25.000 TL → kullanım yönetimi hemşiresi onayı | | Hasta ön kabul ve triyaj | Sağlık | app, voice, whatsapp | 4 | 4 | Her vakada aciliyet seviyesini triyaj hemşiresi onaylar; "acil durum" vakaları doğrudan 112'ye yönlendirilir | | Randevu planlama ve no-show azaltma | Sağlık | whatsapp, sms, batch | 3 | 3 | — | | Tedarikçi faturası üçlü eşleştirme | Üretim & Otomotiv | email, api, batch | 3 | 3 | Fiyat sapması > %2 veya > 1.000 TL → satın alma sorumlusu onayı | | Bakım talebi triyajı | Üretim & Otomotiv | api, app, voice | 4 | 4 | Planlı duruş > 4 saat → vardiya amiri onayı | | Garanti talebi işleme | Üretim & Otomotiv | web, api, email | 3 | 4 | Garanti talebi > 5.000 € → bölge garanti müdürü onayı | | Sevkiyat istisna yönetimi | Lojistik | api, email, whatsapp | 3 | 4 | Ek maliyet > 1.000 € → sevk şefi onayı | | Gümrük belgesi hazırlama | Lojistik | api, email, batch | 3 | 5 | GTİP güven skoru 2.000 € → gümrük uzmanı onayı | | Kesinti bildirimi yönetimi | Enerji | voice, app, whatsapp, web | 4 | 5 | Toplu bilgilendirme > 1.000 müşteri → kontrol odası dispeçeri onayı | | Sayaç okuma itirazı | Enerji | whatsapp, app, web | 3 | 4 | İade / düzeltme > 1.000 TL → faturalama ekip lideri onayı | | Çatı GES bağlantı başvurusu | Enerji | web, email, api | 4 | 4 | Bağlantı görüşü her zaman şebeke planlama mühendisinin imzasıyla | | Talep yanıtı (demand response) olayı | Enerji | api, app, whatsapp, batch | 3 | 4 | Olay > 10 MW → şebeke operatörü onayı; ödeme hesabı denetim iziyle | | Vatandaş başvurusu yönlendirme | Kamu ve Hükümet | web, app, email, voice | 3 | 4 | Sınıflandırma güveni 500.000 € → hukuk müşaviri onayı | | Çalışan işe alıştırma (onboarding) | Kurumsal Fonksiyonlar | api, email, app | 4 | 5 | Şablon dışı erişim veya yönetici yetkisi → bölüm yöneticisi onayı | | BT yardım masası L1 | Kurumsal Fonksiyonlar | app, voice, web | 3 | 3 | Finans veya üretim grubuna erişim → grup sahibi onayı | | Borçlar hesabı (AP) posta kutusu | Kurumsal Fonksiyonlar | email, batch | 3 | 4 | Sipariş (SAS) olmayan fatura > 25.000 TL → masraf yeri sahibi onayı | | Satış fırsatı nitelendirme | Kurumsal Fonksiyonlar | web, api, email | 3 | 3 | Kurumsal (> 1.000 çalışan) veya rakip şüphesi → SDR lideri kabulü | | Ay sonu kapanış ve mutabakat (kiralık muhasebe ajanı) | Kurumsal Fonksiyonlar | batch, email, api | 3 | 4 | Deftere işlenen her kayıt ve beyanname mali müşavir onayıyla | | İhracat kontrolü ve son kullanıcı taraması | Savunma Sanayii | api, email, batch | 3 | 4 | Kontrole tabi her kalem → ihracat uyum yöneticisi onayı; ajan lisans kararı veremez | | Platform bakım (MRO) iş emri hazırlığı | Savunma Sanayii | app, api, batch | 4 | 4 | Uçuşa elverişliliği etkileyen her iş paketi → bakım planlayıcısı onayı | | Tedarikçi güvenlik ve uyum değerlendirmesi | Savunma Sanayii | email, web, api | 3 | 4 | Nitelendirme kararı her zaman tesis güvenlik müdüründe; ajan yalnızca hazırlar | | Cihaz filosu anomali triyajı | IoT ve Bağlı Cihazlar | api, batch, app | 3 | 5 | Toplu cihaz komutu > 100 cihaz → filo operasyon lideri onayı | | Kademeli OTA güncelleme dağıtımı | IoT ve Bağlı Cihazlar | api, batch | 3 | 4 | Kritik filo segmentine dağıtım → ürün sahibi onayı; sağlık kapısı düşerse otomatik geri alma | | Soğuk zincir sapması müdahalesi | IoT ve Bağlı Cihazlar | api, app, whatsapp | 3 | 5 | Ürün taşıma ya da imha kararı → kalite yöneticisi onayı | ## SDLC ajan ekibi Code Aginies içinde beş rol, CI/CD hattında birlikte çalışır. Talepten üretime kadar süreci sahiplenirler; kod incelemesi ve üretim kararı her zaman insanda kalır. Her adım Jira, Git ve CI kayıtlarıyla izlenir. - Proje Yöneticisi Ajanı: talepleri önceliklendirir, kapsamı ve bağımlılıkları netleştirir - Analist Ajanı: kabul kriterlerini ve test senaryolarını yazar - Mimar Ajanı: tasarımı ve etki analizini çıkarır - Geliştirici Ajanı: dalda kodu yazar, PR açar - Test Ajanı: birim ve entegrasyon testlerini üretir, çalıştırır, kapsamı raporlar CI/CD hattında SAST, bağımlılık taraması ve staging duman testleri otomatik koşar. Üretim dağıtımı yayın yöneticisi onayına gider; acil düzeltmelerde aynı kapılar kısaltılmış süreyle uygulanır. ## Canlı örnek: bir ajanın çalışması banking ### Kredi kartı borcu ödeme [Kredi kartı borcu ödeme] çalıştırma #41.286.992süre 0 mstoken 0tahmini maliyet $0,000durum —Onay kuralı: Tutar > 100.000 TL → şube yöneticisi onayı Bu çalıştırmadaki ajanlar Kart Süpervizörüsupervisorplanlar, yönlendirir, onaylar Niyet Ajanıaginies-sunkar-1 · on-premTalebi anlar; kartı, tutarı ve ödeme türünü çıkarır Ödeme Ajanıaginies-tulpar-4 · on-premHesap bakiyesini kontrol eder, OTP ister, ödemeyi çekirdek sistemde yapar Ses Ajanıaginies-voice-tr-v3 · on-premSesli kanalda konuşmayı yazıya döker ve yanıtı seslendirir Denetim izi00:00.00 - ◦Talep alındı, oturum kimliği doğrulandı · kanal: mobil - ∿Niyet: borç ödeme · Kart: ****2210 · Tür: dönem borcunun tamamı - ⚙Kart borcu çekildi: 24.860 TL · son ödeme 14 Eylül - ⌘Plan: hesap seçimi → OTP → ödeme → bildirim - ⚙Hesaplar listelendi: vadesiz ****7741 bakiye yeterli - ✓Politika: tutar ≤ 100.000 TL → onay gerekmez; OTP zorunlu - ➤OTP gönderildi ve doğrulandı (2 deneme hakkı kaldı) - ⇢24.860 TL ****7741 hesabından karta ödendi · işlem #TX-88120 - ➤Müşteriye sesli ve yazılı teyit verildi, karar izi kaydedildi - ■tamamlandı Simülasyon · gerçek ajan tanımlarından türetildi · her ajan Otonom Ajan ile diyalogla kurulabilir ve test korpusuyla doğrulanır --- # Platform dokümantasyonu · Dokümanlar URL: https://www.aginies.com/tr/dokumanlar/platform Language: tr Summary: Sekiz entegre yetenek katmanı, Otonom Ajan, bilgi tabanı, Hub, ses, görüntü, kod, diyalog, gözlemlenebilirlik ve workspace yönetimi. Dokümanlar Bu sayfada A · Orkestrasyon & Yürütme ZekâsıB · AI / Model OrkestrasyonuC · Bilgi & Erişim KatmanıD · Yönetişim, Risk & UyumE · Gözlemlenebilirlik & OptimizasyonF · Entegrasyon & BağlantıG · Çok Kanallı Deneyim & SesH · Dağıtım & Kurumsal MimariOtonom AjanBilgi TabanıHub Aginies: Model Hub, MCP Hub ve proxyVoice AginiesVision AginiesCode AginiesDialog AginiesGözlemlenebilirlik ve raporlamaWorkspace, tenant ve iş birliğiKütüphaneden türetilmiş platform gerçekleriEntegrasyon kataloğuİşletim modeli Platform ve ürünler # Platform dokümantasyonu Sekiz entegre yetenek katmanı, Otonom Ajan, bilgi tabanı, Hub, ses, görüntü, kod, diyalog, gözlemlenebilirlik ve workspace yönetimi. Güncelleme: Eylül 2026 ## A · Orkestrasyon & Yürütme Zekâsı Motor katmanı. Karmaşık işi uçtan uca yürütür; talebi akıllıca yönlendirir ve çok adımlı süreçleri bağımlılıklar başarısız olsa bile kararlı tutar. Her çalıştırma durumludur: adımlar, ara sonuçlar ve kararlar kayıt altındadır. - Süpervizör → alt ajan → süpervizör mimarisi - Dinamik ajan yönlendirme - Durumlu, çok adımlı yürütme; paralel dallar ve döngüler - Ajanlar arası iletişim (A2A) - Zamanlayıcı ve webhook tetikleyiciler - Ajan şablonları ve görsel ajan tasarımcısı - Otonom Ajan: diyalogla ajan üretimi ve test korpusu | Kapsam | Yönetişim | Kanıt | | --- | --- | --- | | Süpervizör planı, ajan ataması ve sonuç birleştirme; başarısız adımda yeniden deneme ve yedek yol · Durumlu çalıştırma: oturum ve uzun dönem hafıza, bekleyen onaylar, zaman aşımı kuralları · Paralel dallar ve döngüler; koşula bağlı sonlanma | Her ajanın yetkisi araç başına tanımlanır; süpervizör kapsam dışı araç çağıramaz. | Çalıştırma planı ve her adımın girdi/çıktısı iz kaydında; yeniden oynatılabilir. | ## B · AI / Model Orkestrasyonu Beyin seçimi ve dayanıklılık. Her iş için doğru modeli kullanır; vendor bağımlılığını azaltır ve bir sağlayıcı düştüğünde operasyonu sürdürür. Model değişimi ajanı yeniden tasarlamayı gerektirmez. - Çoklu model gateway: kapalı ve açık kaynak modeller tek arayüzde - Model proxy: kendi anahtarlarınızla bulut modellerine izlenebilir, maskeli erişim - Yeniden tasarım gerektirmeyen model değişimi - Yedek model stratejisi - Açık kaynak modeller için on-prem servis - Prompt test ortamı ve sürümleme - Hub Aginies: Qwen, GLM ve Kimi tabanlı ince ayarlı model aileleri, güncel açık sürümler ve ses modelleri | Kapsam | Yönetişim | Kanıt | | --- | --- | --- | | Hub kataloğu: Qwen, GLM ve Kimi tabanlı ince ayarlı aileler (tulpar, talay, sunkar, konrul) ve güncel açık sürümler · Model proxy: kurumun kendi anahtarlarıyla bulut modelleri, tek noktadan izleme ve maskeleme · Senaryo başına model seçimi, yedek model ve istem sürümleme | Model değişikliği sürümlüdür ve test korpusundan geçmeden üretime çıkmaz. | Her istekte kullanılan model sürümü, token ve maliyet kaydı; katalog geçmişi. | ## C · Bilgi & Erişim Katmanı Kurumsal hafıza. Dağınık kurumsal bilgiyi yönetilen, aranabilir ve güvenilir bilgiye dönüştürür; cevaplar kaynağa bağlanır. Chunk seviyesinde kontrol, RAG kalitesini doğrudan yönetilebilir kılar. - Doküman gömme hattı - Sürümlü bilgi tabanı yönetimi - Etiket bazlı doküman ayrımı - Hibrit arama ve çoklu kaynak RAG - Chunk seviyesinde inceleme ve düzenleme - Halüsinasyon azaltma mimarisi - Kaynak, etiket ve erişim politikası yönetimi | Kapsam | Yönetişim | Kanıt | | --- | --- | --- | | Doküman gömme hattı, chunk inceleme ve etiketleme; sürümlü bilgi tabanı · Erişim politikası chunk seviyesinde: kim hangi belgeyi hangi ajanla görebilir · Kaynak gösteren cevap; kaynak yoksa “bilmiyorum” | Bilgi tabanı değişiklikleri sahibi tarafından onaylanır; eski sürüm geri alınabilir. | Her cevapta kullanılan chunk’lar ve sürümleri; kapsama ve kalite raporu. | ## D · Yönetişim, Risk & Uyum Kontrol katmanı. AI’ı denetlenebilir, güvenli ve regüle ortamlara uygun kılar; her karar izlenebilir. Yönetişim sonradan eklenmez, mimarinin parçasıdır. - Guardrail’ler ve politika motoru - İnsan onayı kontrol noktaları: eşik ve yetkili bazlı - Ajan karar izlenebilirliği - Hassas veri maskeleme ve PII redaksiyonu - Tüm isteklerin maskelenmesi ve anonimleştirilmesi - Tam girdi/çıktı kaydı ve denetim izi - Acil kapatma (kill-switch) ve otomasyon sınırları | Kapsam | Yönetişim | Kanıt | | --- | --- | --- | | Guardrail ve politika motoru: girdi/çıktı denetimi, maskeleme, yasaklı aksiyonlar · İnsan onayı kontrol noktaları: eşik, yetkili, süre; onay kartı mobil ve web · Kill-switch kullanıcı, ekip ve kurum bazında; rol tabanlı yetki | Politika değişiklikleri dört göz kuralıyla; her kural sürümlü ve gerekçeli. | Onay kayıtları (kim, ne zaman, neye), guardrail müdahaleleri ve kill-switch olayları. | ## E · Gözlemlenebilirlik & Optimizasyon Performans yönetimi. Doğruluk, hız, maliyet ve sağlık görünürlüğü sağlar; sürekli iyileştirmenin verisini üretir. Her çalıştırma adım adım kayıtlıdır. - Çalıştırma bazlı iz ve zaman çizelgesi - Gecikme ve token takibi - Senaryo, birim ve kanal bazında maliyet atfı - Model bazında kullanım ve maliyet; platform sağlığı - Yönlendirme doğruluğu ölçümü - Pano ve analitik katmanı - Canlı loglar, konuşma kayıtları, ajan bazlı başarı ısı haritası | Kapsam | Yönetişim | Kanıt | | --- | --- | --- | | Çalıştırma bazlı iz: adım, süre, token, araç çağrısı, hata · Panolar: ajan başarı ısı haritası, platform sağlığı, model kullanımı, maliyet atfı · SIEM ve veri ambarına aktarım; uyarı kuralları | İzler değiştirilemez ve saklama süresi politikayla belirlenir. | Denetim örneklemi tek tıkla; ajan bazında aylık kalite raporu. | ## F · Entegrasyon & Bağlantı Kurumsal erişim. Mevcut teknoloji ortamını yeniden kurmadan kurumsal sistemlere ve müşteri kanallarına bağlanır. Her araç için çağrı politikası ayrı tanımlanır. - REST API ve veritabanı entegrasyonları - Model Context Protocol (MCP) sunucuları - Özel araç tanımları ve çağrı politikaları - Hazır bağlayıcılar: e-posta, takvim, dosya, iş birliği araçları - Web widget ve mesajlaşma entegrasyonları - MCP Hub: onaylı araç sunucuları kataloğu | Kapsam | Yönetişim | Kanıt | | --- | --- | --- | | REST, veritabanı ve MCP sunucuları; araç başına çağrı politikası · Hazır bağlayıcılar: çekirdek bankacılık, ERP, CRM, bilet, ödeme, santral · Okuma/yazma ayrımı; yazma araçları varsayılan olarak onay ister | Her araç servis hesabı ve kapsamla tanımlanır; sır yönetimi kurumun kasasında. | Araç çağrı günlüğü: parametreler (maskeli), yanıt kodu, süre. | ## G · Çok Kanallı Deneyim & Ses Müşteri etkileşim katmanı. Dijital, mesajlaşma ve ses kanallarında tek ve tutarlı bir orkestrasyon üzerinde AI sunar. Aynı ajan yazılı ve sesli kanalda çalışır. - Gerçek zamanlı konuşma tanıma ve akışlı ses işleme - Çağrı merkezi SIP entegrasyonu - WhatsApp, SMS ve e-posta servisleri - Niyet tespiti ve varlık çıkarımı blokları - Oturum hafızası - Dialog Aginies: diyalog arayüzü ve insan devralma - Voice Aginies: Türkçe STT, TTS ve ASR modelleri | Kapsam | Yönetişim | Kanıt | | --- | --- | --- | | Dialog Aginies web ve mobil; WhatsApp, SMS ve e-posta kanalları · Voice Aginies: akışlı ASR, Türkçe STT/TTS, SIP ile çağrı merkezi · Aynı ajan tüm kanallarda; sıcak aktarım ve konuşma özeti | Kanal başına kimlik doğrulama seviyesi ve işlem eşiği tanımlanır. | Transkript, kayıt ve özetler saklama politikasıyla; kanal bazında kalite puanı. | ## H · Dağıtım & Kurumsal Mimari Kurumsal kısıtlara uyum. Büyük kurumların on-prem, hibrit ve egemenlik gereksinimlerini destekler. Veri nerede olmalıysa orada kalır. - On-prem ve air-gap kurulum - Hibrit mimari: hassas iş yükleri içeride, esnek kapasite dışarıda - Özel bulut / dedike bölge - SSO / SAML, ince taneli RBAC, workspace izolasyonu - Tenant bazında ajan paylaşımı ve ortak geliştirme - GPU-as-a-Service ile ülke sınırları içinde elastik kapasite | Kapsam | Yönetişim | Kanıt | | --- | --- | --- | | Egemen paylaşımlı (Türkiye’deki veri merkezlerimiz), dedike tenant/GPU, kurum veri merkezinde on-prem veya air-gap · Erişim: VPN tüneli veya internet üzerinden allowlist · GPU-as-a-Service ile ülke içi esnek kapasite | Değişiklik yönetimi ve olay bildirimi sözleşmede; uzaktan bakım kurumun açtığı pencerede. | Ağ şeması, tünel/allowlist konfigürasyonu, yedekleme ve tatbikat kayıtları. | ## Otonom Ajan Otonom Ajan, ajanları diyalogla üretir. Ne istediğinizi yazarsınız; Otonom Ajan süpervizörü ve uzman ajanları kurar, araçları ve bilgi tabanını bağlar, modelleri seçer, onay noktalarını tanımlar, test korpusunu yazıp koşturur ve onayınızla üretime alır. Kütüphanedeki her ajan bu yolla üretilebilir. - Tanım: hedef, kanallar ve onay kuralı doğal dille verilir. - Kurulum: süpervizör ve uzman ajanlar oluşturulur; sürüm numarası atanır. - Bağlantı: araçlar (REST/MCP) ve bilgi tabanı bağlanır; chunk sayısı raporlanır. - Model seçimi: niyet, karar ve ses için ayrı modeller; on-prem tercihleri korunur. - Onay noktaları: eşik ve yetkili tanımlanır (ör. tutar > 100.000 → şube yöneticisi). - Test korpusu yazılır ve koşturulur; kontrol aileleri puanlanır. - İnsan onayı: sonuçlar sunulur; üretime alma kararı her zaman insanda kalır. - Dağıtım: yeni sürüm canlıya alınır; loglar ve pano anında aktif olur. Test korpusu üç seviyede yapılandırılır: suite’ler (altın yol, dallanmalar, dayanıklılık enjeksiyonu), her suite içinde senaryolar ve her senaryoda çok turlu konuşmalar. Her tur, kontrol ailelerine göre puanlanır. - customer_resolved: müşteri talebi gerçekten çözüldü mü - pii_before_verification: kimlik doğrulanmadan kişisel veri paylaşıldı mı - Durum makinesi: konuşma tanımlı durumlar ve geçişler içinde kaldı mı - Hafıza gidiş-dönüşü: önceki turlarda verilen bilgi doğru hatırlandı mı - Yargıç boyutları: doğruluk, ton ve politika uyumu gibi nitel ölçütler - Araç ve onay kontrolleri: doğru araç, doğru parametre, eşik aşımında onay tetiklendi mi Kapsam (coverage), tanımlı niyetlerin, dalların ve araçların korpus tarafından ne oranda test edildiğini gösterir. Düşük kapsam üretime alma öncesi uyarı olarak sunulur; sonuçlar sürüm bazında saklanır ve sürümler arasında karşılaştırılır. Otonom Ajan hiçbir zaman kendi başına üretime çıkmaz. Dağıtım, insan onayından sonra gerçekleşir. ## Bilgi Tabanı Bilgi tabanı, ajanların kaynağa bağlı cevap üretmesini sağlar. Dokümanlar gömme hattından geçer, chunk’lara bölünür, etiketlenir ve sürümlenir. Her ajan hangi kaynaklara, hangi etiketlerle ve hangi erişim politikasıyla ulaşacağını bilir. - Kaynaklar: dosya, klasör, web, veritabanı ve bağlayıcılar - Chunk editörü: her parça incelenir, düzenlenir, devre dışı bırakılır - Sürümleme: hangi ajan hangi bilgi tabanı sürümüyle çalıştı, kayıtlı - Hibrit arama: anlamsal ve anahtar kelime aramanın birleşimi - Erişim politikası: etiket ve rol bazında görünürlük - Kaynak gösterimi: cevaptaki her iddia ilgili chunk’a bağlanır ## Hub Aginies: Model Hub, MCP Hub ve proxy Model kataloğu üç grubu aynı yerde toplar: Qwen, GLM ve Kimi tabanlı, kurumsal veriyle ince ayarlanmış Aginies aileleri (tulpar, talay, sunkar, konrul); aynı tabanların güncel açık sürümleri; Türkçe ses ve görüntü modelleri. Aile adı sabittir, sürüm numarası ilerler: yeni bir taban sürümü çıktığında ince ayar tekrarlanır ve katalogdaki sürüm artar; önceki sürüm sözleşme süresince erişilebilir kalır. Her senaryo için doğruluk, gecikme ve maliyet dengesine göre model seçilir. Aşağıdaki sürümler Ekim 2026 kataloğudur. | Model | Taban | Nerede çalışır | Tipik kullanım | | --- | --- | --- | --- | | aginies-tulpar-4 | Büyük sınıf, kurumsal ince ayar | On-prem · ülke içi GPU | Karmaşık karar, çok adımlı akıl yürütme, süpervizör rolü | | aginies-talay-2 | Orta sınıf, kurumsal ince ayar | On-prem · ülke içi GPU | Genel diyalog, özetleme, çıkarım, uzun doküman | | aginies-sunkar-1 | Küçük sınıf, kurumsal ince ayar | On-prem · ülke içi GPU | Niyet tespiti, yönlendirme, guardrail, kod tamamlama | | aginies-konrul-1 | Görüntü-dil, kurumsal ince ayar | On-prem · ülke içi GPU | Doküman ve görüntü anlama, maskeleme | | Qwen · GLM · Kimi (güncel açık sürüm) | Açık ağırlıklı taban modeller | On-prem · ülke içi GPU | İnce ayar gerektirmeyen görevler, karşılaştırma ve yedek model | | Bulut modelleri (proxy) | Kurumun kendi anahtarı | Bulut, geçit ve maskeleme ile | Kurumun tercih ettiği kapalı modeller; kullanım ve maliyet tek noktadan izlenir | | aginies-voice-tr-v3 | Türkçe için eğitilmiş STT | On-prem | Konuşma tanıma, çağrı yazıya dökme | | aginies-tts-tr-v2 | Türkçe için eğitilmiş TTS | On-prem | Doğal ses sentezi, sesli ajan yanıtları | | aginies-asr-stream-v1 | Akışlı ASR | On-prem | Gerçek zamanlı çağrı merkezi tanıma | Model proxy: kendi bulut anahtarlarınızı girersiniz; platform bulut modellerine sizin adınıza proxy olarak erişir. Tüm çağrılar tek noktadan geçer, izlenir ve maskelenir. Anahtarlar şifreli saklanır; kullanım ve maliyet ajan, birim ve kanal bazında raporlanır. - Model değişimi tek seçimle; yedek model stratejisi sağlayıcı kesintisinde devreye girer - Prompt test ortamı ve sürümleme; sürümler arası karşılaştırma - MCP Hub: onaylı araç sunucuları kataloğu; her sunucu için erişim ve çağrı politikası ## Voice Aginies Türkçe için eğitilmiş konuşma modelleri: akışlı konuşma tanıma (STT), doğal ses sentezi (TTS) ve gerçek zamanlı ASR. Çağrı merkezi SIP entegrasyonuyla sesli ajanlar, yazılı kanaldaki ajanla aynı orkestrasyon üzerinde çalışır. - Modeller: aginies-voice-tr-v3 (STT), aginies-tts-tr-v2, aginies-asr-stream-v1 - On-prem servis; ses verisi kurum dışına çıkmaz - Ses analizi: kalite puanlama, duygu ve niyet - Görüşmelerin %100’ü yazıya dökülür ve kalite formuna göre puanlanır - Kişisel veri, yazıya dökme sırasında maskelenir ## Vision Aginies Görüntü ve doküman anlama: fatura, sözleşme ve form çıkarımı, hasar fotoğrafı değerlendirme, görsel kalite kontrol. Kişisel veri tespiti ve doküman maskeleme aynı modellerle, kurum içinde çalışır. - aginies-konrul-1 · on-prem - Alan çıkarımı: her alan için kaynak belge ve güven skoru - Belge sınıflandırma: sayfa düzeninden bağımsız - Kişisel veri tespiti ve doküman maskeleme (KVKK/GDPR) - Düşük güven skorlu alanlar insan doğrulamasına yönlendirilir ## Code Aginies Geliştirici ekipleri için: IDE eklentileri ve kurum içi kod tamamlama, onaylı eklenti ve model dağıtımı yapan kurumsal IDE registry ve CI/CD hattında çalışan SDLC ajan ekibi. Kod kurum dışına çıkmaz. - Kurum içi kod tamamlama; on-prem modellerle - IDE Registry: onaylı eklenti ve model dağıtımı, sürüm kontrolü - SDLC ajan ekibi: PM, analist, mimar, geliştirici, test - Jira, Git ve CI ile çalışır; kod incelemesi ve üretim kararı insanda ## Dialog Aginies Müşteri ve çalışan tarafındaki diyalog yüzeyi: web widget, mobil, WhatsApp ve mesajlaşma. Oturum hafızası, insan devralma ve her mesaj için denetim izi. Aynı ajan yazılı ve sesli kanalda çalışır. - İnsan devralma ve geri verme; bağlam temsilciye özetle aktarılır - Konuşma panosu ve kalite puanlama - Niyet tespiti ve varlık çıkarımı blokları - Mevcut kimlik doğrulama sürecine entegrasyon; yetkisiz talep guardrail’de durur ## Gözlemlenebilirlik ve raporlama Ajan bazında ne oluyor, platform ne durumda, hangi model ne kadara mal oluyor: tek panoda. Her çalıştırma adım adım kayıtlıdır; loglar ve konuşma kayıtları rol bazlı erişimle incelenir. - Canlı çalıştırmalar: adım, süre, token, hedef sistem ve sonuç - Ajan bazında başarı ısı haritası ve hata zaman çizelgesi - Kalite ve hata zaman çizelgeleri; yönlendirme doğruluğu - Maliyet atfı: senaryo, birim, kanal ve şube bazında - Model bazında kullanım ve maliyet; platform sağlığı - Kapasite projeksiyonu: kullanım artışında maliyet ve kaynak tahmini - Denetim raporlaması: girdi, çıktı, model sürümü, kural seti, kullanıcı Raporlar dışa aktarılabilir ve iç denetim ile düzenleyici taleplerine uygun biçimde saklanır. Konuşma kayıtları maskelenmiş hâlde tutulur; maskesiz erişim ayrı yetkilendirme gerektirir. ## Workspace, tenant ve iş birliği Platform workspace ve tenant bazında ayrılır. Her ekip kendi ajanlarını, bilgi tabanlarını ve anahtarlarını izole biçimde yönetir; onaylanan ajanlar kurum içinde paylaşılır ve birlikte geliştirilir. - SSO / SAML ile kimlik; ince taneli RBAC ile yetki - Workspace izolasyonu: veri, log ve maliyet ayrı raporlanır - Tenant bazında ajan paylaşımı ve ortak geliştirme - Şablonlar, kullanıcılar ve lisanslar merkezi yönetimde - Ajan sürümleri ve yayın geçmişi her workspace için kayıtlı ## Kütüphaneden türetilmiş platform gerçekleri Platform sayfasındaki sayılar ve dağılımlar ajan kütüphanesinden derleme sırasında türetilir. Aynı veri burada tablo olarak verilir. | Gösterge | Değer | Kaynak | | --- | --- | --- | | Canlı senaryo | 77 | Ajan kütüphanesi | | Sektör | 14 | Senaryo tanımları | | Uzman ajan | 261 | Senaryo tanımları | | Farklı araç tanımı | 657 | Ajan araç listeleri | | Bağlı sistem | 200 | Senaryo sistem listeleri | | İnsan onayı kuralı olan senaryo | 73 | Onay kuralları | | Kurumsal sistemde işlem adımı | 127 | Adım türleri | | Simüle çalıştırma / ay (toplam) | 3.320.970 | Senaryo metrikleri (tahmini) | | Kanal | Senaryo | | --- | --- | | api | 37 | | app | 34 | | web | 29 | | whatsapp | 27 | | email | 27 | | batch | 25 | | voice | 23 | | sms | 5 | | branch | 3 | | Adım türü | Adım sayısı | | --- | --- | | sistemde işlem | 127 | | araç çağrısı | 110 | | akıl yürütme | 109 | | doğrulama | 89 | | bildirim | 80 | | sinyal alma | 77 | | insan onayı | 61 | | planlama | 48 | | bilgi getirme | 38 | | Model ailesi | Ajan ataması | | --- | --- | | aginies-tulpar | 83 | | aginies-sunkar | 78 | | aginies-talay | 71 | | aginies-konrul | 21 | | aginies-voice-tr | 8 | Bu tablo derleme sırasında ajan kütüphanesinin senaryo tanımlarından türetilir; senaryo eklendikçe değişir. Hacim ve maliyet değerleri simülasyon tanımlarıdır, tahminidir. ## Entegrasyon kataloğu Senaryoların dokunduğu 200 sistem, 12 türde. Yeni bir sistem bağlamak bir araç tanımı ve bir çağrı politikası demektir; yazma araçları varsayılan olarak onay ister. | Sistem türü | Bağlantı | Kütüphanedeki örnekler | | --- | --- | --- | | Kurumsal API’ler | 76 | Access Control, Appointment System, Bank SIM-Swap Signal API, CI Pipeline, CI Runner (GitHub Actions), CI/CD, CI/CD (signed artefacts), Card Management, Card Network, Carrier API, Carrier API (Yurtiçi, DHL), Cart & Checkout API, Civil Registry / Income Services, Connection Portal, Credit Bureau (KKB / Findeks), Credit Bureau (KKB), Customs Broker Portal, Dealer Portal, Dependency Scanner (Snyk), Device / Browser Grid, Device Cloud (MQTT), Enrichment API, Figma, Fraud Engine, Git (self-hosted), GitHub, Insurance Registry (SBM), Insurer Portal, Kubernetes / ArgoCD, Lab / Imaging, MASAK Reporting Portal, Merchant Checkout API, Merchant Portal, Network OSS, OSS Network, OTA Service, Observability (Grafana / Datadog), Open Banking API, Outage Management, Parts & Labour Catalogue (Audatex), Payer Portal, Product Catalogue / Offer Engine, Rating Engine, Repair Network, Repair Network Portal, Review Platform API, Sanctions Screening, Sanctions Screening (Dow Jones), Scheduling, SonarQube, SonarQube + Snyk, Static Analysis, Supplier Portal, Telematics, Terminal Management, Trade Registry (MERSİS), Visa / Mastercard Dispute API, Xcode Cloud, e-Devlet Gateway, e-Government Gateway, e-Invoice Portal, e-Procurement Platform, e-Sign | | Müşteri kanalları | 50 | Customer Portal, Hangar Tablet App, Microsoft Teams, Mobile App Chat, OTP / SMS Gateway, Push & SMS Gateway, SMS Gateway, Shift Channel (Teams), Slack, Slack / Teams, Technician App, Web Chat Widget, WhatsApp Business, WhatsApp Business API | | Çekirdek sistemler | 44 | Active Directory / Entra ID, BSS / Billing, Billing (Amdocs), Billing (CIS), CLM (Contract Lifecycle Management), Card Management, Claims System, Core Banking, Core Ledger, Device Management / OTA, EHR (HL7/FHIR), Guidewire ClaimCenter, HIS Scheduling, HLR / Provisioning, Hospital Information System, Hospital Information System (HIS), IoT Platform, IoT Platform (MQTT / Digital Twin), Loan Origination System, Policy Admin, Policy Admin System, SCADA / DMS, SCADA / Historian, Social Assistance Information System, TMS (Transport Management), Underwriting Workbench, WMS | | Veritabanı ve vektör | 44 | Audit Log, Collections Platform, Control & Sanctions Lists (vector), Eligibility Criteria (vector), Fleet Registry, GIS / Network Model, GIS / Zoning Plan, KYC / Customer 360, Knowledge Base · kart_sss_v3, Known Issues (vector), Legal Playbook Vector Store, Legislation Vector Store, Mailbox Rules & Checklists Vector Store, Manuals & Work Order Vector Store, Meter Data Management, Observability (metrics), PIM / Catalogue, Payer Policy Vector Store, Product Catalogue (PIM), QA Rubric Vector Store, Regulation Vector Store, Regulation Vector Store (EPDK), SCADA Historian, Security Requirements (vector), Smart Meter Data, Specification Vector Store, Stability Rules (vector), Tariff & Regulation Vector Store, Technical Publications (S1000D vector), Time-series DB, Tolerance Policy Vector Store, Triage Protocol Vector Store, Usage Mediation / CDR Store, Vector Store, Vendor Master / Procurement, Warranty Policy Vector Store, Workforce Work Log | | Bilet ve iş emri | 25 | CMMS, CMMS (Maximo), Case Management, Field Service, ITSM / Incident, Jira, Jira (QA), Jira / Xray, Network OSS, QM Scorecard / Case Management, ServiceNow, Ticketing, Trade Compliance Case System, Zendesk | | ERP ve MES | 25 | Billing / CIS (SAP IS-U), Billing / Settlement, ERP, ERP (SAP / Logo), ERP (SAP GTS), ERP Billing, ERP Spare Parts, Inventory, MES, MRO System, Maintenance Management System, Oracle EBS, SAP ERP, SAP S/4HANA, SAP S/4HANA MM, Shopify / OMS, Spares Inventory, Supplier Management (SRM), TMS, WMS / Lot Tracking | | E-posta | 20 | E-mail, Email, Email (Exchange), Exchange / M365 Mail, Procurement Inbox | | Doküman depoları | 19 | Confluence, Contract Repository, Document Store, Document Store (S3), Tech Pub Repository (S1000D) | | CRM | 8 | CRM, Customer CRM / Billing, Customer Profile, HubSpot, Salesforce CRM | | Ödeme sistemleri | 7 | Bank API, PSP / Acquirer API, Payment Gateway (iyzico), Payment Hub, Refund Service, Settlement Ledger | | Ses ve santral | 6 | Call Centre SIP, Genesys SIP, IVR / SIP | | İK sistemleri | 3 | LMS, SAP SuccessFactors, Workday | ## İşletim modeli Canlıya çıktıktan sonra platformun standart işletim modeli. Hedefler sözleşmede kurumla birlikte netleşir. | Konu | Uygulama | | --- | --- | | Sürümleme ve geri alma | Ajan tanımı, kural seti, istem ve model sürümü birlikte kaydedilir; bir önceki sürüme dönüş tek adımdır. | | Test korpusu kapısı | Her değişiklik üretime çıkmadan suite, senaryo ve tur bazlı test korpusundan geçer; kapsama raporu kayıt altındadır. | | Süreklilik hedefleri | Egemen paylaşımlı kurulumda hedef RTO 4 saat, RPO 1 saat; dedike ve on-prem kurulumda kurumla birlikte belirlenir. Yedekler Türkiye içinde. | | Model kataloğu bakımı | Yeni taban sürümü çıktığında ince ayar tekrarlanır, katalogda yeni sürüm açılır; eski sürüm sözleşme süresince kalır. | | Olay yönetimi | Önem derecesine göre bildirim süreleri sözleşmede; kill-switch kurumda, kök neden raporu Aginies’te. | | Aylık gözden geçirme | Ajan bazında kalite, maliyet ve onay oranları; eşik ve kural güncellemeleri kurumun yöneticisiyle birlikte. | ## Ürün ekranları Platform Aginies · Pano · son 30 güntüm workspace’ler · tüm tenant’lar çalıştırma13.120.000 başarı%97,4 p50 gecikme1,4 s platform sağlığısağlıklı en yoğun ajanlar · her hücre ≈ 1 gün dialog_banking_assistant%97,4 card_debt_payment%99,1 kvkk_data_detection%96,2 kvkk_document_masking%98,8 motor_claim_end_to_end%98,1 merchant_operations%100 model kullanımı aginies-tulpar-438% aginies-talay-229% aginies-sunkar-119% aginies-voice-tr-v39% cloud · proxy$2.960 maliyet · senaryo bazında Diyalog bankacılığı26% Kredi ve edinim24% Hasar operasyonları18% KVKK maskeleme14% Üye işyeri10% Diğer8% PII maskeleme: açıkanonimleştirme: açıkdenetim izi: tam1.920+ aktif ajan Platform Aginies · Raporlartüm workspace’ler · Ekim 2026 aylık çalıştırma · son 12 ayson 30 gün: 13.120.000+%31 yıllık | rapor | kırılım | zamanlama | biçim | | --- | --- | --- | --- | | Aylık maliyet raporu | senaryo · birim · kanal · şube | her ayın 1’i · 07:00 | PDF · XLSX | | Ajan performansı | başarı, gecikme, hata sınıfları | haftalık · Pazartesi | PDF | | Denetim dışa aktarımı | çalıştırma izleri, onaylar, model sürümleri | talep üzerine | CSV · JSON | | Model kullanımı | model bazında token, maliyet, proxy payı | günlük · 06:00 | XLSX | | SLA ve kapasite | p95 gecikme, kuyruk, GPU doluluk, projeksiyon | haftalık | PDF | e-posta ve paylaşılan klasöre gönderimiç denetim şablonukişisel veri maskeli Dialog Aginies · web widgetcard_debt_payment v2 Kart borcumun tamamını vadesiz hesabımdan ödemek istiyorum. Dönem borcunuz 24.860 TL, son ödeme 14 Ekim. Vadesiz ****7741 hesabınızda bakiye yeterli. Onaylarsanız tek kullanımlık şifre göndereceğim. ✓ get_card_debt · get_accounts · politika: onay gerekmez Onaylıyorum. Şifre gönderildi. Girdiğinizde ödeme anında yapılır ve teyit mesajı gelir. oturum hafızasıinsan devralma: hazırdenetim izi Mesaj yazın…↑ Voice Aginies · SIP · çağrı 0212-…-4471aginies-voice-tr-v3 · on-prem müşteriDün gece arabama çarptılar, plaka 34 ABC 123, hasar ihbarı açmak istiyorum. ajanGeçmiş olsun. Poliçenizi plakadan buldum; olay yerini ve saati alabilir miyim? STT180 ms niyethasar ihbarı · 0,97 duygusakin TTS · aginies-tts-tr-v2ASR · stream-v1PII maskeleme: açık Vision Aginies · fatura_2026_1012.pdfaginies-konrul-1 · on-prem [Alan tespiti yapılmış fatura] çıkarılan alanlar Fatura noFTR-2026-1012 Tarih12.10.2026 KDV%20 · 3.125,00 Toplam18.750,00 TL TCKNmaskelendi (KVKK) kaynak: sayfa 1güven ≥ 0,95 Code Aginies · billing-service/invoices.tskod tamamlama · on-prem 14export async function exportInvoicePdf(id: string) { 15 const invoice = await repo.findById(id); 16 if (!invoice) throw new NotFound('invoice'); 17 const pdf = await renderer.render('invoice', maskPii(invoice)); 18 await audit.log('invoice.export', { id, user: ctx.user }); 19 return storage.put(`invoices/${id}.pdf`, pdf); 20} Tab: kabul etPROJ-4821test kapsamı %87SAST ✓ IDE RegistryVS Code · JetBrains eklentisiGeliştirici Ajanı · PR #2291 Hub Aginies · Model Hubtüm workspace’ler · Ekim 2026 | model | erişim | kullanım · son 30 gün | maliyet | | --- | --- | --- | --- | | aginies-tulpar-4 | on-prem · GPU 4× | 58% | — | | aginies-talay-2 | on-prem · GPU 2× | 84% | — | | aginies-sunkar-1 | on-prem · GPU 1× | 91% | — | | aginies-konrul-1 | on-prem · vision | 22% | — | | qwen · güncel sürüm | on-prem · açık model | 31% | — | | aginies-voice-tr-v3 | on-prem · STT | 47% | — | | bulut · proxy | kendi anahtarınız | 12% | $288 | anahtarlar şifreli · kullanıcıya görünmeztüm istekler maskeliMCP Hub · 14 sunucu Workforce Aginies · Kadro ve iş kaydıworkspace: fintech-ops · bu hafta · tüm tenant’lar ayrı | ajan | rol · ekip | saat karş. | çıktı | kalite | mod | | --- | --- | --- | --- | --- | --- | | ios-dev-01 | iOS Geliştirici · mobil ekibi | 14 h | 9 PR · 21 puan | 4,7 | otonom · %10 inceleme | | qa-01 | Test Mühendisi · mobil ekibi | 31 h | 412 test · 6 hata | 4,8 | otonom · %10 inceleme | | ba-01 | İş Analisti · ödeme ürünü | 9 h | 11 hikâye · 2 süreç | 4,5 | gölge · %100 inceleme | | acct-01 | Muhasebe · finans | 27 h | 238 kayıt · 3 mutabakat | 4,9 | otonom · eşik onaylı | iş kaydı · bugün 09:12ios-dev-01PR #4821 açıldı: ödeme ekranı erişilebilirlik düzeltmeleri · 2 test eklendiinceleme bekliyor 09:40acct-0138 tedarikçi faturası işlendi, 2 tanesi tolerans dışı → m.ozkan onayınaonay 10:05qa-01Regresyon 3. tur: 412/412 · 1 yeni hata (BUG-2291) kanıtla açıldıtamam 10:30ba-01PAY-118 hikâyesi ve 6 kabul kriteri yazıldı → ürün sahibineinceleme haftalık zaman çizelgesi dışa aktarıldıfaturalama: ajan-ayekip lideri: e.demir (Aginies) --- # Metodoloji ve küresel desenler · Dokümanlar URL: https://www.aginies.com/tr/dokumanlar/processes Language: tr Summary: Yönetim kurulu çerçevesi, beş soru, taker/shaper/maker, beş katmanlı değerlendirme, dönüşüm yolculuğu, bankacılık örneği ve ajanların yansıttığı küresel desenler. Dokümanlar Bu sayfada Yönetim kurulu çerçevesi: iş ve yargıYönetim kurulunun cevaplaması gereken beş soruBenimseme stratejisi: taker, shaper, makerBeş katmanlı değerlendirmeDönüşüm yolculuğuBankacılık örneği: çekirdek yerinde kalırKüresel desenler Süreçler ve global örnekler # Metodoloji ve küresel desenler Yönetim kurulu çerçevesi, beş soru, taker/shaper/maker, beş katmanlı değerlendirme, dönüşüm yolculuğu, bankacılık örneği ve ajanların yansıttığı küresel desenler. Güncelleme: Eylül 2026 ## Yönetim kurulu çerçevesi: iş ve yargı Yapay zekâ, işin yanında yargıyı da otomatikleştiren ilk teknolojidir. Önceki dalgalar veri girişini, hesaplamayı ve iletişimi hızlandırdı; kararı hep insana bıraktı. Ajanlar ise sınıflandırır, önceliklendirir, önerir ve tanımlı sınırlar içinde karar verir. Bu yüzden konu işletim modelinin yeniden tasarımıdır: hangi kararlar ajanlara devredilir, hangileri insanda kalır, sınırlar nasıl ölçülür ve denetlenir. Cevaplar yönetim kurulu seviyesinde verilmelidir. ## Yönetim kurulunun cevaplaması gereken beş soru - Değer havuzu: AI’ın kurumda yaratabileceği değer nerede ve ne büyüklükte? - Süreçler: hangi süreçler ajanlarla yeniden tasarlanacak, hangileri yalnızca hızlanacak? - Organizasyon ve yetenek: önümüzdeki 24 ayda roller, beceriler ve yapı nasıl değişecek? - Benimseme stratejisi ve platform: taker, shaper ya da maker; hangi platform kararları verilecek? - Yönetişim: kararlar nasıl sınırlanacak, denetlenecek ve gerektiğinde durdurulacak? ## Benimseme stratejisi: taker, shaper, maker Her kurum neyi kendisi inşa edeceğini seçer. Doğru soru, hangi alanda hangi rolün üstlenileceğidir. Çoğu kurum üç stratejiyi senaryo bazında karıştırır. | Strateji | Anlamı | Ne zaman uygun | | --- | --- | --- | | Taker | Hazır AI yeteneklerini olduğu gibi kullanmak | Standart, farklılaştırmayan işler; hızlı fayda | | Shaper | Platform üzerinde kurumun verisi ve süreçleriyle ajanlar kurmak | Rekabet avantajı yaratan çekirdek süreçler; egemenlik gereksinimi | | Maker | Kendi modellerini ve altyapısını geliştirmek | Çok büyük ölçek, benzersiz veri ve güçlü mühendislik kapasitesi | Aginies, shaper stratejisinin platformudur: kurum kendi verisi, süreçleri ve sınırlarıyla ajanlarını kurar; model ve altyapı seçimlerini bağımlılık yaratmadan yapar. ## Beş katmanlı değerlendirme Hangi AI fikrinin yapılmaya değdiğini ve ne zaman başlanacağını sayılarla belirleriz. Her senaryo beş katmandan geçer ve her katman net bir kararla biter: devam, düzelt ya da durdur. | Katman | Soru | Çıktı | | --- | --- | --- | | 1 · Değer | Bu iş neden yapılmalı? | Sayılarla konuşan iş gerekçesi ve önceliklendirilmiş senaryo portföyü | | 2 · Hazırlık | Gerekli girdiler elimizde var mı? | Eksiklerin proje öncesi kapatıldığı hazırlık karnesi | | 3 · Uygulanabilirlik | Bunu gerçekten yapabilir miyiz? | Net kapsamlı, riskleri tanımlı, gerçekçi MVP planı | | 4 · Operasyon | Canlıya çıktıktan sonra nasıl yaşayacak? | Yaşayan, ölçülen, sürekli iyileşen AI ürünü | | 5 · Yönetişim | Güvenli, kontrollü ve denetlenebilir mi? | Denetime hazır, kontrollü, güven veren AI işletimi | Değer katmanı kontrolleri: - Problem: hangi iş problemi, bugün nasıl çözülüyor, maliyeti ne? - Fayda: maliyet düşüşü, gelir, risk azaltımı, deneyim; hangisi, ne kadar? - Ölçüm: hedef KPI’lar, beklenen iyileşme, ROI ve geri dönüş süresi - Stratejik uyum: şirket stratejisiyle hizalı mı, yönetim önceliği mi? Hazırlık katmanı kontrolleri: - Veri: kaynaklar mevcut ve dijital mi; kalite, güncellik, sahiplik net mi? - Entegrasyon: sistemler ve API’ler hazır mı; gerçek zamanlı erişim mümkün mü? - Organizasyon: iş birimi sahipleniyor mu; sponsor ve uzmanlar belli mi? - Teknoloji: bulut/güvenlik altyapısı, AI platformu ve lisanslar hazır mı? Uygulanabilirlik katmanı kontrolleri: - Teknik uygunluk: problem AI ile mi çözülmeli; klasik yazılım, klasik ML, ajan? - Karmaşıklık: kaç sistem, kaç veri kaynağı, kaç kullanıcı etkilenecek? - Model performansı: beklenen doğruluk, yanlış sonucun etkisi, insan onayı gerekli mi? - Proje uygulanabilirliği: MVP süresi, ekip, dış bağımlılıklar, teknik riskler Operasyon katmanı kontrolleri: - İşletim modeli: ürün sahibi, süreç sahibi, operasyon ve destek ekibi tanımlı - Kullanıcı yönetimi: eğitim, kullanım takibi, geri bildirim döngüleri - İzleme: performans, metrik ve hata yönetimi gerçek zamanlı - Finansal operasyon: aylık işletme maliyeti, büyüme projeksiyonu, bütçe sahipliği Yönetişim katmanı kontrolleri: - Güvenlik: hassas veri haritası, veri yerleşimi, modele veri akışının kontrolü, maskeleme - Uyum: KVKK ve GDPR etkisi, sektörel regülasyonlar, iç politika gereklilikleri - Risk: halüsinasyon, yanlış karar, veri sızıntısı ve prompt injection değerlendirmesi - Kontrol: insan onayı noktaları, otomasyon sınırları, acil kapatma, karar geçmişi Devam: beş katmanda temel hazır. Düzelt: senaryo değerli, bir iki katmanda eksik var; proje öncesi kapatılır. Durdur: bu hâliyle başlamak bütçe yakar; bunu söylemek de metodolojinin parçasıdır. Her katmanın süresi, katılımcıları, araçları ve kırmızı bayrakları: | Katman | Süre | Katılımcılar | Araçlar ve çıktılar | Kırmızı bayraklar | | --- | --- | --- | --- | --- | | Değer | 1–2 hafta | İş birimi sahibi, finans, sponsor | Senaryo portföy çalıştayı ve değer kartı (problem, hacim, maliyet, KPI) · Ajan kütüphanesindeki benzer senaryoyla karşılaştırma | Sayısız fayda: “verimlilik artacak” ama hacim ve maliyet yok · Sponsor yok; talep BT’den geliyor, iş birimi izliyor | | Hazırlık | 1–2 hafta | Veri sahibi, entegrasyon ekibi, süreç sahibi | Hazırlık karnesi: veri, entegrasyon, süreç, organizasyon (her eksen 0–2) · API ve veri erişim testi; örnek veri seti | Aksiyon alınacak sistemin API’si yok, yalnızca ekran · Süreç kuralları yalnızca kişilerin tecrübesinde yaşıyor | | Uygulanabilirlik | 2–3 hafta | Aginies mühendisliği, kurum mimarisi, bilgi güvenliği | Yaklaşım kararı: klasik yazılım, klasik ML, ajan, üretken AI · Prototip ve test korpusu taslağı; yanlış çıktı etki analizi | Yanlış çıktı geri alınamaz sonuç doğuruyor ve onay noktası tanımsız · Kapsam her toplantıda büyüyor | | Operasyon | 1 hafta (plan) + canlı sonrası sürekli | Ürün sahibi, süreç sahibi, operasyon ve destek | İşletim modeli şablonu: roller, eşikler, inceleme örneklemi, bütçe sahibi · Panolar ve aylık gözden geçirme takvimi | Canlıdan sonra kimin işleteceği belirsiz · Aylık işletme maliyeti için bütçe sahibi yok | | Yönetişim | 2 hafta, diğer katmanlarla paralel | Bilgi güvenliği, uyum, hukuk, iç denetim | Veri haritası ve maskeleme kural seti; onay matrisi · Kontrol seti ve kanıt listesi; sektörel mevzuat eşlemesi | Kişisel veri var, haritası yok · Yönetişim “sonra ekleriz” diye erteleniyor | Örnek: diyalog bankacılığı asistanı beş katmandan nasıl geçti: - Değer: Çağrı merkezine düşen basit işlemler ve işlem başına maliyet sayılarla kondu; hedef %40 azalma. Sponsor: dijital bankacılık. - Hazırlık: Çekirdek bankacılık ve kart sistemlerinin işlem API’leri ve test ortamı mevcut; OTP altyapısı hazır; politika eşikleri bilgi güvenliğiyle yazıldı. - Uygulanabilirlik: Yaklaşım: süpervizör + dört uzman ajan; yanlış çıktı riski için işlem bazlı OTP ve onay bandı; test korpusu 212 senaryo. - Operasyon: Ürün sahibi dijital kanallar, süreç sahibi operasyon; aylık gözden geçirme ve eşik güncellemeleri takvimde. - Yönetişim: Kişisel veri haritası ve maskeleme kuralları; limit artışı > %60, itiraz > 5.000 TL ve yeni alıcı onaya; tüm sistemler Türkiye’de. ## Dönüşüm yolculuğu Değerlendirme dört aşamalı bir yolculuğun içinde tekrar eder. Her aşamanın çıktısı bir sonrakinin girdisidir ve her aşama aynı platform üzerinde ilerler. - Keşif ve portföy: değer havuzları, aday senaryolar, önceliklendirilmiş portföy ve sponsorlar. - Değerlendirme: her aday senaryo beş katmandan geçer; devam, düzelt ya da durdur kararı. - MVP ve doğrulama: ajan kurulur, test korpusu koşturulur, gerçek veriyle KPI’lar doğrulanır. - Üretim ve ölçek: kademeli yetki, izleme, maliyet atfı; portföydeki sonraki senaryoya geçiş. ## Bankacılık örneği: çekirdek yerinde kalır Bankada çekirdek sistem yerinde kalır; AI orkestrasyon katmanı onun üzerine oturur. Çekirdek servisler MCP ve REST üzerinden ajanlaştırılır; ajanlar hesap, kart, kredi ve ödeme servislerini araç olarak görür. Yetki kademeli açılır. - Okuma: bakiye, ekstre, limit ve işlem sorguları; sıfır yazma riski ile ölçüm başlar. - Kontrollü yazma: limit artışı, transfer ve talimatlar eşik ve onay kuralıyla yazılır. - Otonom: kanıtlanmış senaryolarda tanımlı sınırlar içinde onaysız yürütme; kill-switch her an hazır. - Diyalog bankacılığı: mobil ve sesli kanalda işlem yapan asistan - Çekirdek bankacılık AI dönüşümü: çekirdek servislerin kademeli ajanlaştırılması - Operasyonda gerçek zamanlı görev takibi: talep, SLA ve darboğaz görünürlüğü - Müşteri memnuniyeti yönetimi: görüşme analizi, şikâyet ve kırılganlık sinyalleri Her senaryo gözlemlenebilir ve maliyetlendirilir: çalıştırma bazlı iz, birim, kanal ve şube bazında maliyet atfı, kalite ve hata zaman çizelgeleri, kapasite projeksiyonu ve denetim raporlaması. Uyum çerçevesi: BDDK bilgi sistemleri düzenlemesi, birincil ve ikincil sistemler ülke içinde, kritik kararlarda kill-switch ve zorunlu insan onayı. ## Küresel desenler Kütüphanedeki ajanlar dünyada kanıtlanmış uygulama desenlerini yansıtır. Aşağıdaki liste bu desenleri özetler; hacim ve maliyet değerleri simülasyon tanımlarıdır. - FNOL otomasyonu: büyük sigortacılarda hasar bildiriminin ilk temasta, çağrı merkezi yükü olmadan açılması. - AP otomasyonu: ortak hizmet merkezlerinde fatura posta kutusundan ERP kaydına üç yönlü eşleşme. - KYB onboarding: ödeme kuruluşlarında dakikalar süren self-servis üye işyeri edinimi. - Agentic SDLC hattı: büyük yazılım kuruluşlarında talepten üretime ajan destekli teslimat. - WISMO azaltma: e-ticarette proaktif bildirimle “siparişim nerede” taleplerinin self-servise çevrilmesi. - AML alarm triyajı: alarmların zenginleştirilip anlatı taslağı çıkarılması; karar analistte. - Kesinti iletişimi: operatör ve dağıtım şirketlerinde proaktif bilgilendirmeyle gelen çağrı hacminin düşürülmesi. - Diyalog bankacılığı: uygulama içi asistanın bilgi vermekle kalmayıp işlemi tamamlaması. - %100 kalite kapsaması: BPO ve banka çağrı merkezlerinde örnekleme yerine tüm görüşmelerin puanlanması. - Ön yetki otomasyonu: sağlık sistemleri ve ödeyicilerde belge toplama ve kural kontrolünün otomasyonu. - Dijital ön kapı: hasta triyajı ve semptom değerlendirmesinin klinik karar hekimde kalarak otomasyonu. - Sıfır dokunuşlu işe giriş: büyük kurumlarda hesap, donanım ve eğitim atamalarının otomatik koordinasyonu. --- # Güvenlik, uyum ve egemenlik · Dokümanlar URL: https://www.aginies.com/tr/dokumanlar/security Language: tr Summary: Dağıtım seçenekleri, uyum matrisi, model yönetişimi, veri yerleşimi ve keşif oturumunda paylaşılanlar. Dokümanlar Bu sayfada Dağıtım seçenekleriUyum matrisiModel yönetişimiVeri yerleşimi ve egemenlikKod asistanı senaryosu: IDE’den modeleKeşif oturumunda neler paylaşılırSektörel uyum çerçeveleri Güvenlik ve uyum # Güvenlik, uyum ve egemenlik Dağıtım seçenekleri, uyum matrisi, model yönetişimi, veri yerleşimi ve keşif oturumunda paylaşılanlar. Güncelleme: Eylül 2026 ## Dağıtım seçenekleri AI, kurumun diğer kritik sistemleriyle aynı standartlara tabi olmalıdır. Aginies finans, sigorta ve sağlık gibi regüle sektörlerin gereksinimlerine göre kurgulandı; veri nerede olmalıysa orada kalır. Dört dağıtım modeli aynı platformu sunar. | Seçenek | Veri nerede | Modeller nerede | Ne zaman uygun | | --- | --- | --- | --- | | On-premise / air-gap | Kurumun kendi veri merkezi; dış bağlantı yok | İçeride servis edilen oss ve ses modelleri | En hassas veri; düzenleyici zorunluluk | | Hibrit | Hassas iş yükleri içeride | İçeride oss; dışarıda yalnızca maskelenmiş veriyle, güvenli geçit üzerinden | Esnek kapasite ihtiyacı; karma senaryo portföyü | | Özel bulut / dedike bölge | Kuruma ayrılmış tek kiracılı bölge | Bölge içinde; veri yerleşimi garantili | Veri merkezi işletmek istemeyen, yerleşim garantisi isteyen kurumlar | | GPU-as-a-Service | Ülke sınırları içinde | Kullanım bazlı ölçeklenen elastik GPU kapasitesi | Sermaye yatırımı olmadan on-prem model servisi | ## Uyum matrisi Aginies, SOC 2 Type II ve ISO 27001 uyumlu bir platformdur; kontrol seti iç denetim için belgelidir. Aşağıdaki matris uyum alanlarını ve platformdaki karşılıklarını özetler. | Alan | Durum | Kapsam | | --- | --- | --- | | SOC 2 Type II | Uyumlu platform | Kontrol seti iç denetim için belgelidir | | ISO 27001 | Uyumlu platform | Bilgi güvenliği yönetimi | | KVKK & GDPR | Uyum | Kişisel veri haritası, veri yerleşimi garantisi, silme ve erişim talepleri | | BDDK bilgi sistemleri | Uyumlu kurulum | Bankalar için birincil ve ikincil sistemler Türkiye’de; denetime hazır kurulum | | SSO / SAML & RBAC | Yerleşik | İnce taneli roller, workspace izolasyonu, ajan bazlı erişim politikaları | | PII maskeleme & anonimleştirme | Yerleşik | Veri girişinde maskeleme, yapılandırılabilir redaksiyon, tüm isteklerin anonimleştirilmesi | | Denetim izi | Yerleşik | Her ajan aksiyonu ve model çağrısı için girdi, çıktı, model sürümü, kural seti, kullanıcı | | Kill-switch | Yerleşik | Kritik kararlarda acil kapatma ve otomasyon sınırları; zorunlu insan onayı | “Uyumlu platform” ifadesi, kontrol setinin ilgili standartlara göre kurgulandığını ve belgelendiğini anlatır; bağımsız sertifikasyon ayrı bir süreçtir ve ayrıca belgelenir. ## Model yönetişimi Modelden bağımsız, ince ayarlı. Hub Aginies’te kapalı modellerin kurumsal ince ayarlı sürümleri, açık kaynak tabanlı on-prem modeller ve Türkçe ses modelleri aynı katalogdadır. Vendor lock-in yoktur; model değişimi ajanı yeniden yazmayı gerektirmez. - Model proxy: bulut modellerine kurumun kendi anahtarlarıyla, tek noktadan erişim - Anahtarlar şifreli saklanır; kullanım ajan, birim ve kanal bazında raporlanır - Dış modele giden her istek maskelenir ve anonimleştirilir - Yedek model stratejisi: sağlayıcı kesintisinde operasyon sürer - Prompt sürümleme: hangi ajan hangi prompt ve model sürümüyle çalıştı, kayıtlı - Hassas senaryolar için on-prem oss modeller; veri kurum dışına çıkmaz ## Veri yerleşimi ve egemenlik Egemen tasarım, verinin ve modelin kontrolünün kurumda kalması demektir. On-prem ve air-gap kurulumda hiçbir veri kurum dışına çıkmaz; hibrit kurulumda dışarı yalnızca maskelenmiş veri, güvenli geçit üzerinden gider. GPU-as-a-Service ülke sınırları içinde elastik kapasite sağlar. - Kişisel veri haritası: hangi veri, hangi sistemde, hangi amaçla, ne kadar süre - Veri yerleşimi garantisi: dedike bölge ve on-prem seçeneklerinde sözleşmeyle - Bankalar için birincil ve ikincil sistemler ülke içinde - Silme ve erişim talepleri: KVKK ve GDPR süreçlerine uygun kayıt ve yanıt - Ses ve kod verisi: Voice ve Code Aginies on-prem çalışır ## Kod asistanı senaryosu: IDE’den modele Kapalı devre çalışmak isteyen kurumların bilgi güvenliği ekipleri için. Kurumun geliştiricileri IDE’lerindeki Code Aginies eklentisiyle, kurumun VPN’i üzerinden, Türkiye’de konumlu ve BDDK mevzuatına uyumlu veri merkezlerindeki altyapımızdaki Hub Aginies’e bağlanır; kodlama modelleri orada, ülke içi GPU’larda çalışır. Kod bağlamı Türkiye dışına çıkmaz, hiçbir dış model çağrılmaz, kurum verisiyle model eğitilmez, her istek kullanıcı bazında iz bırakır. [Kurum IDE’lerinden VPN tüneliyle Aginies egemen sınırındaki Hub Aginies’e, oradan ülke içi GPU model sunucularına giden istek yolu; dış model çağrısı ve eğitim yok.] - Kimlik: Geliştirici IDE’de kurumun kimlik sağlayıcısıyla (SAML / OIDC, MFA) oturum açar. Eklenti kullanıcıya ait kısa ömürlü bir belirteç taşır; lisans anahtarı kurum düzeyinde tutulur ve eklentiden ayrı durur. - İstek: Eklenti yalnızca ilgili kod bağlamını (açık dosya, imleç çevresi, seçili sembol tanımları) ve istemi gönderir; depo bütünü gönderilmez. Bağlantı kurumun VPN tüneli içinde TLS 1.3 ile şifrelidir; Hub yalnızca kuruma tahsis edilmiş IP’lerden gelen trafiği kabul eder. - Lisans ve kota: Hub Aginies belirteci doğrular, kurumun workspace’ine bağlar, koltuk ve kota kontrolü yapar. Kullanıcı kurumda devre dışı bırakıldığında erişim aynı anda düşer. - Maskeleme ve yönlendirme: İstem, yapılandırılabilir kurallarla sırlar (anahtar, parola, bağlantı dizesi) ve kişisel veri için taranır; eşleşenler maskelenir. İstem, sözleşmede seçilen modele yönlendirilir. - Model: Model, Aginies’in Türkiye’deki veri merkezindeki GPU’larda çalışır. Dış API çağrısı yoktur; istem ve yanıt model tarafında saklanmaz ve eğitimde kullanılmaz. - Yanıt ve iz: Öneri aynı tünelden IDE’ye döner. Platform Aginies isteği kullanıcı, zaman, model sürümü, süre, token ve maliyetle kaydeder; istem ve yanıt içeriğinin saklanıp saklanmayacağı ve süresi kurumla birlikte belirlenir. - Kurumun görünürlüğü: Kurum yöneticileri kullanım ve maliyet panolarını görür; izler istenirse syslog veya API ile kurumun SIEM’ine tek yönlü aktarılır. Kill-switch kullanıcı, ekip veya kurum bazında anında keser. | Katman | Kontrol | Kanıt / denetim | | --- | --- | --- | | Kimlik | Kurumun IdP’si ile SSO (SAML / OIDC), MFA kurumda; kullanıcı ve grup bazlı yetki; anında iptal | Oturum ve yetki logları; koltuk listesi kurumun yöneticisinde | | Ağ | İki erişim modeli: VPN modelinde site-to-site tünel ve Hub dışarıdan erişilemez; internet modelinde TLS 1.3 ve kurum çıkış IP’leri allowlist’te (hız sınırı ve otomatik engelleme muafiyeti) | Tünel veya allowlist konfigürasyonu; bağlantı testleri; IP detaylı ağ şeması | | Uygulama | İmzalı eklenti, IDE Registry’den onaylı sürüm dağıtımı; sürüm sabitleme | İmza doğrulama; sürüm notları; SBOM talep üzerine | | Veri | Sır ve PII maskeleme; bağlam sınırlaması; şifreleme (aktarımda TLS, durağan AES-256); saklama süresi sözleşmeyle | Maskeleme kural seti; saklama politikası; silme kanıtı | | Model | Ülke içi GPU’da servis; dış çağrı yok; eğitim yok; sürümlü modeller, test korpusuyla doğrulama, yedek model | Model kataloğu ve sürüm geçmişi; test raporları | | Kiracı | Kurum workspace’i mantıksal olarak ayrı; istenirse dedike tenant veya dedike donanım | İzolasyon tasarımı; erişim matrisi | | İzleme | İstek bazlı denetim izi; kullanım ve maliyet panoları; SIEM aktarımı; kill-switch | Panolar; log örnekleri; aktarım entegrasyonu | | Operasyon | Olay bildirimi, değişiklik yönetimi, yedekleme ve felaket kurtarma Türkiye içinde | Olay müdahale planı; RTO/RPO; tatbikat kayıtları | | Uyum | SOC 2 Type II ve ISO 27001 uyumlu kontrol seti; KVKK; BDDK bilgi sistemleri mevzuatına uyumlu kurulum | Kontrol seti dokümanı; sızma testi özeti; denetim hakkı sözleşmede | İki erişim modeli: VPN tüneli veya internet üzerinden allowlist. Aynı istek yolu iki şekilde açılır. Kurum tam kapalı devre istiyorsa VPN tüneli; geliştiricilerin ek istemci kurmadan başlamasını istiyorsa internet üzerinden TLS ve kurum çıkış IP’lerinin allowlist’i. İkinci model bir bankanın bilgi güvenliği ekibiyle birlikte kuruldu; aşağıdaki şema ve akış tablosu o kurulumun anonimleştirilmiş hâlidir. - VPN tüneli (tam kapalı devre): Kurum ağ geçidi ile Aginies arasında site-to-site IPsec; Hub yalnızca tünelden ve kuruma tahsisli IP’lerden erişilir, internetten görünmez. - İnternet üzerinden, allowlist ile (hızlı başlangıç): Uç noktalar genel internete açık; eklenti kurumun mevcut internet çıkışından TLS 1.3 ile bağlanır. Kurumun çıkış IP’leri allowlist’e yazılır; böylece hız sınırı ve otomatik engelleme kuralları kurum trafiğini kesmez, erişim kimlik doğrulamayla verilir. [Kurum IDE’lerinden internet üzerinden TLS ile Aginies uygulama sunucusuna, oradan iç ağdaki GPU model sunucusuna giden istek yolu; istemci VPN’i yok, kurum çıkış IP’leri allowlist’te, dış model çağrısı yok.] | Akış | Kaynak | Hedef | Port / protokol | Erişim ve kontrol | | --- | --- | --- | --- | --- | | 1 | Geliştirici IDE’si · Code Aginies eklentisi · kurum çıkış IP’leri | code-api.aginies.com → Aginies genel IP’si | TCP 443 · HTTPS, TLS 1.3 | İnternet; kurum IP’leri hız sınırı ve otomatik engellemeden muaf. SSO belirteci, koltuk ve kota, maskeleme, istek bazında iz. | | 2 | Aginies uygulama sunucusu (iç ağ bacağı) | GPU model sunucusu (iç ağ) | İç ağ · model API | Yalnızca iç ağ; yalnızca uygulama sunucusundan kabul edilir, veri merkezinden çıkmaz. | | 3 | GPU model sunucusu | İnternet | — | Yok: sunucunun genel IP’si ve internete çıkışı yok; istem ve yanıt saklanmaz, eğitimde kullanılmaz. | | 4 | Aginies operasyon ekibi | Aginies uygulama sunucusu · yönetim | TCP 22 · SSH | Yalnızca Aginies operasyon VPN’i; internete ve kuruma kapalı. Kurum tarafında SSH veya VPN gerekmez. | | 5 | Aginies uygulama sunucusu | Dış model sağlayıcıları | — | Yok: kodlama modelleri yalnızca ülke içi GPU’da; kod bağlamı Türkiye dışına çıkmaz. | | 6 | Kurum uygulamaları (isteğe bağlı, doğrudan model erişimi) | hub-api.aginies.com → Aginies genel IP’si | TCP 443 · HTTPS, TLS 1.3 | Aynı allowlist, kimlik ve kota; Hub kataloğundaki modellere API ile erişim. | Şemadaki adresler anonimleştirilmiştir; gerçek IP’ler, alan adları ve güvenlik duvarı kuralları kurumun bilgi güvenliği ekibine teknik ekte verilir. ### Konum ve altyapı **Sistemler nerede çalışıyor? Veri Türkiye dışına çıkıyor mu?** Platformun tamamı Türkiye’deki veri merkezlerimizde çalışır; birincil ve ikincil sistemler ülke içindedir. Kod bağlamı, loglar ve yedekler Türkiye dışına çıkmaz. Bu, BDDK bilgi sistemleri mevzuatının ülke içi konumlanma şartıyla uyumludur. **Veri merkezi hangisi, hangi standartlara sahip?** Sistemler Türkiye’de konumlu, uluslararası Tier ve ulusal TS EN 50600 standartlarına sahip, finans sektörü düzenlemelerine (BDDK, TCMB) uyumlu veri merkezlerinde çalışır. Veri merkezinin adı, belgeleri ve sertifikaları keşif oturumunda ve sözleşme ekinde yazılı paylaşılır. **Kuruma özel bir ortam mı, paylaşımlı mı?** Varsayılan kurulum egemen paylaşımlı altyapıda, kuruma ait mantıksal olarak izole bir workspace’tir. Kurumun tercihine göre dedike tenant, dedike GPU havuzu veya kurumun kendi veri merkezinde on-prem kurulum da mümkündür; mimari üçünde de aynıdır. **Alt yükleniciler kimler? Bulut sağlayıcı kullanıyor musunuz?** Model servisi için bulut sağlayıcı kullanılmaz. Veri merkezi hizmeti Türkiye’de konumlu, düzenleyici kurumlarca uygun görülen sağlayıcılardan alınır; sağlayıcı sözleşmede adıyla yazılır ve bir değişiklik olursa kurum önceden bilgilendirilir. ### Ağ ve bağlantı **Bağlantı nasıl kurulacak? Hangi portlar ve adresler?** İki seçenek vardır. VPN modelinde kurumun ağ geçidi ile Aginies arasında site-to-site IPsec tünel kurulur, tünel içinde yalnızca HTTPS (443) kullanılır ve Hub yalnızca kuruma tahsisli IP’lerden gelen trafiği kabul eder. İnternet modelinde uç noktalar TLS 1.3 ile genel internete açıktır; kurumun güvenlik duvarında tek giden kural (Aginies genel IP’si, TCP 443) açılır ve kurumun çıkış IP’leri allowlist’e yazılır. Allowlist, hız sınırı ve otomatik engelleme kurallarının kurum trafiğini kesmemesini sağlar; erişim kimlik doğrulamayla verilir. Her iki modelde de IP adresleri ve parametreler teknik ekte verilir. **Hub Aginies internete açık mı?** Hayır. Hub yalnızca tünel arayüzünden erişilebilir; herkese açık bir uç noktası yoktur. DDoS veya tarama yüzeyi oluşmaz. **Şifreleme nasıl?** Aktarımda VPN tüneli içinde TLS 1.3 (uçtan uca, tünelden bağımsız); durağan veride AES-256. Anahtar yönetimi Aginies tarafında; kurumun kendi anahtarını getirmesi dedike kurulumda mümkündür. **Kurumun proxy’si veya SSL inceleme cihazı araya girebilir mi?** Eklenti kurumsal proxy ayarlarına uyar. Kurumun kendi sertifikasıyla TLS incelemesi yapması teknik olarak mümkündür; bu durumda inceleme cihazının kod bağlamını gördüğü kurumun kendi politikasına bırakılır. Önerimiz tünel içinde uçtan uca TLS’in korunmasıdır. ### Kimlik ve erişim **Kullanıcılar nasıl doğrulanıyor? Lisans anahtarı paylaşılırsa ne olur?** Kimlik kurumun kimlik sağlayıcısındadır: SAML veya OIDC ile SSO, MFA kurumun politikasıyla. Lisans anahtarı kurum düzeyindedir ve tek başına erişim vermez; her istek kullanıcıya ait kısa ömürlü bir belirteç taşır. Kullanıcı kurumda kapatıldığında Aginies erişimi de aynı anda düşer. **Kim hangi modeli, ne kadar kullanabilir?** Yetki grup bazlıdır: hangi ekip hangi modele, hangi kotayla, hangi özelliklerle (tamamlama, sohbet, ajan) erişir; kurumun yöneticisi Platform Aginies panosundan yönetir. Koltuk sayısı ve kotalar sözleşmeyle belirlenir. **Aginies personeli kurum verisine erişebilir mi?** Üretim erişimi rol bazlı, isimli ve loglanmıştır; istem ve yanıt içeriğine erişim yalnızca kurumun yazılı talebiyle ve destek amaçlı açılır, kaydı kurumla paylaşılır. Varsayılan saklama politikasında içerik zaten tutulmaz. **Yönetici işlemleri iz bırakıyor mu?** Evet. Yönetim işlemleri de aynı denetim izine yazılır: kim, ne zaman, neyi değiştirdi. Politika değişiklikleri sürümlüdür. ### Veri akışı ve saklama **Eklenti tam olarak ne gönderiyor?** Yalnızca isteğin gerektirdiği bağlam: açık dosyanın ilgili bölümü, imleç çevresi, seçili sembol tanımları ve geliştiricinin istemi. Depo bütünü, derleme çıktıları ve ortam dosyaları gönderilmez. Bağlam sınırı ve dışlama listesi (.env, sertifika, anahtar dosyaları) kurumla birlikte yapılandırılır ve eklentiden görülebilir. **Sırlar (parola, API anahtarı) modele gider mi?** Hub’daki maskeleme katmanı istemi anahtar, parola, bağlantı dizesi ve kişisel veri kalıpları için tarar; eşleşenler modele ulaşmadan maskelenir. Kural seti kurumun kalıplarıyla genişletilebilir; maskeleme olayları izde görünür. **İstem ve yanıtlar saklanıyor mu? Ne kadar?** Meta veri (kullanıcı, zaman, model sürümü, süre, token, maliyet) her zaman kaydedilir. İstem ve yanıt içeriğinin saklanması kurumun kararıdır: kapalı (varsayılan), ya da kurumun belirlediği süreyle şifreli saklama. Süre dolunca silme otomatiktir ve kanıtı üretilir. **Verilerimiz model eğitiminde kullanılıyor mu?** Hayır. Kurumun istem ve yanıtları hiçbir modelin eğitim veya ince ayar verisine girmez; bu sözleşmede açık bir madde olarak yer alır. Kuruma özel ince ayar istenirse yalnızca kurumun onayladığı veriyle, kuruma ait ayrı bir model sürümü üretilir. **Yedekler nerede, kim erişiyor?** Yedekler Türkiye içinde, şifreli ve ayrı bir erişim politikasıyla tutulur; içerik saklama kapalıysa yedeklerde de içerik bulunmaz. Yedek saklama süresi ve felaket kurtarma hedefleri sözleşme ekinde yazılıdır. **Sözleşme biterse ne olur?** Kurumun workspace’i, yapılandırması ve izleri istenen formatta dışa aktarılır; ardından tüm veri belirlenen süre içinde silinir ve silme kanıtı verilir. Kuruma özel model sürümü varsa kurum isterse teslim edilir, aksi hâlde imha edilir. ### Model ve yapay zekâ **Hangi modeller kullanılacak? Dışarıdaki bir sağlayıcıya gidiyor mu?** Hub Aginies kataloğundan, ülke içi GPU’larda servis edilen kodlama modelleri: açık kaynak tabanlı on-prem modeller (aginies-talay-2, aginies-sunkar-1) ve ince ayarlı kurumsal sürümler. Hiçbir istem dış sağlayıcıya gitmez; kurumun workspace’inde bulut model proxy’si kapalıdır ve kurum isterse politika düzeyinde kilitlenir. **Modelleri kim seçiyor, nasıl güncelleniyor?** Modeller Aginies tarafından denetlenip test korpusuyla doğrulandıktan sonra kataloğa alınır; her model sürümlüdür. Kurumun workspace’inde model değişikliği kurumun onayıyla ve önceden duyurularak yapılır; eski sürüm belirlenen süre boyunca geri dönüş için tutulur. **Model çıktısı güvenli mi? Zararlı kod üretir mi?** Model çıktısı öneridir; kodu geliştirici gözden geçirir ve kurumun mevcut CI, kod inceleme ve statik analiz süreçlerinden geçer. Aginies tarafında istem enjeksiyonu ve politika dışı çıktı için kontrol aileleri test korpusunda koşar; kuruma özel yasak kalıplar politika olarak eklenebilir. **Lisans ve fikri mülkiyet: üretilen kod kime ait?** Üretilen kodun mülkiyeti kurumdadır. Kullanılan açık kaynak modellerin lisansları ve ticari kullanım koşulları sözleşme ekinde listelenir. ### Eklenti ve istemci tarafı **Eklenti nereden kuruluyor, imzalı mı?** Eklenti imzalıdır ve kurumun kurumsal IDE Registry’si üzerinden onaylı sürüm olarak dağıtılır; herkese açık marketten kurulum gerekmez. Sürüm sabitlenebilir; güncellemeler kurumun onayıyla yayınlanır. **Eklenti başka bir yere bağlanıyor mu? Telemetri var mı?** Eklenti yalnızca kurumun yapılandırdığı Hub adresine bağlanır; başka hiçbir uç nokta yoktur. Hata raporu telemetrisi en az düzeydedir, içerik taşımaz ve kapatılabilir. **Bağlantı koparsa ne olur? Çevrimdışı çalışıyor mu?** Bağlantı yoksa eklenti sessizce devre dışı kalır; yerel kuyruk veya önbellek tutmaz, dolayısıyla cihazda kod bağlamı biriktirmez. Bağlantı gelince kaldığı yerden devam eder. **Kaynak kodu veya SBOM inceleyebilir miyiz?** Eklentinin yazılım malzeme listesi (SBOM) ve bağımlılık taraması talep üzerine paylaşılır; kaynak kod incelemesi gizlilik sözleşmesi altında düzenlenebilir. ### Uyum, denetim ve operasyon **Hangi sertifikalarınız var?** Platform SOC 2 Type II ve ISO 27001 kontrol setleriyle uyumludur; kontrol seti dokümanı iç denetim için paylaşılır. KVKK kapsamında veri işleyen sözleşmesi ve aydınlatma metinleri hazırdır. **Bankalar için: BDDK açısından bu bir destek hizmeti mi? Denetim hakkı var mı?** Kurum bir banka ise kendi sınıflandırmasına göre destek hizmeti kapsamına girebilir; bu durumda gerekli sözleşme hükümleri (denetim hakkı, alt yüklenici bildirimi, çıkış planı, bilgi güvenliği yükümlülükleri, düzenleyicinin ve kurumun denetim erişimi) sözleşmeye eklenir. Ülke içi konumlanma şartı mimaride zaten sağlanmıştır. **Sızma testi yapılıyor mu? Sonuçları görebilir miyiz?** Bağımsız sızma testi düzenli olarak yapılır; yönetici özeti paylaşılır, bulgu kapanış kanıtları sunulur. Kurum kendi testini de yapabilir; kapsam ve zaman önceden anlaşılır. **Güvenlik olayında bildirim süresi nedir?** Kurumu etkileyen bir olayda ilk bildirim 24 saat içinde yapılır; ardından kök neden ve düzeltici eylem raporu verilir. KVKK bildirim yükümlülükleri veri sorumlusu olan kurumla koordineli yürütülür. **Hizmet seviyesi ve kapasite?** Erişilebilirlik ve gecikme hedefleri sözleşmede yazılıdır; kapasite kurumun koltuk sayısına göre ayrılır, GPU havuzu ülke içinde ölçeklenir. Bakım pencereleri önceden duyurulur. **Acil durumda hizmeti anında kesebilir miyiz?** Evet. Kill-switch kurumun yöneticisindedir: kullanıcı, ekip veya kurum bazında anında keser; ayrıca kurum VPN tünelini kendi tarafından kapatarak tüm erişimi sonlandırabilir. **Değişiklik yönetimi nasıl işliyor?** Hub, eklenti ve modellerde kurumu etkileyen her değişiklik sürüm notuyla önceden duyurulur; kritik değişiklikler kurumun onayına bağlanabilir. Politika ve konfigürasyon değişiklikleri denetim izinde sürümlüdür. Aynı mimari üç kurulumda geçerlidir: egemen paylaşımlı altyapı, dedike tenant / GPU, kurumun veri merkezinde on-prem. ## Keşif oturumunda neler paylaşılır Keşif oturumu, kurumun senaryolarını ve kısıtlarını anlamak için yapılan bir çalışma toplantısıdır. Güvenlik ve uyum ekiplerinin sorularını erken cevaplamak için aşağıdaki materyaller paylaşılır. - Referans mimari ve dağıtım seçeneklerinin ayrıntılı şeması - SOC 2 Type II ve ISO 27001 uyumlu kontrol setinin iç denetime yönelik dokümantasyonu - Veri akışı ve maskeleme tasarımı: hangi veri hangi modele, hangi biçimde gider - Model kataloğu ve proxy kurulumu; anahtar yönetimi ve yedek model stratejisi - Örnek denetim raporu ve çalıştırma izi - Seçilen senaryolar için beş katmanlı değerlendirme karnesi Kurum adı ve sektöre özel gereksinimler oturumdan önce paylaşılırsa, materyaller o kısıtlara göre hazırlanır. ## Sektörel uyum çerçeveleri Her sektör sayfasındaki uyum çerçevesi ve önerilen kurulum modeli. İlk 90 gün planı ve ölçülen hedefler sektör sayfasındadır. | Sektör | Çerçeve | Önerilen kurulum | İlk ölçülen hedef | | --- | --- | --- | --- | | Bankacılık | BDDK Bilgi Sistemleri ve Elektronik Bankacılık Yönetmeliği: birincil ve ikincil sistemler Türkiye’de, dış hizmet sözleşmesinde denetim hakkı, bulut modeli bankaya özgü izin · MASAK ve KVKK: şüpheli işlem bildirimi analist kararında, kişisel veri maskeleme ve amaç sınırlaması her senaryoda · İç denetim ve BDDK incelemesi için istek bazlı denetim izi: kullanıcı, model sürümü, veri kaynağı, gerekçe | Başlangıç için Türkiye’deki veri merkezlerimizde egemen paylaşımlı kurulum (banka workspace’i izole, VPN veya allowlist ile erişim); birincil sistem sayılan senaryolar için bankanın veri merkezinde on-prem. | Çağrı merkezine düşen basit işlem: %40 azalma | | Finansal Teknolojiler | TCMB Ödeme ve Elektronik Para Kuruluşları Bilgi Sistemleri Tebliği: topluluk bulutu yalnızca uygunluk verilen sağlayıcıda; birincil sistemler yurt içinde · MASAK yükümlülükleri: işlem izleme uyarıları kanıtla analiste, bildirim kararı insanda · Kart şeması kuralları ve KVKK: ters ibraz süreleri takip edilir, kart verisi maskelenir ve amaç dışı kullanılmaz | TCMB uygunluk listesindeki sağlayıcının topluluk bulutunda veya kurumun kendi veri merkezinde; GPU kapasitesi ülke içinde, modeller Hub kataloğundan. | İşyeri talebi ilk temasta çözüm: %70 ve üzeri | | Sigorta | SEDDK düzenlemeleri ve sigorta bilgi sistemleri gereklilikleri: hasar ve poliçe kararları gerekçeli, zaman damgalı, denetime açık · KVKK özel nitelikli veri: sağlık ve hasar fotoğrafları kurum sınırında işlenir, maskeleme ve erişim kaydı zorunlu · Ödeme ve red kararı her zaman yetkili eksper veya underwriter’da; ajan dosyayı hazırlar | Türkiye içi dedike bölge veya egemen paylaşımlı kurulum; hasar fotoğrafı ve sağlık verisi için görüntü modelleri ülke içi GPU’da. | Basit hasar dosyası kapanış süresi: Haftalardan günlere | | Perakende & E-ticaret | Tüketici mevzuatı ve mesafeli satış kuralları: cayma, iade ve teslimat süreleri ajan kurallarına kodlanır · KVKK ve pazarlama izni: temas yalnızca izinli kanalda, öneriler amaç sınırlı veriyle · Kampanya ve fiyat kuralları: indirim yetkisi matrisle sınırlı, sınır dışı teklif müşteri hizmetlerine | Egemen paylaşımlı kurulum; sezon tepe yüklerinde GPU kapasitesi kullanım bazlı ölçeklenir, müşteri verisi Türkiye’de kalır. | İlk temasta çözüm oranı: +15–25 puan | | Telekom | BTK düzenlemeleri ve Elektronik Haberleşme Kanunu: abonelik, taahhüt ve numara taşıma işlemleri kayıtlı ve geri izlenebilir · KVKK ve trafik verisi: abone verisi Türkiye içinde, ses kayıtları ve transkriptler maskelenir · SIM değişimi ve kimlik işlemleri: ek doğrulama ve insan onayı; şüpheli talep dolandırıcılık ekibine | Egemen paylaşımlı kurulum veya operatörün kendi veri merkezinde dedike tenant; ses modelleri ülke içi GPU’da, SIP ile mevcut santrale bağlı. | Temsilcisiz çözülen talep payı: %60–70 | | Sağlık | Kişisel Sağlık Verileri Yönetmeliği ve KVKK özel nitelikli veri: sağlık verisi hastane sınırında, erişim rol ve amaç bazlı, her erişim kayıtlı · Sağlık Bakanlığı HBYS ve e-Nabız entegrasyon kuralları: ajan HBYS üzerinden işlem yapar, ayrı veri kopyası tutmaz · Klinik karar hekimde: ajan hazırlık, sorgu ve organizasyon yapar; şüphede hekim onay kartı | Hastane veri merkezinde on-prem; modeller kurum içinde, dış bağlantı yok. Hastane grupları için grup veri merkezinde dedike tenant. | Randevuya gelmeme oranı: %30–40 azalma | | Üretim ve Otomotiv | OT güvenliği (IEC 62443 yaklaşımı): ajanlar üretim ağına yalnızca okuma ile bağlanır, komut vermez; OT/IT ayrımı korunur · Ürün güvenliği ve garanti mevzuatı: garanti kararları kurallı, tolerans dışı dosya insana · KVKK ve tedarikçi sözleşmeleri: tedarikçi ve müşteri verisi kurum içinde, paylaşım sözleşme kapsamıyla sınırlı | Hibrit: üretim ağına yakın okuma ajanları ve historian erişimi tesiste, karar ve doküman ajanları Türkiye’deki veri merkezimizde veya kurumun veri merkezinde. | Plansız duruş süresi: %20–30 azalma | | Lojistik | Gümrük mevzuatı: beyanname ve belge hazırlığı ajanla, beyan ve imza yetkili müşavirde · GDP ve soğuk zincir kuralları: sıcaklık sapmaları dokümante edilir, karar kalite sorumlusunda · KVKK ve müşteri sözleşmeleri: sevkiyat ve alıcı verisi kurum içinde, her müşteri mesajı kayıtlı | Egemen paylaşımlı kurulum; TMS, telematik ve taşıyıcı API’leri allowlist ile bağlanır, sevkiyat verisi Türkiye’de kalır. | Gecikme tespit öncülü: Saatler önce | | Enerji | EPDK tüketici hizmetleri ve bağlantı yönetmelikleri: itiraz, düzeltme ve başvuru süreleri ajan kurallarına kodlanır · Kritik altyapı ve OT ayrımı: SCADA ve OMS’e yalnızca okuma erişimi; şebeke kararı kontrol odasında · KVKK ve sayaç verisi: tüketim verisi yalnızca abonenin kendi hesabı için, saklama süresi politikayla | Müşteri operasyonu için egemen paylaşımlı kurulum; şebekeye yakın okuma ajanları şirketin veri merkezinde on-prem, kritik altyapı ayrımı korunarak. | Temsilcisiz çözülen çağrı payı: %55–65 | | Kamu ve Hükümet | Cumhurbaşkanlığı Bilgi ve İletişim Güvenliği Rehberi: kurum veri merkezinde, air-gap dahil kurulum; açık ağırlıklı modeller içeride · İdari usul ve Kamu İhale Kanunu: her çıktı mevzuat maddesi ve kaynak belgeyle; karar yetkili memur veya komisyonda · KVKK ve e-Devlet entegrasyon kuralları: vatandaş verisi amaç sınırlı, erişim yetki bazlı, denetim izi tam | Kurumun kendi veri merkezinde on-prem; gerekiyorsa air-gap. Belediyeler ve bağlı kuruluşlar için kamu bulutunda dedike tenant. | Vatandaştan istenen tekrar belge: %70 azalma | | Savunma Sanayii | SSB ve ihracat kontrol mevzuatı: kontrole tabi liste eşlemesi ajanla, lisans ve paylaşım kararı yetkili uyum sorumlusunda · Tesis güvenlik belgesi ve gizlilik dereceleri: erişim ayrımı, işaretleme kontrolü ve her erişimin denetim günlüğü · Müttefik programlarda ITAR/EAR benzeri kısıtlar: kontrollü teknik veri yalnızca yetkili kullanıcıya, dış ağa çıkış yok | Air-gap kurulum: modeller, platform ve veri kapalı ağın içinde; güncellemeler onaylı paketlerle, uzaktan bakım yalnızca kurumun açtığı pencerede. | Taraf tarama süresi: Günlerden dakikalara | | IoT ve Bağlı Cihazlar | KVKK ve cihaz verisi: telemetri kurum içinde, kullanıcıya ait veri amaç sınırlı · Ürün güvenliği ve siber dayanıklılık gereklilikleri (ihracatta AB CRA dahil): OTA paketleri imzalı, dağıtım kayıtlı, geri alma hazır · Emniyet kritik cihazlar: yalnızca okuma; komut ve toplu işlem ürün sahibi onayında | Hibrit: cihaz bulutu ve zaman serisi verisi mevcut altyapıda, ajanlar ve küçük modeller Türkiye’deki veri merkezimizde veya üreticinin veri merkezinde. | Saha ziyareti gerektiren arıza: %35 azalma | | Yazılım Geliştirme | Kaynak kod ve fikri mülkiyet: kod kurum dışına çıkmaz; Code Aginies ve tamamlama modelleri on-prem · Güvenli yazılım geliştirme gereklilikleri (regüle sektörlerde BDDK/SEDDK beklentileri dahil): SAST, bağımlılık ve sır taraması her PR’da, kanıt sürümle saklanır · Üretim dağıtımı ve mimari kararlar insanda; ajan yalnızca atanan depo ve dallarda yazar | Code Aginies ve kodlama modelleri kurumun veri merkezinde veya kapalı devre kurulumla Türkiye’deki veri merkezimizde; SDLC ajanları Jira, Git ve CI/CD’ye servis hesaplarıyla bağlı. | İlk incelemede kabul edilen PR: %60–75 | --- # Güvenlik ve Egemenlik URL: https://www.aginies.com/tr/guvenlik Language: tr Summary: SOC 2 Type II ve ISO 27001 uyumlu platform; KVKK ve GDPR uyumu; on-prem, air-gap, hibrit ve dedike bölge dağıtımı; GPU-as-a-Service ile ülke içi elastik kapasite; her ajan aksiyonu için tam denetim izi. Güvenlik, uyum ve egemenlik # Veriniz nerede olmalıysa orada. AI, kurumun diğer kritik sistemleriyle aynı standartlara tabi olmalı. Aginies; finans, sigorta ve sağlık gibi regüle sektörlerin gereksinimlerine göre kurgulandı. Güvenlik ve denetlenebilirlik sonradan eklenmez; mimarinin parçasıdır. Platform, modeller ve veri kurumun kendi veri merkezinde. Dış bağlantı yok; Aginies model aileleri ve açık sürümler içeride servis edilir. [AGINIES ORKESTRASYON ÇEKİRDEĞİ] Kavramsal görünüm. Bileşen adları ve sorumluluklar gösterilir; uygulama ayrıntıları keşif oturumunda paylaşılır. Kapalı devre kurulum ## Kod asistanı senaryosu: IDE’den modele Kapalı devre çalışmak isteyen kurumların bilgi güvenliği ekipleri için. Kurumun geliştiricileri IDE’lerindeki Code Aginies eklentisiyle, kurumun VPN’i üzerinden, Türkiye’de konumlu ve BDDK mevzuatına uyumlu veri merkezlerindeki altyapımızdaki Hub Aginies’e bağlanır; kodlama modelleri orada, ülke içi GPU’larda çalışır. Kod bağlamı Türkiye dışına çıkmaz, hiçbir dış model çağrılmaz, kurum verisiyle model eğitilmez, her istek kullanıcı bazında iz bırakır. TürkiyeBirincil ve ikincil sistemler ülke içinde; veri yerleşimi garantili 0Dış model çağrısı; kod bağlamı egemen sınırın dışına çıkmaz 0Kurum verisiyle model eğitimi; istemler öğrenme verisi değildir %100İstek bazında denetim izi: kullanıcı, model sürümü, süre, maliyet [Kurum IDE’lerinden VPN tüneliyle Aginies egemen sınırındaki Hub Aginies’e, oradan ülke içi GPU model sunucularına giden istek yolu; dış model çağrısı ve eğitim yok.] İstek yolu: (1) geliştirici kurum SSO’suyla oturum açar; (2) eklenti kod bağlamını kurumun VPN geçidinden tünelle Hub Aginies’e gönderir; (3) lisans ve kota doğrulanır; (4) maskelenmiş istem ülke içi GPU’daki modele gider; (5) öneri aynı yoldan IDE’ye döner; (6) her istek Platform Aginies’te iz bırakır; (7) izler isteğe bağlı olarak kurumun SIEM’ine aktarılır. ### Bir tamamlama isteğinin yedi adımı - 1 KimlikGeliştirici IDE’de kurumun kimlik sağlayıcısıyla (SAML / OIDC, MFA) oturum açar. Eklenti kullanıcıya ait kısa ömürlü bir belirteç taşır; lisans anahtarı kurum düzeyinde tutulur ve eklentiden ayrı durur. - 2 İstekEklenti yalnızca ilgili kod bağlamını (açık dosya, imleç çevresi, seçili sembol tanımları) ve istemi gönderir; depo bütünü gönderilmez. Bağlantı kurumun VPN tüneli içinde TLS 1.3 ile şifrelidir; Hub yalnızca kuruma tahsis edilmiş IP’lerden gelen trafiği kabul eder. - 3 Lisans ve kotaHub Aginies belirteci doğrular, kurumun workspace’ine bağlar, koltuk ve kota kontrolü yapar. Kullanıcı kurumda devre dışı bırakıldığında erişim aynı anda düşer. - 4 Maskeleme ve yönlendirmeİstem, yapılandırılabilir kurallarla sırlar (anahtar, parola, bağlantı dizesi) ve kişisel veri için taranır; eşleşenler maskelenir. İstem, sözleşmede seçilen modele yönlendirilir. - 5 ModelModel, Aginies’in Türkiye’deki veri merkezindeki GPU’larda çalışır. Dış API çağrısı yoktur; istem ve yanıt model tarafında saklanmaz ve eğitimde kullanılmaz. - 6 Yanıt ve izÖneri aynı tünelden IDE’ye döner. Platform Aginies isteği kullanıcı, zaman, model sürümü, süre, token ve maliyetle kaydeder; istem ve yanıt içeriğinin saklanıp saklanmayacağı ve süresi kurumla birlikte belirlenir. - 7 Kurumun görünürlüğüKurum yöneticileri kullanım ve maliyet panolarını görür; izler istenirse syslog veya API ile kurumun SIEM’ine tek yönlü aktarılır. Kill-switch kullanıcı, ekip veya kurum bazında anında keser. ### İki erişim modeli: VPN tüneli veya internet üzerinden allowlist Aynı istek yolu iki şekilde açılır. Kurum tam kapalı devre istiyorsa VPN tüneli; geliştiricilerin ek istemci kurmadan başlamasını istiyorsa internet üzerinden TLS ve kurum çıkış IP’lerinin allowlist’i. İkinci model bir bankanın bilgi güvenliği ekibiyle birlikte kuruldu; aşağıdaki şema ve akış tablosu o kurulumun anonimleştirilmiş hâlidir. tam kapalı devreVPN tüneli Kurum ağ geçidi ile Aginies arasında site-to-site IPsec; Hub yalnızca tünelden ve kuruma tahsisli IP’lerden erişilir, internetten görünmez. - Kurum tarafında VPN geçidi yapılandırması - Hub dışarıdan erişilemez - Tünel parametreleri teknik ekte hızlı başlangıçİnternet üzerinden, allowlist ile Uç noktalar genel internete açık; eklenti kurumun mevcut internet çıkışından TLS 1.3 ile bağlanır. Kurumun çıkış IP’leri allowlist’e yazılır; böylece hız sınırı ve otomatik engelleme kuralları kurum trafiğini kesmez, erişim kimlik doğrulamayla verilir. - Kurum tarafında yalnızca giden 443 kuralı ve çıkış IP listesi - VPN istemcisi veya sunucu kurulumu yok - Kimlik, kota, maskeleme ve iz aynı [Kurum IDE’lerinden internet üzerinden TLS ile Aginies uygulama sunucusuna, oradan iç ağdaki GPU model sunucusuna giden istek yolu; istemci VPN’i yok, kurum çıkış IP’leri allowlist’te, dış model çağrısı yok.] | Akış | Kaynak | Hedef | Port / protokol | Erişim ve kontrol | | --- | --- | --- | --- | --- | | 1 | Geliştirici IDE’si · Code Aginies eklentisi · kurum çıkış IP’leri | code-api.aginies.com → Aginies genel IP’si | TCP 443 · HTTPS, TLS 1.3 | İnternet; kurum IP’leri hız sınırı ve otomatik engellemeden muaf. SSO belirteci, koltuk ve kota, maskeleme, istek bazında iz. | | 2 | Aginies uygulama sunucusu (iç ağ bacağı) | GPU model sunucusu (iç ağ) | İç ağ · model API | Yalnızca iç ağ; yalnızca uygulama sunucusundan kabul edilir, veri merkezinden çıkmaz. | | 3 | GPU model sunucusu | İnternet | — | Yok: sunucunun genel IP’si ve internete çıkışı yok; istem ve yanıt saklanmaz, eğitimde kullanılmaz. | | 4 | Aginies operasyon ekibi | Aginies uygulama sunucusu · yönetim | TCP 22 · SSH | Yalnızca Aginies operasyon VPN’i; internete ve kuruma kapalı. Kurum tarafında SSH veya VPN gerekmez. | | 5 | Aginies uygulama sunucusu | Dış model sağlayıcıları | — | Yok: kodlama modelleri yalnızca ülke içi GPU’da; kod bağlamı Türkiye dışına çıkmaz. | | 6 | Kurum uygulamaları (isteğe bağlı, doğrudan model erişimi) | hub-api.aginies.com → Aginies genel IP’si | TCP 443 · HTTPS, TLS 1.3 | Aynı allowlist, kimlik ve kota; Hub kataloğundaki modellere API ile erişim. | Şemadaki adresler anonimleştirilmiştir; gerçek IP’ler, alan adları ve güvenlik duvarı kuralları kurumun bilgi güvenliği ekibine teknik ekte verilir. ### Kurumdan Aginies’e gider - Açık dosyanın ilgili bölümü, imleç çevresi ve seçili bağlam (kod bağlamı) - Geliştiricinin istemi (soru, komut) - Kullanıcı kimliği (SSO belirteci), IDE ve eklenti sürümü - Hata raporu için en az düzeyde teknik telemetri (isteğe bağlı, kapatılabilir) ### Hiçbir yere gitmez - Depo bütünü, derleme çıktıları, ortam dosyaları, gizli anahtarlar (maskelenir) - Aginies sınırından dış internete veya bulut model sağlayıcılarına hiçbir istem - Türkiye dışına log, yedek veya kopya - Model eğitimine veya başka bir müşterinin workspace’ine hiçbir kurum verisi ### Bilgi güvenliği ekibinin ilk soruları **Sistemler nerede çalışıyor? Veri Türkiye dışına çıkıyor mu?** Platformun tamamı Türkiye’deki veri merkezlerimizde çalışır; birincil ve ikincil sistemler ülke içindedir. Kod bağlamı, loglar ve yedekler Türkiye dışına çıkmaz. Bu, BDDK bilgi sistemleri mevzuatının ülke içi konumlanma şartıyla uyumludur. **Hub Aginies internete açık mı?** Hayır. Hub yalnızca tünel arayüzünden erişilebilir; herkese açık bir uç noktası yoktur. DDoS veya tarama yüzeyi oluşmaz. **Kullanıcılar nasıl doğrulanıyor? Lisans anahtarı paylaşılırsa ne olur?** Kimlik kurumun kimlik sağlayıcısındadır: SAML veya OIDC ile SSO, MFA kurumun politikasıyla. Lisans anahtarı kurum düzeyindedir ve tek başına erişim vermez; her istek kullanıcıya ait kısa ömürlü bir belirteç taşır. Kullanıcı kurumda kapatıldığında Aginies erişimi de aynı anda düşer. **Aginies personeli kurum verisine erişebilir mi?** Üretim erişimi rol bazlı, isimli ve loglanmıştır; istem ve yanıt içeriğine erişim yalnızca kurumun yazılı talebiyle ve destek amaçlı açılır, kaydı kurumla paylaşılır. Varsayılan saklama politikasında içerik zaten tutulmaz. **Eklenti tam olarak ne gönderiyor?** Yalnızca isteğin gerektirdiği bağlam: açık dosyanın ilgili bölümü, imleç çevresi, seçili sembol tanımları ve geliştiricinin istemi. Depo bütünü, derleme çıktıları ve ortam dosyaları gönderilmez. Bağlam sınırı ve dışlama listesi (.env, sertifika, anahtar dosyaları) kurumla birlikte yapılandırılır ve eklentiden görülebilir. **Sırlar (parola, API anahtarı) modele gider mi?** Hub’daki maskeleme katmanı istemi anahtar, parola, bağlantı dizesi ve kişisel veri kalıpları için tarar; eşleşenler modele ulaşmadan maskelenir. Kural seti kurumun kalıplarıyla genişletilebilir; maskeleme olayları izde görünür. **İstem ve yanıtlar saklanıyor mu? Ne kadar?** Meta veri (kullanıcı, zaman, model sürümü, süre, token, maliyet) her zaman kaydedilir. İstem ve yanıt içeriğinin saklanması kurumun kararıdır: kapalı (varsayılan), ya da kurumun belirlediği süreyle şifreli saklama. Süre dolunca silme otomatiktir ve kanıtı üretilir. **Verilerimiz model eğitiminde kullanılıyor mu?** Hayır. Kurumun istem ve yanıtları hiçbir modelin eğitim veya ince ayar verisine girmez; bu sözleşmede açık bir madde olarak yer alır. Kuruma özel ince ayar istenirse yalnızca kurumun onayladığı veriyle, kuruma ait ayrı bir model sürümü üretilir. **Hangi modeller kullanılacak? Dışarıdaki bir sağlayıcıya gidiyor mu?** Hub Aginies kataloğundan, ülke içi GPU’larda servis edilen kodlama modelleri: açık kaynak tabanlı on-prem modeller (aginies-talay-2, aginies-sunkar-1) ve ince ayarlı kurumsal sürümler. Hiçbir istem dış sağlayıcıya gitmez; kurumun workspace’inde bulut model proxy’si kapalıdır ve kurum isterse politika düzeyinde kilitlenir. **Güvenlik olayında bildirim süresi nedir?** Kurumu etkileyen bir olayda ilk bildirim 24 saat içinde yapılır; ardından kök neden ve düzeltici eylem raporu verilir. KVKK bildirim yükümlülükleri veri sorumlusu olan kurumla koordineli yürütülür. **Acil durumda hizmeti anında kesebilir miyiz?** Evet. Kill-switch kurumun yöneticisindedir: kullanıcı, ekip veya kurum bazında anında keser; ayrıca kurum VPN tünelini kendi tarafından kapatarak tüm erişimi sonlandırabilir. Tam soru-cevap listesi ve kontrol matrisi dokümanlarda. ### Aynı mimari, üç kurulum önerilen başlangıçEgemen paylaşımlı Türkiye’deki veri merkezlerimiz, kuruma izole workspace, VPN ile erişim. - Günler içinde başlar - Model güncellemeleri ve kapasite Aginies’te - Kuruma sabit IP, izole tenant, SIEM aktarımı daha yüksek izolasyonDedike tenant / GPU Aynı veri merkezinde kuruma ayrılmış tenant ve GPU havuzu; istenirse kurumun anahtarıyla şifreleme. - Kaynak paylaşımı yok - Kuruma özel model sürümleri - Kapasite taahhüdü sözleşmeyle tam kontrolKurum veri merkezinde on-prem Hub, model sunucuları ve Platform kurumun kendi veri merkezinde; uzaktan bakım yalnızca kurumun açtığı pencerede. - Hiçbir veri kurum sınırından çıkmaz - Kurumun GPU yatırımı gerekir - Kurulum ve güncelleme kurumun değişiklik takvimiyle ## Dağıtım seçenekleri ### On-premise / air-gap Hassas veriler kurum dışına çıkmadan, kendi veri merkezinizde. Modeller içeride servis edilir; dış bağlantı yoktur. ### Hibrit Hassas iş yükleri içeride, esnek kapasite dışarıda. Dış modeller yalnızca maskelenmiş veriyle ve güvenli geçit üzerinden. ### Özel bulut / dedike bölge Kurum için ayrılmış tek kiracılı bölge; veri yerleşimi garantili. ### GPU-as-a-Service Sermaye yatırımı yapmadan, kullanım bazlı ölçeklenen, ülke sınırları içinde elastik GPU kapasitesi. ## Uyum matrisi | SOC 2 Type II | Uyumlu platform; kontrol seti iç denetim için belgelidir | | ISO 27001 | Uyumlu bilgi güvenliği yönetimi | | KVKK & GDPR | Kişisel veri haritası, veri yerleşimi garantisi, silme ve erişim talepleri | | BDDK bilgi sistemleri | Bankacılık için birincil ve ikincil sistemler Türkiye’de; denetime hazır kurulum | | SSO / SAML & RBAC | İnce taneli roller, workspace izolasyonu, workflow bazlı erişim politikaları | | PII maskeleme | Veri girişinde hassas veri maskeleme, yapılandırılabilir redaksiyon | | Denetim izi | Her ajan aksiyonu ve model çağrısı için girdi, çıktı, model sürümü, kural seti, kullanıcı | | Kill-switch | Kritik kararlarda acil kapatma ve otomasyon sınırları; zorunlu insan onayı | ## Model bağımsız, ince ayarlı Hub Aginies’te Qwen, GLM ve Kimi tabanlı ince ayarlı Aginies aileleri (tulpar, talay, sunkar, konrul), aynı tabanların güncel açık sürümleri ve Türkçe ses modelleri aynı katalogda; hepsi ülke içi GPU’da. Kurumun kendi anahtarıyla bulut modelleri proxy üzerinden eklenebilir. Vendor lock-in yok. Model değişimi hattı yeniden yazmayı gerektirmez. Her senaryo için doğruluk, gecikme ve maliyet dengesine göre model seçilir; yedek model stratejisi sağlayıcı kesintisinde operasyonu sürdürür. aginies-tulpar-4 · on-premaginies-talay-2 · on-premaginies-sunkar-1 · on-premaginies-konrul-1 · on-premqwen · güncel sürümglm · güncel sürümkimi · güncel sürümaginies-voice-tr-v3 · on-premaginies-tts-tr-v2 · on-prem Hub Aginies model kataloğu: Qwen, GLM ve Kimi tabanlı ince ayarlı Aginies aileleri (tulpar, talay, sunkar, konrul), aynı tabanların güncel açık sürümleri ve Türkçe ses modelleri; hepsi ülke içi GPU’da. Kurumun kendi anahtarıyla bulut modelleri proxy üzerinden eklenebilir. Sektörel uyum çerçeveleri ## Her sektörde hangi çerçeveye, hangi kurulumla cevap verilir. Sektör sayfalarındaki uyum çerçevesi ve önerilen kurulum, tek tabloda. Ayrıntı ve ilk 90 gün planı sektör sayfasında. | Sektör | Çerçeve | Önerilen kurulum | | --- | --- | --- | | Bankacılık → | BDDK Bilgi Sistemleri ve Elektronik Bankacılık Yönetmeliği: birincil ve ikincil sistemler Türkiye’de, dış hizmet sözleşmesinde denetim hakkı, bulut modeli bankaya özgü izin | Başlangıç için Türkiye’deki veri merkezlerimizde egemen paylaşımlı kurulum (banka workspace’i izole, VPN veya allowlist ile erişim); birincil sistem sayılan senaryolar için bankanın veri merkezinde on-prem. | | Finansal Teknolojiler → | TCMB Ödeme ve Elektronik Para Kuruluşları Bilgi Sistemleri Tebliği: topluluk bulutu yalnızca uygunluk verilen sağlayıcıda; birincil sistemler yurt içinde | TCMB uygunluk listesindeki sağlayıcının topluluk bulutunda veya kurumun kendi veri merkezinde; GPU kapasitesi ülke içinde, modeller Hub kataloğundan. | | Sigorta → | SEDDK düzenlemeleri ve sigorta bilgi sistemleri gereklilikleri: hasar ve poliçe kararları gerekçeli, zaman damgalı, denetime açık | Türkiye içi dedike bölge veya egemen paylaşımlı kurulum; hasar fotoğrafı ve sağlık verisi için görüntü modelleri ülke içi GPU’da. | | Perakende & E-ticaret → | Tüketici mevzuatı ve mesafeli satış kuralları: cayma, iade ve teslimat süreleri ajan kurallarına kodlanır | Egemen paylaşımlı kurulum; sezon tepe yüklerinde GPU kapasitesi kullanım bazlı ölçeklenir, müşteri verisi Türkiye’de kalır. | | Telekom → | BTK düzenlemeleri ve Elektronik Haberleşme Kanunu: abonelik, taahhüt ve numara taşıma işlemleri kayıtlı ve geri izlenebilir | Egemen paylaşımlı kurulum veya operatörün kendi veri merkezinde dedike tenant; ses modelleri ülke içi GPU’da, SIP ile mevcut santrale bağlı. | | Sağlık → | Kişisel Sağlık Verileri Yönetmeliği ve KVKK özel nitelikli veri: sağlık verisi hastane sınırında, erişim rol ve amaç bazlı, her erişim kayıtlı | Hastane veri merkezinde on-prem; modeller kurum içinde, dış bağlantı yok. Hastane grupları için grup veri merkezinde dedike tenant. | | Üretim ve Otomotiv → | OT güvenliği (IEC 62443 yaklaşımı): ajanlar üretim ağına yalnızca okuma ile bağlanır, komut vermez; OT/IT ayrımı korunur | Hibrit: üretim ağına yakın okuma ajanları ve historian erişimi tesiste, karar ve doküman ajanları Türkiye’deki veri merkezimizde veya kurumun veri merkezinde. | | Lojistik → | Gümrük mevzuatı: beyanname ve belge hazırlığı ajanla, beyan ve imza yetkili müşavirde | Egemen paylaşımlı kurulum; TMS, telematik ve taşıyıcı API’leri allowlist ile bağlanır, sevkiyat verisi Türkiye’de kalır. | | Enerji → | EPDK tüketici hizmetleri ve bağlantı yönetmelikleri: itiraz, düzeltme ve başvuru süreleri ajan kurallarına kodlanır | Müşteri operasyonu için egemen paylaşımlı kurulum; şebekeye yakın okuma ajanları şirketin veri merkezinde on-prem, kritik altyapı ayrımı korunarak. | | Kamu ve Hükümet → | Cumhurbaşkanlığı Bilgi ve İletişim Güvenliği Rehberi: kurum veri merkezinde, air-gap dahil kurulum; açık ağırlıklı modeller içeride | Kurumun kendi veri merkezinde on-prem; gerekiyorsa air-gap. Belediyeler ve bağlı kuruluşlar için kamu bulutunda dedike tenant. | | Savunma Sanayii → | SSB ve ihracat kontrol mevzuatı: kontrole tabi liste eşlemesi ajanla, lisans ve paylaşım kararı yetkili uyum sorumlusunda | Air-gap kurulum: modeller, platform ve veri kapalı ağın içinde; güncellemeler onaylı paketlerle, uzaktan bakım yalnızca kurumun açtığı pencerede. | | IoT ve Bağlı Cihazlar → | KVKK ve cihaz verisi: telemetri kurum içinde, kullanıcıya ait veri amaç sınırlı | Hibrit: cihaz bulutu ve zaman serisi verisi mevcut altyapıda, ajanlar ve küçük modeller Türkiye’deki veri merkezimizde veya üreticinin veri merkezinde. | | Yazılım Geliştirme → | Kaynak kod ve fikri mülkiyet: kod kurum dışına çıkmaz; Code Aginies ve tamamlama modelleri on-prem | Code Aginies ve kodlama modelleri kurumun veri merkezinde veya kapalı devre kurulumla Türkiye’deki veri merkezimizde; SDLC ajanları Jira, Git ve CI/CD’ye servis hesaplarıyla bağlı. | ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # İletişim URL: https://www.aginies.com/tr/iletisim Language: tr Summary: 30 dakikalık keşif oturumu: öncelikli iş problemlerinizi ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. Sonraki adım # Aginies’i kendi akışlarınızda görün. Ekibimizle 30 dakikalık bir oturum planlayın. Kullanım senaryonuzu birlikte ele alır, platformu gösterir, üretime giden yolu çizeriz. Oturumda ne olur - 1Önceliklendirdiğiniz 2–3 iş problemini birlikte ele alırız. - 2Aginies üzerinde benzer bir senaryonun canlı demosunu görürsünüz. - 3Sizin için Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkarırız. Getirmeniz yeterli olanlar - Bir iş problemi ve bugünkü hacmi (aylık işlem, çağrı, belge sayısı) - Aksiyon alınacak sistemlerin adları; API olup olmadığını bilmeniz gerekmez - Hangi kararın insanda kalması gerektiğine dair ilk fikriniz Oturumdan sonra alacaklarınız - Senaryonuza en yakın kütüphane senaryosu ve çözüm paketi - Değer ve hazırlık katmanı için doldurulmuş karne taslağı - İlk 90 gün için pilot kapsamı ve kurulum modeli önerisi E-posta dahi@aginies.com Konum İstanbul, Türkiye · Global müşterilere hizmet Sektörünüz için başlangıç noktaları Keşif oturumundan önce sektörünüzün ilk 90 gün planına göz atabilirsiniz. - Bankacılıkilk senaryo: Diyalog bankacılığı: sesli ve yazılı işlem asistanı - Finansal Teknolojilerilk senaryo: Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit - Sigortailk senaryo: Uçtan uca kasko hasarı: ihbardan ödemeye - Perakende & E-ticaretilk senaryo: Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin - Telekomilk senaryo: Abone hizmet asistanı: fatura, paket, arıza ve taahhüt - Sağlıkilk senaryo: Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuç - Üretim ve Otomotivilk senaryo: Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna - Lojistikilk senaryo: Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA - Enerjiilk senaryo: Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda - Kamu ve Hükümetilk senaryo: Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi - Savunma Sanayiiilk senaryo: Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolü - IoT ve Bağlı Cihazlarilk senaryo: Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyen - Yazılım Geliştirmeilk senaryo: Özellik teslimat hattı --- # Karşılaştırma URL: https://www.aginies.com/tr/karsilastirma Language: tr Summary: Aginies’i iş akışı otomasyon araçları, ajan çerçeveleri, chatbot platformları, hyperscaler ajan stüdyoları, AI kodlama ajanları ve altyapı / model servisleriyle kategori düzeyinde karşılaştırın: orkestrasyon, modeller, ses, bilgi, yönetişim, dağıtım ve işletim modeli. Karşılaştırma # Aginies hangi kategoriye giriyor? Aginies, kurumun tüm AI senaryolarının koştuğu ve egemen olarak kurulabilen yürütme katmanıdır; aşağıdaki altı araç kategorisinin hiçbiriyle birebir örtüşmez. Karşılaştırma kategori düzeyinde ve özellik bazındadır. ## İş akışı otomasyon araçları Tetikleyici-aksiyon mantığıyla uygulamaları birbirine bağlayan, geniş entegrasyon kataloglu araçlar. Neyi kastediyoruz Tetikleyici-aksiyon kurgusuyla uygulamaları birbirine bağlayan, çoğunlukla SaaS olarak sunulan düşük kodlu otomasyon araçları ve entegrasyon platformları (iPaaS); yapay zekâ bunlara bir adım olarak eklenir. Ne zaman uygun Belirli, kural tabanlı otomasyonlar; küçük ekipler; hızlı prototip. Üretimde karşılaşılan boşluk AI bir adım olarak eklenir; supervisor mimarisi, insan onayı, denetim izi ve on-prem egemenlik ürünün çevresinde kalır. 4 var · 12 kısmen · 29 yok ## Ajan geliştirme çerçeveleri Geliştiricilerin kod yazarak ajan grafikleri kurduğu açık kaynak kütüphaneler. Neyi kastediyoruz Geliştiricinin ajan grafiğini kodla kurduğu açık kaynak Python veya TypeScript kütüphaneleri: çok ajanlı orkestrasyon, araç çağrısı ve durum yönetimi çerçeveleri. Ne zaman uygun Güçlü mühendislik ekibi olan, kendi platformunu inşa etmek isteyen kurumlar. Üretimde karşılaşılan boşluk Gözlemlenebilirlik, maliyet atfı, RBAC, ses kanalı ve işletim modeli sıfırdan yazılır; sorumluluk tamamen kurumda kalır. 7 var · 12 kısmen · 26 yok ## Chatbot ve sanal asistan platformları Müşteri hizmetleri diyaloglarını tasarlamak için görsel konuşma tasarımcıları. Neyi kastediyoruz Müşteri hizmetleri diyalogları için görsel konuşma tasarımcısı, niyet tanıma ve kanal entegrasyonlarıyla gelen chatbot ve sanal asistan platformları. Ne zaman uygun Bilgi veren, yönlendiren, sınırlı işlem yapan müşteri asistanları. Üretimde karşılaşılan boşluk Diyalog dışı arka plan süreçleri (doküman, uzlaştırma, SDLC) ve çekirdek sistemlerde yazma işlemleri kapsam dışıdır. 5 var · 20 kısmen · 20 yok ## Hyperscaler ajan stüdyoları Büyük bulut ve CRM sağlayıcılarının kendi ekosistemlerine bağlı ajan araçları. Neyi kastediyoruz Büyük bulut, ofis yazılımı ve CRM sağlayıcılarının kendi ürünlerine gömülü ajan stüdyoları; o sağlayıcının modelleri ve bulutuyla çalışır. Ne zaman uygun Tek bir ekosisteme standartlaşmış ve verisini o bulutta tutabilen kurumlar. Üretimde karşılaşılan boşluk Model ve bulut bağımlılığı; on-prem / air-gap ve ülke içi GPU kapasitesi çoğunlukla sunulmaz; Türkçe ses ve yerel regülasyon uyumu ek proje ister. 8 var · 27 kısmen · 10 yok ## AI kodlama ajanları Terminal ve IDE içinde kod tabanını okuyan, dosya düzenleyen ve komut çalıştıran geliştirici ajanları; tek sağlayıcılı bulut araçlarından model bağımsız açık kaynak araçlara uzanır. Neyi kastediyoruz Terminal veya IDE içinde çalışan geliştirici ajanları: tek bir sağlayıcının bulut modeline bağlı ticari kodlama asistanları ve herhangi bir model uç noktasına bağlanabilen açık kaynak araçlar. Ne zaman uygun Bireysel geliştirici verimliliği; kodunu bulut modele gönderebilen ekipler ya da kendi model uç noktasını işletmeyi göze alan mühendislik ekipleri. Üretimde karşılaşılan boşluk Geliştirici aracıdır: iş süreçleri için supervisor mimarisi, ses ve diyalog kanalları, birim bazında maliyet atfı yoktur; yönetişim geliştirici başına izin sorusuyla sınırlıdır, kurum düzeyinde denetim izi ve onay akışı bulunmaz; yerel model için GPU altyapısını kurum kendisi kurar. Code Aginies aynı IDE yüzeyini egemen platformun içinde, ülke içi GPU, kurum lisansı ve maskelemeyle sunar. 2 var · 17 kısmen · 26 yok ## Altyapı ve model servisleri Modeli her donanımda servis eden, GPU ve Kubernetes kapasitesini yöneten ve token bazlı model erişimi veren altyapı katmanları; ülke içi bulut hizmeti olarak ya da kurumun kendi veri merkezinde çalışan yazılım olarak sunulur. Neyi kastediyoruz Kurumsal Linux ve Kubernetes sağlayıcılarının yapay zekâ platformları ile ülke içi bulut sağlayıcılarının GPU ve model-as-a-service hizmetleri; modeli servis ederler, ajanı kurum yazar. Ne zaman uygun Yalnızca model erişimi ve GPU kapasitesi isteyen, ajanı kendi mühendisleriyle yazacak kurumlar; regüle sektörde ortak altyapı tercih edenler. Üretimde karşılaşılan boşluk Modeli çalıştırır, işi yapan ajanı getirmez: iş süreci kütüphanesi, Türkçe ses ve diyalog kanalları, kod asistanı, iş bazında maliyet atfı ve sonuç sorumluluğu kurumda kalır. Aginies bu katmanın üstünde çalışır; altyapı kararı Aginies kararını değiştirmez. 11 var · 19 kısmen · 15 yok Konumlandırma ## Kurulum modeli ve yönetişim ekseninde nerede duruyoruz Yatay eksen ajanın nasıl kurulduğunu (kod → görsel tasarımcı → diyalogla üretim), dikey eksen yönetişim ve egemenliği (denetim izi, insan onayı, on-prem) gösterir. Aginies her ikisinde de en uçta: ajanlar diyalogla üretilir, kurum kendi veri merkezinde işletir. [Konumlandırma grafiği] ## Özellik karşılaştırması Kategori düzeyinde, Ekim 2026 itibarıyla. Ürünler sürekli değişir; belirli bir ürünle birebir karşılaştırma için keşif oturumunda birlikte bakalım. | Özellik | Aginies | İş akışı otomasyon araçları | Ajan geliştirme çerçeveleri | Chatbot ve sanal asistan platformları | Hyperscaler ajan stüdyoları | AI kodlama ajanları | Altyapı ve model servisleri | | --- | --- | --- | --- | --- | --- | --- | --- | | Ajan iş gücü | | Rol bazında kiralık ajan (geliştirici, analist, muhasebeci, testçi)Dış kaynak insan yerine dış kaynak ajan; ajan sizin araçlarınızda çalışır | Var | Yok | Yok | Yok | Kısmen | Yok | Yok | | İş kaydı: görev, çıktı, saat karşılığı, kalite, onay, maliyet | Var | Yok | Yok | Yok | Yok | Yok | Yok | | Gölge dönem → örneklemeli inceleme; adı belli ekip lideri | Var | Yok | Yok | Yok | Yok | Yok | Yok | | Orkestrasyon | | Supervisor → alt ajan mimarisi | Var | Yok | Var | Kısmen | Kısmen | Kısmen | Kısmen | | Görsel ajan tasarımcısı (kod yazmadan) | Var | Var | Yok | Var | Var | Yok | Kısmen | | Durumlu, çok adımlı yürütme ve paralel dallar | Var | Kısmen | Var | Kısmen | Kısmen | Kısmen | Kısmen | | Ajanlar arası iletişim (A2A) | Var | Yok | Kısmen | Yok | Kısmen | Yok | Kısmen | | Otonom Ajan: diyalogla ajan üretimi, test korpusu ve dağıtım | Var | Yok | Yok | Yok | Kısmen | Yok | Yok | | Modeller | | Model bağımsız geçit: kapalı ve açık kaynak modellerModelHub | Var | Kısmen | Var | Kısmen | Yok | Kısmen | Var | | Açık kaynak modellerin on-prem servisi | Var | Yok | Kısmen | Yok | Yok | Kısmen | Var | | Yedek model stratejisi ve prompt sürümleme | Var | Yok | Kısmen | Kısmen | Kısmen | Yok | Kısmen | | Model proxy: kendi anahtarlarınızla bulut modelleri, izlenebilir ve maskeli | Var | Kısmen | Kısmen | Yok | Yok | Kısmen | Kısmen | | Görüntü ve doküman modelleri (Vision) | Var | Yok | Kısmen | Yok | Kısmen | Yok | Kısmen | | OpenAI uyumlu model API’si kurum uygulamalarına servis edilir: REST, SSE ve WebSocketSağlayıcı değişince istemci kodu değişmez | Var | Yok | Yok | Yok | Kısmen | Yok | Var | | Model-as-a-Service: token bazlı kullanım, grup ve kullanıcı kotası | Var | Yok | Yok | Yok | Kısmen | Yok | Var | | Kanallar | | Sesli ajan: akışlı konuşma tanıma ve çağrı merkezi SIP | Var | Yok | Yok | Kısmen | Kısmen | Yok | Yok | | Diyalog: oturum hafızası, niyet ve varlık blokları | Var | Yok | Kısmen | Var | Var | Yok | Kısmen | | WhatsApp, SMS, e-posta, web widget | Var | Var | Yok | Var | Kısmen | Yok | Yok | | Türkçe konuşma ve yerel dil kalitesi | Var | Yok | Yok | Kısmen | Kısmen | Yok | Kısmen | | Code: IDE eklentileri, kod tamamlama, IDE registry | Var | Yok | Yok | Yok | Kısmen | Var | Yok | | Kod asistanı kapalı devre: VPN, ülke içi model, kod bağlamı kurum sınırındaGüvenlik sayfasındaki senaryo | Var | Yok | Yok | Yok | Yok | Kısmen | Yok | | Bilgi | | Çoklu kaynak RAG, hibrit arama | Var | Kısmen | Var | Var | Var | Kısmen | Kısmen | | Chunk seviyesinde inceleme ve düzenleme | Var | Yok | Yok | Kısmen | Kısmen | Yok | Yok | | Entegrasyon | | MCP sunucuları ve özel araçlar, araç başına çağrı politikası | Var | Kısmen | Var | Kısmen | Kısmen | Var | Kısmen | | Çekirdek sistemlerde yazma işlemleri (core, ERP, CRM) | Var | Var | Var | Kısmen | Kısmen | Yok | Yok | | Hazır bağlayıcı kataloğu (binlerce uygulama)Aginies kurumsal sistemlere REST, veritabanı ve MCP ile bağlanır; her bağlantı politika taşıyan bir araç olarak tanımlanır. Hazır bağlayıcılar e-posta, takvim, dosya ve iş birliği araçlarıyla sınırlıdır; binlerce SaaS uygulaması için bağlayıcı pazaryeri sunulmaz | Kısmen | Var | Yok | Kısmen | Var | Yok | Yok | | Yönetişim | | İnsan onayı kontrol noktaları (eşik ve yetkili bazlı) | Var | Kısmen | Kısmen | Kısmen | Kısmen | Kısmen | Yok | | Her ajan aksiyonu ve model çağrısı için tam denetim izi | Var | Kısmen | Yok | Kısmen | Kısmen | Kısmen | Kısmen | | PII maskeleme, guardrail’ler, kill-switch | Var | Yok | Kısmen | Kısmen | Var | Kısmen | Kısmen | | SSO / SAML, ince taneli RBAC, workspace izolasyonu | Var | Kısmen | Yok | Var | Var | Kısmen | Var | | Workspace / tenant ayrımı; ajanların kurum içinde paylaşımı ve ortak geliştirme | Var | Kısmen | Yok | Kısmen | Var | Kısmen | Kısmen | | Tüm isteklerin maskelenmesi ve anonimleştirilmesi | Var | Yok | Yok | Kısmen | Kısmen | Yok | Kısmen | | Model ve ajan değerlendirme: test korpusu, güvenlik taraması, regülasyon uyum testi | Var | Yok | Kısmen | Kısmen | Kısmen | Yok | Kısmen | | Gözlemlenebilirlik | | Çalıştırma bazlı iz, gecikme ve token takibi | Var | Kısmen | Kısmen | Kısmen | Kısmen | Kısmen | Var | | Senaryo, birim ve kanal bazında maliyet atfı | Var | Yok | Yok | Yok | Kısmen | Yok | Kısmen | | Model bazında kullanım ve maliyet, platform sağlığı panosu | Var | Yok | Yok | Kısmen | Kısmen | Kısmen | Var | | Dağıtım | | On-prem / air-gap kurulum | Var | Kısmen | Var | Yok | Yok | Kısmen | Kısmen | | Hibrit ve dedike bölge seçenekleri | Var | Kısmen | Kısmen | Kısmen | Kısmen | Kısmen | Var | | Ülke içi GPU-as-a-Service | Var | Yok | Yok | Yok | Yok | Yok | Var | | KVKK, BDDK ve GDPR gereksinimlerine göre kurulum | Var | Yok | Yok | Kısmen | Kısmen | Yok | Kısmen | | Self-servis GPU ve Kubernetes kapasitesi, kullandıkça ödeAginies’te kapasite sözleşmeyle ayrılır | Kısmen | Yok | Yok | Yok | Var | Yok | Var | | Çıkarım motoru ve dağıtık GPU zamanlaması ürün olarak (vLLM tabanlı servis, çok küme)Aginies açık motorları işletir, ürünleştirmez | Kısmen | Yok | Yok | Yok | Kısmen | Yok | Var | | İşletim modeli | | Tek sorumlu muhatap: ekip + platform + altyapı | Var | Yok | Yok | Yok | Yok | Yok | Yok | | 5 katmanlı değerlendirme metodolojisi | Var | Yok | Yok | Yok | Yok | Yok | Yok | | SDLC ajan ekibi: PM, analist, mimar, geliştirici, test | Var | Yok | Kısmen | Yok | Kısmen | Kısmen | Yok | Kütüphaneden kanıt ## Tablodaki iddialar senaryo tanımlarıyla desteklenir. Her sayı ajan kütüphanesindeki tanımlardan derleme sırasında türetilir; ilgili satırın kanıtı olarak okunabilir. 73senaryoda insan onayı kuralı 127kurumsal sistemde işlem adımı 23sesli kanalda çalışan senaryo 200bağlı kurumsal sistem 14sektör, tek yürütme katmanı 5model ailesi, ülke içi GPU Örnek ürünlerle tam sürüm Bu sayfa kategorilerle karşılaştırır ve ürün adı vermez. Her kategorinin örnek ürünlerini içeren tam karşılaştırma dokümanlar alanında, erişim koduyla açılır. Dokümanlar › Karşılaştırma ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Metodoloji URL: https://www.aginies.com/tr/metodoloji Language: tr Summary: 5 katmanlı değerlendirme çerçevesi: Değer, Hazırlık, Uygulanabilirlik, Operasyon, Yönetişim. Her aşamanın sonunda net bir karar noktası: devam, düzelt veya durdur. Metodoloji # Her AI fikri yapılmaya değmez. Yapılmaya değenler de hazırlıksız başlatılmamalı. Her kullanım senaryosunu beş katmanlı değerlendirme çerçevesinden geçiririz. Bu çerçeve, portföy önceliklendirmesinden canlı işletime kadar tüm kararların ortak dilidir. Her aşamanın sonunda net bir karar noktası vardır: devam, düzelt veya durdur. 01 ## Değer Bu iş neden yapılmalı? İş gerekçesiyle başlarız: problem, fayda, ölçüm ve stratejik uyum sayılarla tanımlanır. - Problem: hangi iş problemi, bugün nasıl çözülüyor, maliyeti ne? - Fayda: maliyet düşüşü, gelir artışı, risk azaltımı, deneyim; hangisi, ne kadar? - Ölçüm: hedef KPI’lar, beklenen iyileşme, ROI ve geri dönüş süresi - Stratejik uyum: şirket stratejisiyle hizalı mı, yönetim önceliği mi? Süre1–2 hafta Katılımcılarİş birimi sahibi, finans, sponsor Araçlar ve çıktılar - Senaryo portföy çalıştayı ve değer kartı (problem, hacim, maliyet, KPI) - Ajan kütüphanesindeki benzer senaryoyla karşılaştırma Kırmızı bayraklar - Sayısız fayda: “verimlilik artacak” ama hacim ve maliyet yok - Sponsor yok; talep BT’den geliyor, iş birimi izliyor Çıktı Sayılarla konuşan bir iş gerekçesi ve önceliklendirilmiş senaryo portföyü 02 ## Hazırlık Gerekli girdiler elimizde var mı? En iyi fikir bile veri ve organizasyon hazır değilse durur. Başlamadan dört eksende kontrol ederiz. - Veri: kaynaklar mevcut ve dijital mi; kalite, güncellik, sahiplik net mi? - Entegrasyon: bağlanılacak sistemler ve API’ler hazır mı; gerçek zamanlı erişim mümkün mü? - Organizasyon: iş birimi sahipleniyor mu; sponsor ve konu uzmanları belli mi? - Teknoloji: bulut/güvenlik altyapısı, AI platformu ve lisanslar hazır mı? Süre1–2 hafta KatılımcılarVeri sahibi, entegrasyon ekibi, süreç sahibi Araçlar ve çıktılar - Hazırlık karnesi: veri, entegrasyon, süreç, organizasyon (her eksen 0–2) - API ve veri erişim testi; örnek veri seti Kırmızı bayraklar - Aksiyon alınacak sistemin API’si yok, yalnızca ekran - Süreç kuralları yalnızca kişilerin tecrübesinde yaşıyor Çıktı Eksiklerin proje öncesi kapatıldığı bir hazırlık karnesi; sürpriz yok 03 ## Uygulanabilirlik Bunu gerçekten yapabilir miyiz? Dürüst mühendislik değerlendirmesi: AI her zaman doğru cevap değildir. Klasik yazılım, klasik ML, ajan ya da üretken AI; hangisi gerekli? - Teknik uygunluk: problem AI ile mi çözülmeli; hangi yaklaşım? - Karmaşıklık: kaç sistem, kaç veri kaynağı, kaç kullanıcı etkilenecek? - Model performansı: beklenen doğruluk, yanlış sonucun etkisi, insan onayı gerekli mi? - Proje uygulanabilirliği: MVP süresi, ekip, dış bağımlılıklar, teknik riskler Süre2–3 hafta KatılımcılarAginies mühendisliği, kurum mimarisi, bilgi güvenliği Araçlar ve çıktılar - Yaklaşım kararı: klasik yazılım, klasik ML, ajan, üretken AI - Prototip ve test korpusu taslağı; yanlış çıktı etki analizi Kırmızı bayraklar - Yanlış çıktı geri alınamaz sonuç doğuruyor ve onay noktası tanımsız - Kapsam her toplantıda büyüyor Çıktı Net kapsamlı, riskleri tanımlanmış, gerçekçi bir MVP planı 04 ## Operasyon Canlıya çıktıktan sonra nasıl yaşayacak? AI sistemleri bir ürün gibi işletilir; go-live sonrasında izleme, bakım ve sürüm yönetimi işin parçasıdır. - İşletim modeli: ürün sahibi, süreç sahibi, operasyon ve destek ekibi tanımlı - Kullanıcı yönetimi: eğitim, kullanım takibi, geri bildirim döngüleri - İzleme: performans, metrik ve hata yönetimi gerçek zamanlı - Finansal operasyon: aylık işletme maliyeti, kullanım artışında maliyet projeksiyonu, bütçe sahipliği Süre1 hafta (plan) + canlı sonrası sürekli KatılımcılarÜrün sahibi, süreç sahibi, operasyon ve destek Araçlar ve çıktılar - İşletim modeli şablonu: roller, eşikler, inceleme örneklemi, bütçe sahibi - Panolar ve aylık gözden geçirme takvimi Kırmızı bayraklar - Canlıdan sonra kimin işleteceği belirsiz - Aylık işletme maliyeti için bütçe sahibi yok Çıktı Yaşayan, ölçülen ve sürekli iyileşen bir AI ürünü 05 ## Yönetişim Güvenli, kontrollü ve denetlenebilir mi? AI, kurumun diğer kritik sistemleriyle aynı standartlara tabi olmalı. Yönetişim sonradan eklenmez; mimarinin parçasıdır. - Güvenlik: hassas ve kişisel veri haritası, veri yerleşimi, modele veri akışının kontrolü, maskeleme - Uyum: KVKK ve GDPR etkisi, sektörel regülasyonlar, iç politika gereklilikleri - Risk: halüsinasyon, yanlış karar, veri sızıntısı ve prompt injection risklerinin değerlendirilmesi - Kontrol ve denetlenebilirlik: insan onayı noktaları, otomasyon sınırları, acil kapatma, karar geçmişi Süre2 hafta, diğer katmanlarla paralel KatılımcılarBilgi güvenliği, uyum, hukuk, iç denetim Araçlar ve çıktılar - Veri haritası ve maskeleme kural seti; onay matrisi - Kontrol seti ve kanıt listesi; sektörel mevzuat eşlemesi Kırmızı bayraklar - Kişisel veri var, haritası yok - Yönetişim “sonra ekleriz” diye erteleniyor Çıktı Denetime hazır, kontrollü ve güven veren bir AI işletimi Dönüşüm yolculuğu ## Nasıl ilerliyoruz - 1 ### Keşif & portföy İş birimleriyle senaryo toplama, Değer katmanıyla önceliklendirme - 2 ### Değerlendirme Seçilen senaryoların Hazırlık + Uygulanabilirlik analizi, MVP tasarımı - 3 ### MVP & doğrulama Aginies üzerinde geliştirme, gerçek verilerle pilot, KPI doğrulaması - 4 ### Üretim & ölçekleme Operasyon + Yönetişim katmanlarıyla canlıya alma, portföyün büyütülmesi Uygulanmış örnek ## Örnek: diyalog bankacılığı asistanı beş katmandan nasıl geçti - 01 Değer · Bu iş neden yapılmalı? Çağrı merkezine düşen basit işlemler ve işlem başına maliyet sayılarla kondu; hedef %40 azalma. Sponsor: dijital bankacılık. - 02 Hazırlık · Gerekli girdiler elimizde var mı? Çekirdek bankacılık ve kart sistemlerinin işlem API’leri ve test ortamı mevcut; OTP altyapısı hazır; politika eşikleri bilgi güvenliğiyle yazıldı. - 03 Uygulanabilirlik · Bunu gerçekten yapabilir miyiz? Yaklaşım: süpervizör + dört uzman ajan; yanlış çıktı riski için işlem bazlı OTP ve onay bandı; test korpusu 212 senaryo. - 04 Operasyon · Canlıya çıktıktan sonra nasıl yaşayacak? Ürün sahibi dijital kanallar, süreç sahibi operasyon; aylık gözden geçirme ve eşik güncellemeleri takvimde. - 05 Yönetişim · Güvenli, kontrollü ve denetlenebilir mi? Kişisel veri haritası ve maskeleme kuralları; limit artışı > %60, itiraz > 5.000 TL ve yeni alıcı onaya; tüm sistemler Türkiye’de. Senaryoyu canlı izleyin: Diyalog bankacılığı: sesli ve yazılı işlem asistanı → Öz-değerlendirme ## Senaryonuz beş katmandan geçer mi? Aynı çerçeveyi kendi senaryonuza uygulayın. Sonuç, keşif oturumunda konuşacağımız ilk şey olur. ### Senaryonuz beş katmandan geçer mi? 10 soru, 3 dakika. Cevaplar tarayıcınızda kalır; sonucu isterseniz keşif oturumuna taşırsınız. ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Platform URL: https://www.aginies.com/tr/platform Language: tr Summary: Aginies: sekiz entegre yetenek katmanıyla kurumsal agentic AI orkestrasyon platformu. Görsel ajan tasarımcısı, model geçidi, RAG, yönetişim, gözlemlenebilirlik, MCP entegrasyonu ve egemen dağıtım. Platform # Tek platform. AI yürütmenin her katmanı. Ajanları kur, bilgiyi yönet, çalıştırmaları izle, sistemleri bağla. Hepsi tek, üretim sınıfı kontrol düzleminden. Aginies chatbot değildir; kurumun tüm AI senaryolarının koştuğu yürütme katmanıdır. Kütüphaneden türetilmiş platform gerçekleri ## Rakamlar senaryo tanımlarından gelir, elle yazılmaz. Ajan kütüphanesindeki 77 canlı senaryo derleme sırasında okunur; aşağıdaki sayılar ve dağılımlar o tanımlardan türetilir, senaryo eklendikçe değişir. 77canlı senaryo 14sektör 261uzman ajan 657farklı araç 200bağlı sistem 73onay kuralı 127sistemde işlem adımı 3.320.970simüle çalıştırma / ay Kanal dağılımı (senaryo sayısı) api37 app34 web29 whatsapp27 email27 batch25 voice23 sms5 branch3 Adım türleri (toplam adım) sistemde işlem127 araç çağrısı110 akıl yürütme109 doğrulama89 bildirim80 sinyal alma77 insan onayı61 planlama48 bilgi getirme38 Model aileleri (ajan ataması) aginies-tulpar83 aginies-sunkar78 aginies-talay71 aginies-konrul21 aginies-voice-tr8 Otonom Ajan ## Ajanı diyalogla kurun; test edin, onaylayın, üretime alın. Ne istediğinizi yazın: Otonom Ajan süpervizörü ve uzman ajanları kurar, araçları ve bilgi tabanını bağlar, modelleri seçer, test korpusunu yazıp koşturur ve onayınızla dağıtır. Ajan kütüphanesindeki her şey bu diyalogla üretilebilir; çalıştırmalar loglarda ve panoda izlenir. Otonom AjanAginies Workspace · Diyalog bankacılığı: sesli ve yazılı işlem asistanı Otonom Ajan’a yazın… AjanTestlerCanlı Ajan burada oluşacak - Test korpusu: suite, senaryo ve tur bazında; altın yol, dallanmalar, dayanıklılık enjeksiyonu - Kontrol aileleri: müşteri çözümü, PII doğrulama öncesi, durum makinesi, hafıza tur dönüşü, yargıç boyutları - Kapsam raporu: hangi kontroller puanlanıyor, hangileri kör - Değişiklikler versiyonlanır; dağıtım her zaman bir insanın onayında durur - Aynı diyalogdan loglar, konuşmalar ve pano sorgulanır Ürün ailesi ## 7 ürün, tek platform. Platform Aginies çatıdır; Voice, Code, Hub, Vision, Workforce ve Dialog onun modülleridir. Aynı yönetişim, aynı loglar, aynı lisans. ### Platform Aginies Hepsi burada toplanır Orkestrasyon çekirdeği, görsel ajan tasarımcısı, Otonom Ajan, bilgi tabanı, loglar ve panolar. Tüm ajanların kurulduğu, izlendiği ve yönetildiği tek kontrol düzlemi. - Görsel ajan tasarımcısı: ajan, yönlendirici, hafıza, guardrail, insan onayı blokları - Otonom Ajan: diyalogla ajan üretimi, test korpusu, dağıtım - Bilgi tabanı: doküman hattı, sürümleme, chunk editörü - Loglar ve panolar: canlı çalıştırmalar, konuşmalar, ajan bazlı başarı ısı haritası - Workspace ve tenant ayrımı: ekipler ajanlarını kurum içinde paylaşır, birlikte geliştirir - İzleme: ajan bazında ne oluyor, platform sağlığı, model kullanımı ve maliyet - Raporlar: zamanlanmış maliyet, performans, SLA ve denetim raporları; birim, kanal ve şube kırılımı - Şablonlar, kullanıcılar, lisanslar Neler dahil - Görsel ajan tasarımcısı, workspace ve tenant yönetimi, SSO ile roller - Çalıştırma motoru, onay kartları, kill-switch, panolar ve raporlar - Bilgi tabanı yönetimi ve Otonom Ajan Entegrasyon Kurumun kimlik sağlayıcısı (SAML/OIDC), SIEM, veri ambarı ve bilet sistemleriyle. Yönetişim Her workspace ayrı politika seti; dört göz kuralı ve sürümleme platform genelinde. 77 / 77 senaryoda kullanılıyor Platform Aginies · Pano · son 30 güntüm workspace’ler · tüm tenant’lar çalıştırma13.120.000 başarı%97,4 p50 gecikme1,4 s platform sağlığısağlıklı en yoğun ajanlar · her hücre ≈ 1 gün dialog_banking_assistant%97,4 card_debt_payment%99,1 kvkk_data_detection%96,2 kvkk_document_masking%98,8 motor_claim_end_to_end%98,1 merchant_operations%100 model kullanımı aginies-tulpar-438% aginies-talay-229% aginies-sunkar-119% aginies-voice-tr-v39% cloud · proxy$2.960 maliyet · senaryo bazında Diyalog bankacılığı26% Kredi ve edinim24% Hasar operasyonları18% KVKK maskeleme14% Üye işyeri10% Diğer8% PII maskeleme: açıkanonimleştirme: açıkdenetim izi: tam1.920+ aktif ajan ### Voice Aginies STT · TTS · ASR Türkçe için eğitilmiş konuşma modelleri: akışlı konuşma tanıma, doğal ses sentezi ve gerçek zamanlı ASR. Çağrı merkezi SIP entegrasyonuyla sesli ajanlar. - aginies-voice-tr-v3 (STT), aginies-tts-tr-v2, aginies-asr-stream-v1 - On-prem servis; ses verisi kurum dışına çıkmaz - Ses analizi: kalite puanlama, duygu ve niyet Neler dahil - Akışlı Türkçe ASR ve STT (aginies-voice-tr-v3), doğal TTS (tts-tr-v2) - Ses izi doğrulama, OTP okutma, sıcak aktarım - Çağrı kalite analizi: transkript, duygu, uyum kontrolü Entegrasyon Genesys, Avaya ve açık kaynak santrallerle SIP; kayıt sistemleriyle API. Yönetişim Ses kayıtları ülke içinde; kanal bazında kimlik seviyesi ve işlem eşiği. 23 / 77 senaryoda kullanılıyor Voice Aginies · SIP · çağrı 0212-…-4471aginies-voice-tr-v3 · on-prem müşteriDün gece arabama çarptılar, plaka 34 ABC 123, hasar ihbarı açmak istiyorum. ajanGeçmiş olsun. Poliçenizi plakadan buldum; olay yerini ve saati alabilir miyim? STT180 ms niyethasar ihbarı · 0,97 duygusakin TTS · aginies-tts-tr-v2ASR · stream-v1PII maskeleme: açık ### Code Aginies IDE Registry · IDE eklentileri · kod tamamlama Geliştirici ekipleri için: IDE eklentileri ve kod tamamlama, kurumsal IDE registry ve SDLC ajan ekibi (PM, analist, mimar, geliştirici, test) ile CI/CD hattı. - Kurum içi kod tamamlama; kod dışarı çıkmaz - IDE Registry: onaylı eklenti ve model dağıtımı - SDLC ajanları Jira, Git ve CI ile çalışır Neler dahil - IDE eklentileri (VS Code, JetBrains) ve kurum içi kod tamamlama - IDE Registry: onaylı eklenti ve model dağıtımı, sürüm sabitleme - SDLC ajan ekibi: analiz, tasarım, geliştirme, test, inceleme Entegrasyon Jira, Git (GitHub, GitLab, Bitbucket, self-hosted), CI/CD ve statik analiz araçlarıyla. Yönetişim Kod kurum dışına çıkmaz; PR birleştirme ve üretim dağıtımı her zaman insanda. 8 / 77 senaryoda kullanılıyor Code Aginies · billing-service/invoices.tskod tamamlama · on-prem 14export async function exportInvoicePdf(id: string) { 15 const invoice = await repo.findById(id); 16 if (!invoice) throw new NotFound('invoice'); 17 const pdf = await renderer.render('invoice', maskPii(invoice)); 18 await audit.log('invoice.export', { id, user: ctx.user }); 19 return storage.put(`invoices/${id}.pdf`, pdf); 20} Tab: kabul etPROJ-4821test kapsamı %87SAST ✓ IDE RegistryVS Code · JetBrains eklentisiGeliştirici Ajanı · PR #2291 ### Hub Aginies Model Hub · MCP Hub · Model proxy Model kataloğu: Qwen, GLM ve Kimi tabanlı, kurumsal veriyle ince ayarlanmış Aginies model aileleri (tulpar, talay, sunkar, konrul) ve aynı tabanların güncel açık sürümleri, ülke içi GPU’larda. Katalog sürekli güncellenir; yeni sürüm çıktığında Hub’a eklenir, eski sürüm sözleşme süresince kalır. Kendi anahtarlarınızı girin, platform bulut modellerine proxy olarak erişsin: tek noktadan yönetim, kullanım ve maliyet izleme, maskeleme. MCP Hub: onaylı araç sunucuları. - aginies-tulpar-4 · talay-2 · sunkar-1 · konrul-1 (ince ayarlı aileler, on-prem) - Qwen · GLM · Kimi güncel açık sürümleri aynı katalogda; ses için aginies-voice-tr-v3 ve tts-tr-v2 - Model proxy: kendi anahtarlarınızla bulut modelleri, tek noktadan izlenir ve maskelenir - Model değişimi tek seçimle; yedek model stratejisi Neler dahil - Model kataloğu: tulpar, talay, sunkar, konrul aileleri ve güncel açık sürümler - Model proxy ve anahtar yönetimi; kullanım ve maliyet izleme - MCP Hub: onaylı araç sunucuları, çağrı politikaları, bağlantı testleri Entegrasyon Ülke içi GPU havuzlarıyla; kurumun seçtiği bulut model sağlayıcılarıyla proxy üzerinden. Yönetişim Model ve araç değişiklikleri sürümlü ve test kapısından geçer; anahtarlar şifreli kasada. 77 / 77 senaryoda kullanılıyor Hub Aginies · Model Hubtüm workspace’ler · Ekim 2026 | model | erişim | kullanım · son 30 gün | maliyet | | --- | --- | --- | --- | | aginies-tulpar-4 | on-prem · GPU 4× | 58% | — | | aginies-talay-2 | on-prem · GPU 2× | 84% | — | | aginies-sunkar-1 | on-prem · GPU 1× | 91% | — | | aginies-konrul-1 | on-prem · vision | 22% | — | | qwen · güncel sürüm | on-prem · açık model | 31% | — | | aginies-voice-tr-v3 | on-prem · STT | 47% | — | | bulut · proxy | kendi anahtarınız | 12% | $288 | anahtarlar şifreli · kullanıcıya görünmeztüm istekler maskeliMCP Hub · 14 sunucu ### Vision Aginies Doküman · görüntü · maskeleme Görüntü ve doküman anlama: fatura, sözleşme ve form çıkarımı, hasar fotoğrafı değerlendirme, görsel kalite kontrol. KVKK doküman maskeleme ve kişisel veri tespiti aynı modellerle çalışır. - aginies-konrul-1 · görüntü ve doküman · on-prem - Alan çıkarımı: kaynak belge ve güven skoruyla - Kişisel veri tespiti ve doküman maskeleme Neler dahil - Fatura, sözleşme, form ve kimlik belgesi çıkarımı; alan bazında güven skoru - Hasar fotoğrafı değerlendirme ve görsel kalite kontrol - Kişisel veri tespiti ve doküman maskeleme Entegrasyon Doküman depoları (S3, SharePoint), tarayıcı ve ECM sistemleriyle; sonuçlar ERP ve hasar sistemlerine araçla yazılır. Yönetişim Belgeler kurum sınırında işlenir; kaynak belge ve çıkarılan değer birlikte saklanır. 21 / 77 senaryoda kullanılıyor Vision Aginies · fatura_2026_1012.pdfaginies-konrul-1 · on-prem [Alan tespiti yapılmış fatura] çıkarılan alanlar Fatura noFTR-2026-1012 Tarih12.10.2026 KDV%20 · 3.125,00 Toplam18.750,00 TL TCKNmaskelendi (KVKK) kaynak: sayfa 1güven ≥ 0,95 ### Workforce Aginies Kiralık ajan iş gücü Dış kaynak insan yerine dış kaynak ajan: iOS geliştirici, iş analisti, muhasebeci, test mühendisi gibi roller kurumunuzun araçlarında çalışan ajanlar olarak kiralanır. Her görev, çıktı ve saat karşılığı iş kaydında; Aginies tarafında adı belli bir ekip lideri. - Rol kataloğu: mühendislik, iş analizi, finans, destek, İK, hukuk - İş kaydı: görev, çıktı, saat karşılığı, kalite puanı, inceleme durumu - Gölge dönem → örneklemeli inceleme → otonom; eşikler ve kill-switch - Ajan-ay ya da çıktı bazlı faturalama; çıktıların mülkiyeti sizde Neler dahil - Rol kataloğu: mühendislik, iş analizi, finans, destek, İK, hukuk - İş kaydı: görev, çıktı, saat karşılığı, kalite puanı, inceleme durumu - Gölge dönem → örneklemeli inceleme → otonom; adı belli ekip lideri Entegrasyon Kurumun kendi araçlarıyla: Jira, Git, ERP, bilet sistemi, e-posta. Yönetişim Üretim ve finansal kayıt değişiklikleri kurum onayında; çıktıların mülkiyeti kurumda. 17 / 77 senaryoda kullanılıyor Workforce Aginies· kadro ve iş kaydı · fintech-opsPzt 09:00 kiralık ajanlar · 4 ios-dev-01iOS Geliştirici hazır mobil0 h bu hafta qa-01Test Mühendisi hazır mobil0 h bu hafta ba-01İş Analisti hazır ödeme0 h bu hafta acct-01Muhasebe hazır finans0 h bu hafta ### Dialog Aginies Diyalog arayüzü Müşteri ve çalışan tarafındaki diyalog yüzeyi: web widget, mobil, WhatsApp ve mesajlaşma. Oturum hafızası, insan devralma ve her mesajın denetim izi. - Aynı ajan yazılı ve sesli kanalda - İnsan devralma ve geri verme - Konuşma panosu ve kalite puanlama Neler dahil - Web ve mobil diyalog arayüzü; gömülebilir widget - WhatsApp, SMS ve e-posta kanalları aynı ajanla - Onay kartları, dosya ve görsel paylaşımı, sıcak aktarım Entegrasyon Kurumun kimlik ve OTP altyapısıyla; WhatsApp Business API ve SMS geçitleriyle. Yönetişim İşlem bazlı kimlik eşiği; her konuşma ve işlem iz kaydında. 60 / 77 senaryoda kullanılıyor Dialog Aginies · web widgetcard_debt_payment v2 Kart borcumun tamamını vadesiz hesabımdan ödemek istiyorum. Dönem borcunuz 24.860 TL, son ödeme 14 Ekim. Vadesiz ****7741 hesabınızda bakiye yeterli. Onaylarsanız tek kullanımlık şifre göndereceğim. ✓ get_card_debt · get_accounts · politika: onay gerekmez Onaylıyorum. Şifre gönderildi. Girdiğinizde ödeme anında yapılır ve teyit mesajı gelir. oturum hafızasıinsan devralma: hazırdenetim izi Mesaj yazın…↑ Mimari ## Nerede çalışması gerekiyorsa orada. Aynı platform, aynı yönetişim; farklı sınırlar. Dağıtım modunu değiştirin, egemen sınırın nereden geçtiğini görün. Platform, modeller ve veri kurumun kendi veri merkezinde. Dış bağlantı yok; Aginies model aileleri ve açık sürümler içeride servis edilir. [AGINIES ORKESTRASYON ÇEKİRDEĞİ] Kavramsal görünüm. Bileşen adları ve sorumluluklar gösterilir; uygulama ayrıntıları keşif oturumunda paylaşılır. Yetenek katmanları ## Sekiz katman, tek yürütme modeli. A ### Orkestrasyon & Yürütme Zekâsı motor katmanı Karmaşık işi uçtan uca yürütür; talebi akıllıca yönlendirir, çok adımlı süreçleri bağımlılıklar başarısız olsa bile kararlı tutar. - Supervisor → sub-agent → supervisor mimarisi - Dinamik ajan yönlendirme - Durumlu, çok adımlı yürütme - Ajanlar arası iletişim (A2A) - Zamanlayıcı ve webhook tetikleyiciler - Ajan şablonları - Otonom Ajan: diyalogla ajan üretimi ve test korpusu Kapsam - Süpervizör planı, ajan ataması ve sonuç birleştirme; başarısız adımda yeniden deneme ve yedek yol - Durumlu çalıştırma: oturum ve uzun dönem hafıza, bekleyen onaylar, zaman aşımı kuralları - Paralel dallar ve döngüler; koşula bağlı sonlanma Yönetişim Her ajanın yetkisi araç başına tanımlanır; süpervizör kapsam dışı araç çağıramaz. Kanıt Çalıştırma planı ve her adımın girdi/çıktısı iz kaydında; yeniden oynatılabilir. B ### AI / Model Orkestrasyonu beyin seçimi ve dayanıklılık Her iş için doğru modeli kullanır; vendor bağımlılığını azaltır, bir sağlayıcı düştüğünde operasyonu sürdürür. - Çoklu model gateway - Model proxy: kendi anahtarlarınızla bulut modellerine tek noktadan, izlenebilir erişim - Yeniden tasarım gerektirmeyen model değişimi - Yedek model stratejisi - Açık kaynak modeller için on-prem servis - Prompt test ortamı ve sürümleme - Hub Aginies: ince ayarlı kurumsal modeller (c5, g5, g4, oss) ve ses modelleri Kapsam - Hub kataloğu: Qwen, GLM ve Kimi tabanlı ince ayarlı aileler (tulpar, talay, sunkar, konrul) ve güncel açık sürümler - Model proxy: kurumun kendi anahtarlarıyla bulut modelleri, tek noktadan izleme ve maskeleme - Senaryo başına model seçimi, yedek model ve istem sürümleme Yönetişim Model değişikliği sürümlüdür ve test korpusundan geçmeden üretime çıkmaz. Kanıt Her istekte kullanılan model sürümü, token ve maliyet kaydı; katalog geçmişi. C ### Bilgi & Erişim Katmanı kurumsal hafıza Dağınık kurumsal bilgiyi yönetilen, aranabilir ve güvenilir bilgiye dönüştürür; cevaplar kaynağa bağlanır. - Doküman gömme hattı - Sürümlü bilgi tabanı yönetimi - Etiket bazlı doküman ayrımı - Hibrit arama ve çoklu kaynak RAG - Chunk seviyesinde inceleme ve düzenleme - Halüsinasyon azaltma mimarisi - Bilgi tabanı yönetimi: kaynak, etiket ve erişim politikası Kapsam - Doküman gömme hattı, chunk inceleme ve etiketleme; sürümlü bilgi tabanı - Erişim politikası chunk seviyesinde: kim hangi belgeyi hangi ajanla görebilir - Kaynak gösteren cevap; kaynak yoksa “bilmiyorum” Yönetişim Bilgi tabanı değişiklikleri sahibi tarafından onaylanır; eski sürüm geri alınabilir. Kanıt Her cevapta kullanılan chunk’lar ve sürümleri; kapsama ve kalite raporu. D ### Yönetişim, Risk & Uyum kontrol katmanı AI’ı denetlenebilir, güvenli ve regüle ortamlara uygun kılar; her karar izlenebilir. - Guardrail’ler ve politika motoru - İnsan onayı kontrol noktaları - Ajan karar izlenebilirliği - Hassas veri maskeleme ve PII redaksiyonu - Tüm isteklerin maskelenmesi ve anonimleştirilmesi - Tam girdi/çıktı kaydı ve denetim izi - Acil kapatma ve otomasyon sınırları Kapsam - Guardrail ve politika motoru: girdi/çıktı denetimi, maskeleme, yasaklı aksiyonlar - İnsan onayı kontrol noktaları: eşik, yetkili, süre; onay kartı mobil ve web - Kill-switch kullanıcı, ekip ve kurum bazında; rol tabanlı yetki Yönetişim Politika değişiklikleri dört göz kuralıyla; her kural sürümlü ve gerekçeli. Kanıt Onay kayıtları (kim, ne zaman, neye), guardrail müdahaleleri ve kill-switch olayları. E ### Gözlemlenebilirlik & Optimizasyon performans yönetimi Doğruluk, hız, maliyet ve sağlık görünürlüğü sağlar; sürekli iyileştirmenin verisini üretir. - Çalıştırma bazlı iz ve zaman çizelgesi - Gecikme ve token takibi - Senaryo, birim ve kanal bazında maliyet atfı - Model bazında kullanım ve maliyet; platform sağlığı - Yönlendirme doğruluğu ölçümü - Pano ve analitik katmanı - Canlı loglar, konuşma kayıtları, ajan bazlı başarı ısı haritası - Raporlar: zamanlanmış, dışa aktarılabilir (PDF, XLSX, CSV); iç denetim şablonu Kapsam - Çalıştırma bazlı iz: adım, süre, token, araç çağrısı, hata - Panolar: ajan başarı ısı haritası, platform sağlığı, model kullanımı, maliyet atfı - SIEM ve veri ambarına aktarım; uyarı kuralları Yönetişim İzler değiştirilemez ve saklama süresi politikayla belirlenir. Kanıt Denetim örneklemi tek tıkla; ajan bazında aylık kalite raporu. F ### Entegrasyon & Bağlantı kurumsal erişim Mevcut teknoloji ortamını yeniden kurmadan kurumsal sistemlere ve müşteri kanallarına bağlanır. - REST API ve veritabanı entegrasyonları - Model Context Protocol (MCP) sunucuları - Özel araç tanımları ve çağrı politikaları - Hazır bağlayıcılar: e-posta, takvim, dosya, iş birliği araçları - Web widget ve mesajlaşma entegrasyonları - MCP Hub: onaylı araç sunucuları kataloğu Kapsam - REST, veritabanı ve MCP sunucuları; araç başına çağrı politikası - Hazır bağlayıcılar: çekirdek bankacılık, ERP, CRM, bilet, ödeme, santral - Okuma/yazma ayrımı; yazma araçları varsayılan olarak onay ister Yönetişim Her araç servis hesabı ve kapsamla tanımlanır; sır yönetimi kurumun kasasında. Kanıt Araç çağrı günlüğü: parametreler (maskeli), yanıt kodu, süre. G ### Çok Kanallı Deneyim & Ses müşteri etkileşim katmanı Dijital, mesajlaşma ve ses kanallarında tutarlı bir orkestrasyon üzerinde AI sunar. - Gerçek zamanlı konuşma tanıma ve akışlı ses işleme - Çağrı merkezi SIP entegrasyonu - WhatsApp, SMS ve e-posta servisleri - Niyet tespiti ve varlık çıkarımı blokları - Oturum hafızası - Dialog Aginies: diyalog arayüzü, insan devralma - Voice Aginies: Türkçe STT, TTS ve ASR modelleri Kapsam - Dialog Aginies web ve mobil; WhatsApp, SMS ve e-posta kanalları - Voice Aginies: akışlı ASR, Türkçe STT/TTS, SIP ile çağrı merkezi - Aynı ajan tüm kanallarda; sıcak aktarım ve konuşma özeti Yönetişim Kanal başına kimlik doğrulama seviyesi ve işlem eşiği tanımlanır. Kanıt Transkript, kayıt ve özetler saklama politikasıyla; kanal bazında kalite puanı. H ### Dağıtım & Kurumsal Mimari kurumsal kısıtlara uyum Büyük kurumların on-prem, hibrit ve egemenlik gereksinimlerini destekler. - On-prem ve air-gap kurulum - Hibrit mimari: hassas iş yükleri içeride, esnek kapasite dışarıda - Özel bulut / dedike bölge - SSO / SAML, ince taneli RBAC, workspace izolasyonu - Tenant bazında ajan paylaşımı ve ortak geliştirme - GPU-as-a-Service ile ülke sınırları içinde elastik kapasite Kapsam - Egemen paylaşımlı (Türkiye’deki veri merkezlerimiz), dedike tenant/GPU, kurum veri merkezinde on-prem veya air-gap - Erişim: VPN tüneli veya internet üzerinden allowlist - GPU-as-a-Service ile ülke içi esnek kapasite Yönetişim Değişiklik yönetimi ve olay bildirimi sözleşmede; uzaktan bakım kurumun açtığı pencerede. Kanıt Ağ şeması, tünel/allowlist konfigürasyonu, yedekleme ve tatbikat kayıtları. Görsel ajan tasarımcısı ## Ajan hatlarını yapıştırıcı kod yazmadan tasarlayın. Sürükle, bağla, yapılandır. Ajan, yönlendirici, hafıza, guardrail ve insan onayı gibi her blok birinci sınıf bir ilkeldir; herhangi bir akışa kompoze edilir. İş birimi ile BT aynı ekrana bakar. ### Ajan Kendi modeli, araçları ve hafızasıyla bir uzman. ### Süpervizör Hedefi adımlara böler, ajanlara dağıtır, sonucu birleştirir. ### Yönlendirici Niyete, kurala ya da skora göre akışı dallandırır. ### Hafıza Oturum ve uzun dönem bağlamı ajanlar arasında paylaşır. ### Guardrail Girdi ve çıktıyı politikaya göre denetler, durdurur ya da maskeler. ### İnsan onayı Kritik kararda akışı durdurur, yetkiliye sunar, kararı kaydeder. ### Araç / MCP Kurumsal sisteme okuma ya da yazma; çağrı politikası araç başına. ### Paralel & döngü Bağımsız adımları aynı anda, tekrarlı işleri koşula kadar çalıştırır. İzleme ve raporlar ## Her çalıştırma adım adım görünür. Ajan bazında ne olduğu, platformun sağlığı, hangi modelin ne kadar kullanıldığı ve maliyeti tek panoda. Her çalıştırmanın adım adım kaydı: girdi, çıktı, model, süre, token, onaylayan kişi. Tüm istekler maskelenebilir ve anonimleştirilebilir; workspace ve tenant bazında ayrılır. - Ajan bazında çalıştırma, başarı ve hata ısı haritası - Platform sağlığı: gecikme, kuyruk, GPU ve model erişilebilirliği - Model bazında kullanım ve maliyet; proxy üzerinden bulut modelleri dahil - Senaryo, birim ve kanal bazında maliyet atfı - Tüm isteklerde maskeleme ve anonimleştirme - Workspace / tenant ayrımı; ekipler ajanları paylaşır, birlikte geliştirir - Raporlar: zamanlanmış maliyet, performans, SLA ve denetim raporları; PDF, XLSX, CSV; e-posta ve paylaşılan klasöre gönderim - İç denetim için tek rapor: model sürümü, kural seti, kullanıcı Platform Aginies · Pano · son 30 güntüm workspace’ler · tüm tenant’lar çalıştırma13.120.000 başarı%97,4 p50 gecikme1,4 s platform sağlığısağlıklı en yoğun ajanlar · her hücre ≈ 1 gün dialog_banking_assistant%97,4 card_debt_payment%99,1 kvkk_data_detection%96,2 kvkk_document_masking%98,8 motor_claim_end_to_end%98,1 merchant_operations%100 model kullanımı aginies-tulpar-438% aginies-talay-229% aginies-sunkar-119% aginies-voice-tr-v39% cloud · proxy$2.960 maliyet · senaryo bazında Diyalog bankacılığı26% Kredi ve edinim24% Hasar operasyonları18% KVKK maskeleme14% Üye işyeri10% Diğer8% PII maskeleme: açıkanonimleştirme: açıkdenetim izi: tam1.920+ aktif ajan Platform Aginies · Raporlartüm workspace’ler · Ekim 2026 aylık çalıştırma · son 12 ayson 30 gün: 13.120.000+%31 yıllık | rapor | kırılım | zamanlama | biçim | | --- | --- | --- | --- | | Aylık maliyet raporu | senaryo · birim · kanal · şube | her ayın 1’i · 07:00 | PDF · XLSX | | Ajan performansı | başarı, gecikme, hata sınıfları | haftalık · Pazartesi | PDF | | Denetim dışa aktarımı | çalıştırma izleri, onaylar, model sürümleri | talep üzerine | CSV · JSON | | Model kullanımı | model bazında token, maliyet, proxy payı | günlük · 06:00 | XLSX | | SLA ve kapasite | p95 gecikme, kuyruk, GPU doluluk, projeksiyon | haftalık | PDF | e-posta ve paylaşılan klasöre gönderimiç denetim şablonukişisel veri maskeli Workforce Aginies · Kadro ve iş kaydıworkspace: fintech-ops · bu hafta · tüm tenant’lar ayrı | ajan | rol · ekip | saat karş. | çıktı | kalite | mod | | --- | --- | --- | --- | --- | --- | | ios-dev-01 | iOS Geliştirici · mobil ekibi | 14 h | 9 PR · 21 puan | 4,7 | otonom · %10 inceleme | | qa-01 | Test Mühendisi · mobil ekibi | 31 h | 412 test · 6 hata | 4,8 | otonom · %10 inceleme | | ba-01 | İş Analisti · ödeme ürünü | 9 h | 11 hikâye · 2 süreç | 4,5 | gölge · %100 inceleme | | acct-01 | Muhasebe · finans | 27 h | 238 kayıt · 3 mutabakat | 4,9 | otonom · eşik onaylı | iş kaydı · bugün 09:12ios-dev-01PR #4821 açıldı: ödeme ekranı erişilebilirlik düzeltmeleri · 2 test eklendiinceleme bekliyor 09:40acct-0138 tedarikçi faturası işlendi, 2 tanesi tolerans dışı → m.ozkan onayınaonay 10:05qa-01Regresyon 3. tur: 412/412 · 1 yeni hata (BUG-2291) kanıtla açıldıtamam 10:30ba-01PAY-118 hikâyesi ve 6 kabul kriteri yazıldı → ürün sahibineinceleme haftalık zaman çizelgesi dışa aktarıldıfaturalama: ajan-ayekip lideri: e.demir (Aginies) Entegrasyon ## MCP ile herhangi bir aracı takın. Sunucu adresini ekleyin, başlıkları yapılandırın, bağlantıyı test edin: ajanlar araçları güvenle çağırabilir. Chunk seviyesinde erişim kontrolü ile her bilgi parçası incelenebilir, düzenlenebilir, etiketlenebilir. serverhttps://mcp.erp.internal/v1connected toolsget_purchase_order · post_invoice · get_vendor policypost_invoice → insan onayı > 50.000 authSSO · workspace: finance-ops chunks4.212 · etiket: politika/v3 Entegrasyon kataloğu ## 200 sistem, 12 tür: senaryoların dokunduğu kurumsal sistemler. Her satır bir sistem türü ve kütüphanedeki örnekleri. Yeni bir sistem bağlamak, bir araç tanımı ve çağrı politikası demektir. Kurumsal API’ler76 bağlantı Access ControlAppointment SystemBank SIM-Swap Signal APICI PipelineCI Runner (GitHub Actions)CI/CDCI/CD (signed artefacts)Card Management+55 Müşteri kanalları50 bağlantı Customer PortalHangar Tablet AppMicrosoft TeamsMobile App ChatOTP / SMS GatewayPush & SMS GatewaySMS GatewayShift Channel (Teams)+6 Çekirdek sistemler44 bağlantı Active Directory / Entra IDBSS / BillingBilling (Amdocs)Billing (CIS)CLM (Contract Lifecycle Management)Card ManagementClaims SystemCore Banking+19 Veritabanı ve vektör44 bağlantı Audit LogCollections PlatformControl & Sanctions Lists (vector)Eligibility Criteria (vector)Fleet RegistryGIS / Network ModelGIS / Zoning PlanKYC / Customer 360+29 Bilet ve iş emri25 bağlantı CMMSCMMS (Maximo)Case ManagementField ServiceITSM / IncidentJiraJira (QA)Jira / Xray+6 ERP ve MES25 bağlantı Billing / CIS (SAP IS-U)Billing / SettlementERPERP (SAP / Logo)ERP (SAP GTS)ERP BillingERP Spare PartsInventory+12 E-posta20 bağlantı E-mailEmailEmail (Exchange)Exchange / M365 MailProcurement Inbox Doküman depoları19 bağlantı ConfluenceContract RepositoryDocument StoreDocument Store (S3)Tech Pub Repository (S1000D) CRM8 bağlantı CRMCustomer CRM / BillingCustomer ProfileHubSpotSalesforce CRM Ödeme sistemleri7 bağlantı Bank APIPSP / Acquirer APIPayment Gateway (iyzico)Payment HubRefund ServiceSettlement Ledger Ses ve santral6 bağlantı Call Centre SIPGenesys SIPIVR / SIP İK sistemleri3 bağlantı LMSSAP SuccessFactorsWorkday İşletim ## Canlıya çıktıktan sonra nasıl işletilir. Hedefler sözleşmede kurumla birlikte netleşir; aşağıdakiler platformun standart işletim modelidir. ### Sürümleme ve geri alma Ajan tanımı, kural seti, istem ve model sürümü birlikte kaydedilir; bir önceki sürüme dönüş tek adımdır. ### Test korpusu kapısı Her değişiklik üretime çıkmadan suite, senaryo ve tur bazlı test korpusundan geçer; kapsama raporu kayıt altındadır. ### Süreklilik hedefleri Egemen paylaşımlı kurulumda hedef RTO 4 saat, RPO 1 saat; dedike ve on-prem kurulumda kurumla birlikte belirlenir. Yedekler Türkiye içinde. ### Model kataloğu bakımı Yeni taban sürümü çıktığında ince ayar tekrarlanır, katalogda yeni sürüm açılır; eski sürüm sözleşme süresince kalır. ### Olay yönetimi Önem derecesine göre bildirim süreleri sözleşmede; kill-switch kurumda, kök neden raporu Aginies’te. ### Aylık gözden geçirme Ajan bazında kalite, maliyet ve onay oranları; eşik ve kural güncellemeleri kurumun yöneticisiyle birlikte. Kurulum modelleri ## Aynı platform, üç kurulum. Egemen paylaşımlıdan air-gap’e kadar; erişim VPN tüneli veya allowlist ile. Ayrıntı güvenlik sayfasında. Güvenlik → önerilen başlangıçEgemen paylaşımlı Türkiye’deki veri merkezlerimiz, kuruma izole workspace, VPN ile erişim. - Günler içinde başlar - Model güncellemeleri ve kapasite Aginies’te - Kuruma sabit IP, izole tenant, SIEM aktarımı daha yüksek izolasyonDedike tenant / GPU Aynı veri merkezinde kuruma ayrılmış tenant ve GPU havuzu; istenirse kurumun anahtarıyla şifreleme. - Kaynak paylaşımı yok - Kuruma özel model sürümleri - Kapasite taahhüdü sözleşmeyle tam kontrolKurum veri merkezinde on-prem Hub, model sunucuları ve Platform kurumun kendi veri merkezinde; uzaktan bakım yalnızca kurumun açtığı pencerede. - Hiçbir veri kurum sınırından çıkmaz - Kurumun GPU yatırımı gerekir - Kurulum ve güncelleme kurumun değişiklik takvimiyle ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Sektörler URL: https://www.aginies.com/tr/sektorler Language: tr Summary: Bankacılık, fintech, sigorta, perakende, telekom, sağlık, üretim, lojistik, enerji, kamu ve hükümet, savunma sanayii, IoT ve yazılım geliştirme için Aginies: sektörün regülasyonuna, sistemlerine ve senaryolarına göre kurgulanmış AI orkestrasyonu. Sektörler # On üç sektör, altmış ajan: her sektörün kendi sistemleri, kuralları ve onay noktaları var. Aynı platform, sektöre göre kurgulanmış senaryolar. Regüle sektörlerde uyum sonradan eklenmez; mimarinin parçasıdır. Savunma ve kamuda kapalı ağ, IoT’de küçük on-prem modeller, enerjide şebeke sistemlerine salt okuma: her sektörün kurulumu kendi şartlarına göre. Finans ve sigorta BDDK, MASAK, SEDDK: denetim en sıkı, onay noktaları en net sektörler. ## Bankacılık 9 ajan Çekirdek sistem yerinde kalır; üzerine karar veren, aksiyon alan bir AI katmanı gelir. PoC çukuruLegacy çekirdekGörünmeyen maliyet 4 uyum ilkesi · 3 çözüm Devamı → ## Finansal Teknolojiler 5 ajan Onboarding, AML, kredi kararı ve uzlaştırma: hız ile uyumun aynı anda gerektiği yerde ajanlar. Onboarding darboğazıUyarı yorgunluğuSaniyelik kredi kararı 4 uyum ilkesi · 3 çözüm Devamı → ## Sigorta 6 ajan Hasar bildiriminden tahsilata: her adımı izlenen, onay noktaları belli süreçler. Fırtına günü kapasitesiBelge kaosuAçıklanamayan karar 4 uyum ilkesi · 3 çözüm Devamı → Yüksek hacimli müşteri operasyonu Milyonlarca etkileşim, binlerce senaryo: kanal ve çekirdek sistem arasında ajanlar. ## Perakende & E-ticaret 6 ajan Her müşteri sorusunun sonunda bir işlem bekler; ajanlar o işlemi sipariş sisteminde yapar. Sezon zirvesiCevap veren, işlem yapmayan botKatalog borcu 3 uyum ilkesi · 3 çözüm Devamı → ## Telekom 5 ajan Milyonlarca abone, binlerce senaryo: fatura, arıza, dolandırıcılık ve teklif tek katmanda. Fatura itirazı hacmiSIM değişikliği dolandırıcılığıArıza iletişimi 3 uyum ilkesi · 3 çözüm Devamı → ## Enerji 5 ajan Kesinti, sayaç, bağlantı başvurusu ve talep yanıtı: şebeke sistemlerine bağlı, müşteriyle konuşan ajanlar. Kesinti çağrı yığınıTahmini fatura itirazlarıDağıtık üretim başvuruları 3 uyum ilkesi · 3 çözüm Devamı → ## Sağlık 4 ajan Klinik karar hekimde kalır; ön yetkilendirme, kabul ve randevu operasyonunu ajanlar taşır. Ön yetkilendirme gecikmesiKabul ve triyaj yüküRandevu kayıpları 3 uyum ilkesi · 3 çözüm Devamı → Operasyon ve altyapı ERP, MES, TMS ve IoT platformlarının üstünde karar veren, sevk eden ajanlar. ## Üretim ve Otomotiv 4 ajan Tedarikçi faturasından bakım talebine: ERP ve MES’in üstünde karar veren, iş emri açan ajanlar. Sistemler arası kopuklukBakım talebi yığınıFatura ve mal kabul farkları 3 uyum ilkesi · 3 çözüm Devamı → ## Lojistik 3 ajan Sevkiyat istisnaları, gümrük belgeleri ve soğuk zincir: olay olduğu anda karar veren ajanlar. İstisna geç fark edilirGümrük belgesi hatalarıSoğuk zincir görünürlüğü 3 uyum ilkesi · 3 çözüm Devamı → ## IoT ve Bağlı Cihazlar 4 ajan Binlerce cihazdan gelen sinyali tek olaya, tek karara indiren ajanlar: anomali, OTA, soğuk zincir. Alarm gürültüsüRiskli OTA dağıtımıSensörden aksiyona boşluk 3 uyum ilkesi · 3 çözüm Devamı → Kamu, savunma ve teknoloji Kapalı ağ, açıklanabilir karar, kaynak kodun içeride kalması. ## Kamu ve Hükümet 5 ajan Vatandaş başvurusundan ihale dosyasına: mevzuata bağlı, açıklanabilir, kararı memurda bırakan ajanlar. Başvuru yönlendirmeDosya ön inceleme yüküAçıklanabilirlik 4 uyum ilkesi · 3 çözüm Devamı → ## Savunma Sanayii 4 ajan İhracat kontrolü, bakım (MRO) ve tedarik zinciri güvenliği: tamamen kapalı ağda çalışan ajanlar. İhracat uyum yüküTeknik yayın karmaşasıTedarikçi güvenliği 4 uyum ilkesi · 3 çözüm Devamı → ## Yazılım Geliştirme 8 ajan PM, analist, mimar, geliştirici ve test ajanları: özellik talebinden üretime giden hattın tamamı. Gereksinim belirsizliğiİnceleme darboğazıSürüm riski 3 uyum ilkesi · 3 çözüm Devamı → ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Bankacılık URL: https://www.aginies.com/tr/sektorler/banking Language: tr Summary: Çekirdek sistem yerinde kalır; üzerine karar veren, aksiyon alan bir AI katmanı gelir. Bankacılık # Çekirdek sistem yerinde kalır; üzerine karar veren, aksiyon alan bir AI katmanı gelir. Türkiye’deki bankaların neredeyse tamamı yapay zekâ ile pilot yürütüyor; çok azı bunu çekirdek süreçlerde, ölçekte ve denetime hazır biçimde işletebiliyor. Aginies, kartlar, krediler, çağrı merkezi ve dijital kanallar için ortak bir orkestrasyon katmanı kurar: her ajan, her servis, her limit ve her onay merkezi ve denetlenebilir kurallarla tanımlanır. Kısaca Aginies, Bankacılık sektöründe 9 canlı ajan senaryosu ve 3 çözüm paketi sunar: Diyalog bankacılığı: sesli ve yazılı işlem asistanı, Kredi kartı limit artışı, Dijital müşteri edinimi (KYC), Kart dolandırıcılık alarmı triyajı, Ters ibraz (chargeback) itirazı, Erken dönem tahsilat iletişimi, Konut kredisi belge incelemesi, Sesli bankacılıkla para transferi; çözümler: Diyalog bankacılığı, Kredi ve müşteri edinimi operasyonları, Dolandırıcılık ve itiraz operasyonları. Üretime geçişi engelleyen başlıklar PoC çukuru, Legacy çekirdek, Görünmeyen maliyet, Uyum sonradan geliyor; uyum tarafında 4 gereklilik (BDDK bilgi sistemleri gerekliliklerine uyumlu kurulum; birincil ve ikincil sistemler Türkiye’de dahil) kurulumun parçasıdır. Neden üretime ulaşamıyor 01 ### PoC çukuru Chatbot pilotları etkileyici, ama çekirdek bankacılık işlemine dokunan bir üretim yolu tanımsız. 02 ### Legacy çekirdek Ana sistemler AI’ın karar alıp aksiyon almasına göre tasarlanmadı; her entegrasyon ayrı proje. 03 ### Görünmeyen maliyet Hangi senaryo ne kadar token tüketiyor, hangi karar hangi veriyle verildi, raporlanamıyor. 04 ### Uyum sonradan geliyor BDDK gereklilikleri, KVKK ve iç denetim beklentileri projenin sonunda gündeme gelip projeyi durduruyor. Nereden başlanır: ilk 90 gün Üç başlangıç noktası; her biri mevcut sistemlerle kısa bir pilotta ölçülebilir. - 01 ajan senaryosuDiyalog bankacılığı: sesli ve yazılı işlem asistanı → En görünür etki: çağrı merkezi yükü ve müşteri deneyimi aynı anda; mevcut kimlik altyapısıyla pilot 6 haftada. - 02 ajan senaryosuKart dolandırıcılık alarmı triyajı → Analist kuyruğunu hemen rahatlatır; karar hâlâ insanda olduğu için bilgi güvenliği onayı hızlı gelir. - 03 çözüm paketiKredi ve müşteri edinimi operasyonları → Belge inceleme ve KYC, ölçülebilir en büyük iş yükü; otomatik onay bandı kredi politikasıyla birlikte yazılır. Senaryolar ## Bu sektörde canlı izleyebileceğiniz ajanlar BankacılıkHITL ### Diyalog bankacılığı: sesli ve yazılı işlem asistanı Müşteri konuşur ya da yazar; fatura, limit, itiraz ve transfer aynı oturumda çekirdek sistemde tamamlanır. 186.000 çalıştırma / ay %94,7 başarı 5 ajan BankacılıkHITL ### Kredi kartı limit artışı Müşteri mobilde yazar ya da çağrı merkezinde konuşur; karar saniyeler içinde verilir, eşik üstü yönetici onayına gider. 18.400 çalıştırma / ay %96,4 başarı 4 ajan BankacılıkHITL ### Dijital müşteri edinimi (KYC) Kimlik, canlılık ve yaptırım kontrolü tek akışta; temiz başvurular şubeye gitmeden dakikalarda hesap açar. 42.000 çalıştırma / ay %93,8 başarı 5 ajan BankacılıkHITL ### Kart dolandırıcılık alarmı triyajı Kart saniyeler içinde bloklanır, müşteriye WhatsApp'tan sorulur; yanıta göre kart açılır ya da iptal edilir. 65.000 çalıştırma / ay %97,1 başarı 4 ajan BankacılıkHITL ### Ters ibraz (chargeback) itirazı İtiraz alınır, kanıt paketi toplanır ve kart şemasına dosyalanır; eşik üstü geçici alacak onaya gider. 9.800 çalıştırma / ay %95,2 başarı 4 ajan BankacılıkHITL ### Erken dönem tahsilat iletişimi Gecikmiş hesaplar segmentlenir, doğru kanaldan nazikçe hatırlatılır; müşteri ödeme planını kendisi kurar. 96.000 çalıştırma / ay %91,2 başarı 4 ajan BankacılıkHITL ### Konut kredisi belge incelemesi Bordro, SGK dökümü ve ekspertiz otomatik okunur; gelir, LTV ve eksik belge kontrolü danışman beklemeden biter. 6.200 çalıştırma / ay %94,7 başarı 4 ajan BankacılıkHITL ### Sesli bankacılıkla para transferi Müşteri arar ve konuşur; ses izi, güvenlik sorusu ve OTP ile doğrulanır, transfer temsilcisiz tamamlanır. 27.000 çalıştırma / ay %93,1 başarı 5 ajan Sektör mimarisi Sektördeki 9 senaryodan türetilmiştir: süpervizörler, uzman ajanlar, Hub model aileleri ve dokunulan sistemler. Süpervizörler Diyalog Bankacılığı SüpervizörüKart SüpervizörüMüşteri Edinim SüpervizörüDolandırıcılık Süpervizörüİtiraz SüpervizörüTahsilat SüpervizörüKredi Dosya SüpervizörüSesli İşlem Süpervizörü Uzman ajanlar Diyalog AjanıKimlik Ajanıİşlem AjanıRisk AjanıNiyet AjanıBelge AjanıCanlılık AjanıTarama AjanıKayıt AjanıTriyaj Ajanıİletişim AjanıKart AjanıKabul AjanıKanıt AjanıDosyalama AjanıSegment AjanıÖdeme Planı AjanıÇıkarım AjanıGelir AjanıDeğerleme AjanıKonuşma AjanıTransfer AjanıÖdeme AjanıSes Ajanı Modeller (Hub kataloğu) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-konrul-1aginies-voice-tr-v3 Dokunulan sistemler Mobile App ChatCore BankingCard ManagementFraud EngineSMS GatewayKYC / Customer 360Push & SMS GatewaySanctions Screening (Dow Jones)Document Store (S3)WhatsApp Business APISalesforce CRMVisa / Mastercard Dispute APIEmail (Exchange)Collections PlatformLoan Origination SystemCredit Bureau (KKB)Genesys SIPKnowledge Base · kart_sss_v3OTP / SMS Gateway Kanallar appvoicewhatsappwebsmsapiemailbranchbatch Simülasyon hacimleri 502.400çalıştırma / ay %94,6başarı oranı 6,5 smedyan süre $0,0028tahmini maliyet / çalıştırma Sektör senaryolarının simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Uyum ve egemenlik - BDDK bilgi sistemleri gerekliliklerine uyumlu kurulum; birincil ve ikincil sistemler Türkiye’de - Kritik kararlarda kill-switch ve zorunlu insan onayı - Model, platform ve veri bankanın kendi veri merkezinde; dış bağlantı olmadan - Her ajan aksiyonu ve model çağrısı için tam denetim izi Uyum çerçevesi - BDDK Bilgi Sistemleri ve Elektronik Bankacılık Yönetmeliği: birincil ve ikincil sistemler Türkiye’de, dış hizmet sözleşmesinde denetim hakkı, bulut modeli bankaya özgü izin - MASAK ve KVKK: şüpheli işlem bildirimi analist kararında, kişisel veri maskeleme ve amaç sınırlaması her senaryoda - İç denetim ve BDDK incelemesi için istek bazlı denetim izi: kullanıcı, model sürümü, veri kaynağı, gerekçe Güvenlik → Önerilen kurulum Başlangıç için Türkiye’deki veri merkezlerimizde egemen paylaşımlı kurulum (banka workspace’i izole, VPN veya allowlist ile erişim); birincil sistem sayılan senaryolar için bankanın veri merkezinde on-prem. Onay kuralları Sektör senaryolarının insan onayı eşikleri. - Diyalog bankacılığı: sesli ve yazılı işlem asistanıLimit artışı > %60, itiraz tutarı > 5.000 TL veya yeni alıcıya transfer → operasyon onayı - Kredi kartı limit artışıLimit > 100.000 TL → şube yöneticisi onayı - Dijital müşteri edinimi (KYC)Yaptırım/PEP bulanık eşleşme skoru > 0,65 → uyum uzmanı onayı - Kart dolandırıcılık alarmı triyajıTeyitli dolandırıcılık zararı > 2.500 TL → analist onayı ile kalıcı iptal - Ters ibraz (chargeback) itirazıTutar > 2.500 TL → geçici alacak öncesi itiraz ekibi lideri onayı - Erken dönem tahsilat iletişimiPlan bakiyesi > 25.000 TL veya > 3 taksit → tahsilat ekip lideri onayı - Konut kredisi belge incelemesiKredi tutarı > 3.000.000 TL veya DTI > %45 → kredi tahsis uzmanı onayı - Sesli bankacılıkla para transferiTutar > 50.000 TL veya yeni alıcı → çağrı merkezi temsilcisine sıcak aktarım - Kredi kartı borcu ödemeTutar > 100.000 TL → şube yöneticisi onayı Ölçülen hedefler | Çağrı merkezine düşen basit işlem | %40 azalma | | Dolandırıcılık alarmı inceleme süresi | %70 kısalma | | Şubeye gitmeden açılan hesap payı | %70–85 | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef pilotun ilk haftasında birlikte sabitlenir. İlgili çözüm paketleri Diyalog bankacılığıMobil bankacılık işlemlerini yazarak ya da konuşarak gerçekleştir.Kredi ve müşteri edinimi operasyonlarıDijital hesap açılışı, limit artışı ve kredi belgeleri tek karar mantığıyla, şubeye gitmeden.Dolandırıcılık ve itiraz operasyonlarıAlarm triyajı, ters ibraz dosyaları, kart bloke ve erken tahsilat tek operasyon masasında. ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Savunma Sanayii URL: https://www.aginies.com/tr/sektorler/defence Language: tr Summary: İhracat kontrolü, bakım (MRO) ve tedarik zinciri güvenliği: tamamen kapalı ağda çalışan ajanlar. Savunma Sanayii # İhracat kontrolü, bakım (MRO) ve tedarik zinciri güvenliği: tamamen kapalı ağda çalışan ajanlar. Savunma ve havacılık şirketlerinde hız ile güvenlik aynı anda gerekir: ihracat lisansı tespiti, platform bakım planlaması, tedarikçi nitelendirme. Aginies bu süreçleri kurumun kapalı ağında, açık kaynak modellerle, dış bağlantı olmadan çalıştırır. Kontrole tabi her karar, uçuşa elverişliliği etkileyen her iş paketi ve her nitelendirme insanda kalır; ajanlar sınıflandırır, eşler, hazırlar. Kısaca Aginies, Savunma Sanayii sektöründe 4 canlı ajan senaryosu ve 3 çözüm paketi sunar: Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolü, İhracat kontrolü ve son kullanıcı taraması, Platform bakım (MRO) iş emri hazırlığı, Tedarikçi güvenlik ve uyum değerlendirmesi, Sözleşme madde çıkarımı, Kod incelemesi ve güvenlik kapısı; çözümler: Bilgi yönetimi ve karar desteği, Saha servis ve sevk otomasyonu, Savunma ihracat kontrolü ve tedarik zinciri güvenliği. Üretime geçişi engelleyen başlıklar İhracat uyum yükü, Teknik yayın karmaşası, Tedarikçi güvenliği, Bulut yasağı; uyum tarafında 4 gereklilik (Air-gap kurulum; modeller, platform ve veri kapalı ağın içinde dahil) kurulumun parçasıdır. Neden üretime ulaşamıyor 01 ### İhracat uyum yükü Her kalem için sınıflandırma ve taraf taraması elle; uzmanlar dosya başına saatler harcıyor. 02 ### Teknik yayın karmaşası Görev kartı ve parça listesi binlerce sayfada; doğru revizyonu bulmak zaman alıyor. 03 ### Tedarikçi güvenliği Anketler ve belgeler elle okunuyor; eksikler nitelendirme sonrası fark ediliyor. 04 ### Bulut yasağı Gizlilik dereceli ortamlarda hiçbir dış model kullanılamıyor. Nereden başlanır: ilk 90 gün Üç başlangıç noktası; her biri mevcut sistemlerle kısa bir pilotta ölçülebilir. - 01 ajan senaryosuTeknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolü → Hangar ve bakım ekiplerinin her gün kullandığı bilgi; kapalı ağda düşük riskle başlar. - 02 ajan senaryosuİhracat kontrolü ve son kullanıcı taraması → Taraf tarama ve kontrol listesi eşleme kurallı; lisans kararı uyum yöneticisinde kalır. - 03 ajan senaryosuTedarikçi güvenlik ve uyum değerlendirmesi → Tedarikçi anketleri belge ağırlıklı; değerlendirme süresi hemen ölçülür. Senaryolar ## Bu sektörde canlı izleyebileceğiniz ajanlar Savunma SanayiiHITL ### Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolü Teknisyen hangarda sorar; doğru revizyondaki prosedür, parça numarası ve emniyet notu yetki kontrolüyle gelir, her erişim iz bırakır. 9.800 çalıştırma / ay %95,7 başarı 5 ajan Savunma SanayiiHITL ### İhracat kontrolü ve son kullanıcı taraması Parça, alıcı ve son kullanım tek akışta sınıflandırılır; lisans gereği tespit edilir, karar her zaman uyum yöneticisinde kalır. 2.600 çalıştırma / ay %96,1 başarı 4 ajan Savunma SanayiiHITL ### Platform bakım (MRO) iş emri hazırlığı Arıza raporu teknik yayınlarla eşlenir, görev kartı ve parça listesi çıkar, iş emri planlayıcı onayıyla açılır. 4.100 çalıştırma / ay %95,7 başarı 5 ajan Savunma SanayiiHITL ### Tedarikçi güvenlik ve uyum değerlendirmesi Tedarikçi anketleri ve belgeleri okunur, tesis güvenlik gereklilikleriyle karşılaştırılır, eksikler listelenir; nitelendirme kararı güvenlik müdürünün. 900 çalıştırma / ay %95,2 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Sözleşme madde çıkarımı Sözleşme maddeleri çıkarılır, hukuk playbook'u ile karşılaştırılır; sapmalar redline önerisiyle avukata gider. 1.400 çalıştırma / ay %95,1 başarı 4 ajan Yazılım GeliştirmeHITL ### Kod incelemesi ve güvenlik kapısı Her PR ajanlarca incelenir, taranır, kapsamı ölçülür; geliştirici düzeltir, insan yalnızca temiz PR'ı onaylar. 4.200 çalıştırma / ay %93,6 başarı 5 ajan Sektör mimarisi Sektördeki 4 senaryodan türetilmiştir: süpervizörler, uzman ajanlar, Hub model aileleri ve dokunulan sistemler. Süpervizörler Teknik Yayın Süpervizörüİhracat Uyum SüpervizörüBakım SüpervizörüTedarikçi Güvenlik Süpervizörü Uzman ajanlar Yetki AjanıYayın AjanıProsedür AjanıParça AjanıSınıflandırma AjanıTaraf Tarama AjanıLisans AjanıArıza AjanıTeknik Yayın AjanıPlanlama AjanıBelge AjanıDeğerlendirme AjanıDosya Ajanı Modeller (Hub kataloğu) aginies-sunkar-1aginies-tulpar-4aginies-talay-2aginies-konrul-1 Dokunulan sistemler Hangar Tablet AppAccess ControlTech Pub Repository (S1000D)MRO SystemAudit LogERP (SAP GTS)Control & Sanctions Lists (vector)Trade Compliance Case SystemE-mailMaintenance Management SystemTechnical Publications (S1000D vector)Spares InventoryTechnician AppSupplier Management (SRM)Security Requirements (vector)Document Store Kanallar appapiemailbatchweb Simülasyon hacimleri 17.400çalıştırma / ay %95,7başarı oranı 7,6 smedyan süre $0,0053tahmini maliyet / çalıştırma Sektör senaryolarının simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Uyum ve egemenlik - Air-gap kurulum; modeller, platform ve veri kapalı ağın içinde - Kontrole tabi kalemlerde lisans kararı yalnızca ihracat uyum yöneticisinde - Gizlilik derecesine göre erişim ayrımı, tam denetim izi, kill-switch - Tesis güvenliği ve tedarik zinciri gereklilikleriyle uyumlu kontrol seti Uyum çerçevesi - SSB ve ihracat kontrol mevzuatı: kontrole tabi liste eşlemesi ajanla, lisans ve paylaşım kararı yetkili uyum sorumlusunda - Tesis güvenlik belgesi ve gizlilik dereceleri: erişim ayrımı, işaretleme kontrolü ve her erişimin denetim günlüğü - Müttefik programlarda ITAR/EAR benzeri kısıtlar: kontrollü teknik veri yalnızca yetkili kullanıcıya, dış ağa çıkış yok Güvenlik → Önerilen kurulum Air-gap kurulum: modeller, platform ve veri kapalı ağın içinde; güncellemeler onaylı paketlerle, uzaktan bakım yalnızca kurumun açtığı pencerede. Onay kuralları Sektör senaryolarının insan onayı eşikleri. - Teknik yayın asistanı: bakım el kitabı, parça, prosedür ve revizyon kontrolüProsedürden sapma önerisi → yetkili mühendis onayı; ihracat kontrollü içerik → yetki seviyesi şart - İhracat kontrolü ve son kullanıcı taramasıKontrole tabi her kalem → ihracat uyum yöneticisi onayı; ajan lisans kararı veremez - Platform bakım (MRO) iş emri hazırlığıUçuşa elverişliliği etkileyen her iş paketi → bakım planlayıcısı onayı - Tedarikçi güvenlik ve uyum değerlendirmesiNitelendirme kararı her zaman tesis güvenlik müdüründe; ajan yalnızca hazırlar Ölçülen hedefler | Taraf tarama süresi | Günlerden dakikalara | | Teknik yayın sorgusu yanıt süresi | Saniyeler, doğru revizyonla | | Dış ağa çıkan veri | Sıfır | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef pilotun ilk haftasında birlikte sabitlenir. İlgili çözüm paketleri Bilgi yönetimi ve karar desteğiMevzuat, politika ve sözleşmelere saniyeler içinde kaynaklı erişim.Saha servis ve sevk otomasyonuSensör alarmı ya da müşteri bildirimi → doğru teknisyen, doğru parça, doğru zaman.Savunma ihracat kontrolü ve tedarik zinciri güvenliğiSınıflandırma, taraf tarama, lisans dosyası ve tedarikçi nitelendirme: kapalı ağda, kararı uzmanda bırakan ajanlar. ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Enerji URL: https://www.aginies.com/tr/sektorler/energy Language: tr Summary: Kesinti, sayaç, bağlantı başvurusu ve talep yanıtı: şebeke sistemlerine bağlı, müşteriyle konuşan ajanlar. Enerji # Kesinti, sayaç, bağlantı başvurusu ve talep yanıtı: şebeke sistemlerine bağlı, müşteriyle konuşan ajanlar. Dağıtım ve perakende enerji şirketleri iki dünyayı aynı anda yönetir: SCADA, GIS ve sayaç verisiyle şebeke; CRM ve faturalama ile milyonlarca müşteri. Aginies iki tarafı bağlayan ajanlar kurar: kesinti bildirimlerini kümeler, sayaç itirazlarını görüntüden çözer, çatı GES başvurularını ön değerlendirir, talep yanıtı olaylarını yürütür. Mühendislik ve regülasyon kararları insanda kalır. Kısaca Aginies, Enerji sektöründe 5 canlı ajan senaryosu ve 3 çözüm paketi sunar: Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda, Kesinti bildirimi yönetimi, Sayaç okuma itirazı, Çatı GES bağlantı başvurusu, Talep yanıtı (demand response) olayı, Fatura itirazı çözümü; çözümler: Fatura ve doküman veri çıkarımı, Saha servis ve sevk otomasyonu, Enerji müşteri operasyonları. Üretime geçişi engelleyen başlıklar Kesinti çağrı yığını, Tahmini fatura itirazları, Dağıtık üretim başvuruları, Esneklik potansiyeli atıl; uyum tarafında 3 gereklilik (EPDK mevzuatına uygun fatura düzeltme, bağlantı ve bildirim süreçleri dahil) kurulumun parçasıdır. Neden üretime ulaşamıyor 01 ### Kesinti çağrı yığını Aynı fider için yüzlerce arama; kimse bilgilendirilmeden çağrı merkezi tıkanıyor. 02 ### Tahmini fatura itirazları Sayaç okunamayınca tahmin, tahmin yanlışsa itiraz; her dosya elle. 03 ### Dağıtık üretim başvuruları Çatı GES talepleri katlanıyor; mühendisler belge kontrolüne zaman harcıyor. 04 ### Esneklik potansiyeli atıl Talep yanıtı sözleşmeleri var, olay yönetimi elle ve yavaş. Nereden başlanır: ilk 90 gün Üç başlangıç noktası; her biri mevcut sistemlerle kısa bir pilotta ölçülebilir. - 01 ajan senaryosuAbone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda → Fatura ve kesinti çağrıları en büyük hacim; CIS ve OMS bağlantısıyla pilot ölçülebilir. - 02 ajan senaryosuKesinti bildirimi yönetimi → Kesinti dönemlerinde çağrı yükü katlanır; proaktif bilgilendirme ilk fırtınada fark eder. - 03 ajan senaryosuÇatı GES bağlantı başvurusu → Başvuru belge ağırlıklı ve süreli; EPDK takvimine uyum ölçülür. Senaryolar ## Bu sektörde canlı izleyebileceğiniz ajanlar EnerjiHITL ### Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogda Abone arar ya da yazar; yüksek fatura açıklanır, kesinti süresi verilir, sayaç verisi gösterilir, taksit planı kurulur. 520.000 çalıştırma / ay %94,6 başarı 5 ajan EnerjiHITL ### Kesinti bildirimi yönetimi Bildirimler fider bazında kümelenir, tek OSS kaydı açılır, etkilenen herkes ekip yola çıkmadan bilgilenir. 38.000 çalıştırma / ay %97,7 başarı 5 ajan EnerjiHITL ### Sayaç okuma itirazı Müşteri sayaç fotoğrafı yollar; endeks görüntüden okunur, tahmini fatura düzeltilir, büyük iade lidere gider. 9.200 çalıştırma / ay %94,9 başarı 4 ajan EnerjiHITL ### Çatı GES bağlantı başvurusu Başvuru belgeleri okunur, trafo kapasitesi ve mevzuat kontrol edilir, eksikler aynı gün bildirilir; bağlantı görüşü mühendisin. 5.400 çalıştırma / ay %95,3 başarı 5 ajan EnerjiHITL ### Talep yanıtı (demand response) olayı Şebeke sıkışıklığında uygun tüketiciler seçilir, teklif kanaldan gönderilir, katılım ölçülür ve ödeme hesaplanır. 140 çalıştırma / ay %96,2 başarı 4 ajan TelekomHITL ### Fatura itirazı çözümü Fatura kalemleri ve kullanım kayıtları saniyede açıklanır; iyi niyet kredisi politikaya göre uygulanır. 88.000 çalıştırma / ay %96,3 başarı 4 ajan Sektör mimarisi Sektördeki 5 senaryodan türetilmiştir: süpervizörler, uzman ajanlar, Hub model aileleri ve dokunulan sistemler. Süpervizörler Abone Hizmet SüpervizörüKesinti SüpervizörüFatura İtiraz SüpervizörüBağlantı SüpervizörüTalep Yanıtı Süpervizörü Uzman ajanlar Diyalog AjanıFatura AjanıŞebeke AjanıTahsilat AjanıSes AjanıBildirim AjanıKümeleme Ajanıİletişim AjanıGörüntü Ajanıİşlem AjanıBelge AjanıMevzuat AjanıDosya AjanıSeçim AjanıTeklif AjanıUzlaştırma Ajanı Modeller (Hub kataloğu) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-voice-tr-v3aginies-konrul-1 Dokunulan sistemler IVR / SIPBilling (CIS)Outage ManagementSmart Meter DataSMS GatewayCustomer CRM / BillingSCADA / HistorianNetwork OSSWhatsApp Business APIBilling / CIS (SAP IS-U)Meter Data ManagementRegulation Vector Store (EPDK)Connection PortalGIS / Network ModelE-mailSCADA / DMSBilling / Settlement Kanallar voicewhatsappappwebemailapibatch Simülasyon hacimleri 572.740çalıştırma / ay %94,8başarı oranı 6,4 smedyan süre $0,0016tahmini maliyet / çalıştırma Sektör senaryolarının simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Uyum ve egemenlik - EPDK mevzuatına uygun fatura düzeltme, bağlantı ve bildirim süreçleri - Şebeke operasyonunu etkileyen her karar kontrol odası ya da mühendis onayında - SCADA ve OT sistemlerine yalnızca okuma erişimi; kritik altyapı ayrımı korunur Uyum çerçevesi - EPDK tüketici hizmetleri ve bağlantı yönetmelikleri: itiraz, düzeltme ve başvuru süreleri ajan kurallarına kodlanır - Kritik altyapı ve OT ayrımı: SCADA ve OMS’e yalnızca okuma erişimi; şebeke kararı kontrol odasında - KVKK ve sayaç verisi: tüketim verisi yalnızca abonenin kendi hesabı için, saklama süresi politikayla Güvenlik → Önerilen kurulum Müşteri operasyonu için egemen paylaşımlı kurulum; şebekeye yakın okuma ajanları şirketin veri merkezinde on-prem, kritik altyapı ayrımı korunarak. Onay kuralları Sektör senaryolarının insan onayı eşikleri. - Abone asistanı: fatura, kesinti, tüketim ve taksit tek diyalogdaTaksit > 6 ay, fatura iptali veya sayaç şüphesi → müşteri hizmetleri onayı - Kesinti bildirimi yönetimiToplu bilgilendirme > 1.000 müşteri → kontrol odası dispeçeri onayı - Sayaç okuma itirazıİade / düzeltme > 1.000 TL → faturalama ekip lideri onayı - Çatı GES bağlantı başvurusuBağlantı görüşü her zaman şebeke planlama mühendisinin imzasıyla - Talep yanıtı (demand response) olayıOlay > 10 MW → şebeke operatörü onayı; ödeme hesabı denetim iziyle Ölçülen hedefler | Temsilcisiz çözülen çağrı payı | %55–65 | | Fatura itirazı çözüm süresi | Günlerden dakikalara | | Kesinti dönemi çağrı yükü | %50 azalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef pilotun ilk haftasında birlikte sabitlenir. İlgili çözüm paketleri Fatura ve doküman veri çıkarımıOCR → doğrulama → ERP: elle veri girişi olmadan temiz kayıt.Saha servis ve sevk otomasyonuSensör alarmı ya da müşteri bildirimi → doğru teknisyen, doğru parça, doğru zaman.Enerji müşteri operasyonlarıKesinti, sayaç, bağlantı ve talep yanıtı: şebeke verisiyle konuşan müşteri ajanları. ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Finansal Teknolojiler URL: https://www.aginies.com/tr/sektorler/fintech Language: tr Summary: Onboarding, AML, kredi kararı ve uzlaştırma: hız ile uyumun aynı anda gerektiği yerde ajanlar. Finansal Teknolojiler # Onboarding, AML, kredi kararı ve uzlaştırma: hız ile uyumun aynı anda gerektiği yerde ajanlar. Ödeme kuruluşları, elektronik para şirketleri ve BNPL sağlayıcıları bankalarla aynı denetime, ama çok daha hızlı büyüme temposuna tabi. Aginies; üye iş yeri onboarding’i, işlem izleme uyarıları, anlık kredi kararı ve PSP uzlaştırma istisnalarını tek orkestrasyon katmanında çalıştırır; MASAK bildirimi ve limit üstü kararlar insanda kalır. Kısaca Aginies, Finansal Teknolojiler sektöründe 5 canlı ajan senaryosu ve 3 çözüm paketi sunar: Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit, Üye işyeri KYB onboarding, AML işlem izleme alarm triyajı, BNPL anlık taksit kararı, Ödeme mutabakatı istisnaları, Ters ibraz (chargeback) itirazı, Kart dolandırıcılık alarmı triyajı; çözümler: Diyalog bankacılığı, Finansal analiz ve operasyonel zekâ, Üye işyeri ve ödeme operasyonları. Üretime geçişi engelleyen başlıklar Onboarding darboğazı, Uyarı yorgunluğu, Saniyelik kredi kararı, Uzlaştırma istisnaları; uyum tarafında 4 gereklilik (MASAK ve KVKK gereklilikleri; şüpheli işlem bildirimi her zaman analist kararıyla dahil) kurulumun parçasıdır. Neden üretime ulaşamıyor 01 ### Onboarding darboğazı KYB, UBO ve yaptırım kontrolleri günler sürüyor; üye iş yeri rakibe gidiyor. 02 ### Uyarı yorgunluğu İşlem izleme kurallarının %95’i yanlış pozitif; analistler gerçek riski kaçırıyor. 03 ### Saniyelik kredi kararı Ödeme anında karar gerekiyor; bürodan açık bankacılığa kadar veri tek akışta birleşmeli. 04 ### Uzlaştırma istisnaları PSP ekstreleriyle defter arasındaki farklar ay sonunda elle kapatılıyor. Nereden başlanır: ilk 90 gün Üç başlangıç noktası; her biri mevcut sistemlerle kısa bir pilotta ölçülebilir. - 01 ajan senaryosuÜye işyeri operasyon asistanı: POS, hakediş, iade ve limit → İşyeri destek hattı en yoğun kanal; WhatsApp üzerinde pilot haftalar içinde ölçülür. - 02 ajan senaryosuÖdeme mutabakatı istisnaları → Ay sonunu bekleyen istisnalar finans ekibini kilitliyor; otomatik eşleştirme ilk ay fark yaratır. - 03 ajan senaryosuÜye işyeri KYB onboarding → KYB süresi doğrudan gelirle ilişkili; politika motoru risk ekibiyle birlikte kurulur. Senaryolar ## Bu sektörde canlı izleyebileceğiniz ajanlar Finansal TeknolojilerHITL ### Üye işyeri operasyon asistanı: POS, hakediş, iade ve limit İşyeri WhatsApp'tan yazar; POS arızası, hakediş sorgusu, iade ve limit artışı tek diyalogda çözülür. 64.000 çalıştırma / ay %93,8 başarı 5 ajan Finansal TeknolojilerHITL ### Üye işyeri KYB onboarding Sicil, gerçek faydalanıcı ve yaptırım kontrolü tek akışta; düşük riskli işyeri aynı gün ödeme almaya başlar. 1.400 çalıştırma / ay %93,4 başarı 5 ajan Finansal TeknolojilerHITL ### AML işlem izleme alarm triyajı Alarm zenginleştirilir, anlatı taslağı hazır gelir; kararı analist verir, bildirim dakikalarda dosyalanır. 8.500 çalıştırma / ay %95,8 başarı 5 ajan Finansal TeknolojilerHITL ### BNPL anlık taksit kararı Ödemede saniyeler içinde taksit kararı; büro ve açık bankacılık verisi birleşir, limit üstü uzmana gider. 54.000 çalıştırma / ay %97,3 başarı 5 ajan Finansal TeknolojilerHITL ### Ödeme mutabakatı istisnaları PSP mutabakat dosyası ana defterle eşlenir; küçük farklar otomatik düzeltilir, büyükleri finans onayına gider. 620 çalıştırma / ay %95,1 başarı 4 ajan BankacılıkHITL ### Ters ibraz (chargeback) itirazı İtiraz alınır, kanıt paketi toplanır ve kart şemasına dosyalanır; eşik üstü geçici alacak onaya gider. 9.800 çalıştırma / ay %95,2 başarı 4 ajan BankacılıkHITL ### Kart dolandırıcılık alarmı triyajı Kart saniyeler içinde bloklanır, müşteriye WhatsApp'tan sorulur; yanıta göre kart açılır ya da iptal edilir. 65.000 çalıştırma / ay %97,1 başarı 4 ajan Sektör mimarisi Sektördeki 5 senaryodan türetilmiştir: süpervizörler, uzman ajanlar, Hub model aileleri ve dokunulan sistemler. Süpervizörler Üye İşyeri SüpervizörüÜye İşyeri Edinim SüpervizörüAML SüpervizörüKredi Karar SüpervizörüMutabakat Süpervizörü Uzman ajanlar Diyalog AjanıPOS AjanıOperasyon AjanıRisk AjanıKimlik AjanıOrtaklık AjanıSözleşme AjanıZenginleştirme AjanıTarama AjanıAnlatı AjanıDosyalama AjanıKabul AjanıBüro AjanıKarar AjanıKayıt AjanıEşleştirme Ajanıİnceleme AjanıDüzeltme Ajanı Modeller (Hub kataloğu) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-konrul-1 Dokunulan sistemler WhatsApp BusinessTerminal ManagementSettlement LedgerRefund ServiceTicketingMerchant PortalTrade Registry (MERSİS)Sanctions Screeninge-SignEmailCore LedgerCase ManagementMASAK Reporting PortalMerchant Checkout APICredit Bureau (KKB / Findeks)Open Banking APISMS GatewayPSP / Acquirer APICard Network Kanallar whatsappwebvoiceemailapibatchappsms Simülasyon hacimleri 128.520çalıştırma / ay %95,4başarı oranı 7,8 smedyan süre $0,0031tahmini maliyet / çalıştırma Sektör senaryolarının simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Uyum ve egemenlik - MASAK ve KVKK gereklilikleri; şüpheli işlem bildirimi her zaman analist kararıyla - Yüksek riskli üye iş yerleri ve limit üstü kredi kararlarında insan onayı - Her karar için veri kaynağı, model sürümü ve gerekçe kaydı - Veri Türkiye içinde; on-prem ya da dedike bölge Uyum çerçevesi - TCMB Ödeme ve Elektronik Para Kuruluşları Bilgi Sistemleri Tebliği: topluluk bulutu yalnızca uygunluk verilen sağlayıcıda; birincil sistemler yurt içinde - MASAK yükümlülükleri: işlem izleme uyarıları kanıtla analiste, bildirim kararı insanda - Kart şeması kuralları ve KVKK: ters ibraz süreleri takip edilir, kart verisi maskelenir ve amaç dışı kullanılmaz Güvenlik → Önerilen kurulum TCMB uygunluk listesindeki sağlayıcının topluluk bulutunda veya kurumun kendi veri merkezinde; GPU kapasitesi ülke içinde, modeller Hub kataloğundan. Onay kuralları Sektör senaryolarının insan onayı eşikleri. - Üye işyeri operasyon asistanı: POS, hakediş, iade ve limitİade > 25.000 TL veya limit artışı > %50 → risk ekibi onayı - Üye işyeri KYB onboardingRisk sınıfı yüksek veya UBO bulanık eşleşme > 0,65 → uyum yöneticisi onayı - AML işlem izleme alarm triyajıHer alarm → uyum analisti kararı zorunlu; ajan tek başına STR dosyalamaz, vaka kapatmaz - BNPL anlık taksit kararıSepet > 15.000 TL veya büro skoru < 1.200 → kredi uzmanı onayı - Ödeme mutabakatı istisnalarıTek fark > 10.000 TL veya toplam > 50.000 TL → finans müdürü onayı ile ayarlama kaydı Ölçülen hedefler | İşyeri talebi ilk temasta çözüm | %70 ve üzeri | | KYB tamamlanma süresi | Günlerden saatlere | | Mutabakat istisnası kapanış süresi | %60 kısalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef pilotun ilk haftasında birlikte sabitlenir. İlgili çözüm paketleri Diyalog bankacılığıMobil bankacılık işlemlerini yazarak ya da konuşarak gerçekleştir.Finansal analiz ve operasyonel zekâFinansal veriyi ve belgeyi okuyan, yorumlayan, karar destekleyen ajanlar.Üye işyeri ve ödeme operasyonlarıİşyeri edinimi, POS ve hakediş desteği, mutabakat istisnaları ve işlem izleme tek masada. ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Kamu ve Hükümet URL: https://www.aginies.com/tr/sektorler/government Language: tr Summary: Vatandaş başvurusundan ihale dosyasına: mevzuata bağlı, açıklanabilir, kararı memurda bırakan ajanlar. Kamu ve Hükümet # Vatandaş başvurusundan ihale dosyasına: mevzuata bağlı, açıklanabilir, kararı memurda bırakan ajanlar. Kamu kurumlarında her işlem bir mevzuat maddesine, bir süreye ve bir sorumluya bağlıdır. Aginies vatandaş başvurularını yönlendiren, ruhsat ve sosyal yardım dosyalarını ön değerlendiren, ihale tekliflerini şartnameyle karşılaştıran ajanlar kurar. Her çıktı madde numarasıyla gerekçelendirilir; karar her zaman yetkili memur ya da komisyondadır. Kurulum kurumun kendi veri merkezinde, dış bağlantı olmadan yapılabilir. Kısaca Aginies, Kamu ve Hükümet sektöründe 5 canlı ajan senaryosu ve 3 çözüm paketi sunar: Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi, Vatandaş başvurusu yönlendirme, Ruhsat başvurusu ön inceleme, Kamu ihalesi teklif dosyası ön kontrolü, Sosyal yardım başvurusu ön değerlendirme, İhracat kontrolü ve son kullanıcı taraması; çözümler: KVKK kişisel veri koruma, Vatandaş hizmetleri ve dosya ön inceleme, Kamu alımı ve ihale uyumu. Üretime geçişi engelleyen başlıklar Başvuru yönlendirme, Dosya ön inceleme yükü, Açıklanabilirlik, Egemenlik şartı; uyum tarafında 4 gereklilik (Kurumun kendi veri merkezinde, air-gap dahil kurulum; açık kaynak modeller içeride dahil) kurulumun parçasıdır. Neden üretime ulaşamıyor 01 ### Başvuru yönlendirme Dilekçeler yanlış birime gidiyor, süre işliyor, vatandaş tekrar başvuruyor. 02 ### Dosya ön inceleme yükü Ruhsat, yardım ve ihale dosyalarında eksik belge kontrolü memur zamanının çoğunu alıyor. 03 ### Açıklanabilirlik Her karar gerekçe ister; kara kutu araçlar kamuda kullanılamıyor. 04 ### Egemenlik şartı Veri yurt dışına çıkamaz, çoğu bulut asistanı bu yüzden eleniyor. Nereden başlanır: ilk 90 gün Üç başlangıç noktası; her biri mevcut sistemlerle kısa bir pilotta ölçülebilir. - 01 ajan senaryosuVatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi → Başvuru ve belge süreci vatandaşın en çok hissettiği yer; e-Devlet entegrasyonu tekrar belgeyi ortadan kaldırır. - 02 ajan senaryosuVatandaş başvurusu yönlendirme → Dilekçe yönlendirme kurallı ve hacimli; ilk pilot için en düşük riskli senaryo. - 03 çözüm paketiKamu alımı ve ihale uyumu → İhale dosyası ön incelemesi komisyon zamanını açar; karar komisyonda kalır. Senaryolar ## Bu sektörde canlı izleyebileceğiniz ajanlar Kamu ve HükümetHITL ### Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibi Vatandaş yazar ya da arar; uygunluk kontrol edilir, belgeler okunur, başvuru açılır, randevu verilir, her adım izlenebilir. 230.000 çalıştırma / ay %93,8 başarı 5 ajan Kamu ve HükümetHITL ### Vatandaş başvurusu yönlendirme e-Devlet başvurusu saniyeler içinde sınıflandırılır, yasal süresi hesaplanır ve doğru birime SLA ile atanır. 54.000 çalıştırma / ay %98,1 başarı 4 ajan Kamu ve HükümetHITL ### Ruhsat başvurusu ön inceleme Eksik belge ve mevzuat uyumu dakikalar içinde raporlanır; kararı her zaman ruhsat memuru verir. 3.100 çalıştırma / ay %95,8 başarı 4 ajan Kamu ve HükümetHITL ### Kamu ihalesi teklif dosyası ön kontrolü Teklif dosyaları şartnameyle karşılaştırılır, eksik ve uygunsuz belgeler listelenir; değerlendirme kararı komisyonun. 380 çalıştırma / ay %94,8 başarı 4 ajan Kamu ve HükümetHITL ### Sosyal yardım başvurusu ön değerlendirme Başvuru kurum kayıtlarıyla doğrulanır, uygunluk kriterleri puanlanır, eksikler vatandaşa bildirilir; karar sosyal inceleme görevlisinin. 21.000 çalıştırma / ay %96,6 başarı 4 ajan Savunma SanayiiHITL ### İhracat kontrolü ve son kullanıcı taraması Parça, alıcı ve son kullanım tek akışta sınıflandırılır; lisans gereği tespit edilir, karar her zaman uyum yöneticisinde kalır. 2.600 çalıştırma / ay %96,1 başarı 4 ajan Sektör mimarisi Sektördeki 5 senaryodan türetilmiştir: süpervizörler, uzman ajanlar, Hub model aileleri ve dokunulan sistemler. Süpervizörler Vatandaş Hizmetleri SüpervizörüBaşvuru SüpervizörüRuhsat Süpervizörüİhale SüpervizörüSosyal Yardım Süpervizörü Uzman ajanlar Diyalog AjanıUygunluk AjanıBelge AjanıDosya AjanıSınıflandırma AjanıSLA AjanıMevzuat AjanıŞartname AjanıTutanak AjanıDoğrulama AjanıVatandaş İletişim Ajanı Modeller (Hub kataloğu) aginies-talay-2aginies-tulpar-4aginies-konrul-1aginies-sunkar-1 Dokunulan sistemler WhatsApp Businesse-Government GatewayCase ManagementDocument StoreAppointment Systeme-Devlet GatewayLegislation Vector StoreSMS GatewayGIS / Zoning PlanRegulation Vector StoreExchange / M365 Maile-Procurement PlatformSpecification Vector StoreSocial Assistance Information SystemCivil Registry / Income ServicesEligibility Criteria (vector) Kanallar whatsappvoicewebappemailbatch Simülasyon hacimleri 308.480çalıştırma / ay %94,8başarı oranı 7 smedyan süre $0,0021tahmini maliyet / çalıştırma Sektör senaryolarının simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Uyum ve egemenlik - Kurumun kendi veri merkezinde, air-gap dahil kurulum; açık kaynak modeller içeride - Her çıktı mevzuat maddesi ve kaynak belgeyle gerekçelendirilir - İdari karar her zaman yetkili memur ya da komisyonda; ajan yalnızca hazırlar - KVKK ve kamu bilgi güvenliği gerekliliklerine uyumlu erişim ve denetim izi Uyum çerçevesi - Cumhurbaşkanlığı Bilgi ve İletişim Güvenliği Rehberi: kurum veri merkezinde, air-gap dahil kurulum; açık ağırlıklı modeller içeride - İdari usul ve Kamu İhale Kanunu: her çıktı mevzuat maddesi ve kaynak belgeyle; karar yetkili memur veya komisyonda - KVKK ve e-Devlet entegrasyon kuralları: vatandaş verisi amaç sınırlı, erişim yetki bazlı, denetim izi tam Güvenlik → Önerilen kurulum Kurumun kendi veri merkezinde on-prem; gerekiyorsa air-gap. Belediyeler ve bağlı kuruluşlar için kamu bulutunda dedike tenant. Onay kuralları Sektör senaryolarının insan onayı eşikleri. - Vatandaş hizmet asistanı: başvuru, belge, randevu ve durum takibiUygunluk sınırda, belge tutarsız veya aynı hanede başka başvuru → memur incelemesi - Vatandaş başvurusu yönlendirmeSınıflandırma güveni < 0,85 veya personel şikâyeti → evrak kayıt memuru onayı - Ruhsat başvurusu ön incelemeHer ruhsat kararı ruhsat memuru tarafından verilir; sistem yalnızca ön inceleme yapar - Kamu ihalesi teklif dosyası ön kontrolüUygunluk ve eleme kararı her zaman ihale komisyonunda; ajan yalnızca hazırlar - Sosyal yardım başvurusu ön değerlendirmeYardım kararı her zaman sosyal inceleme görevlisinde; ajan ön değerlendirme yapar Ölçülen hedefler | Vatandaştan istenen tekrar belge | %70 azalma | | Başvuru ön inceleme süresi | Günlerden dakikalara | | Doğru birime ilk yönlendirme | %95 ve üzeri | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef pilotun ilk haftasında birlikte sabitlenir. İlgili çözüm paketleri KVKK kişisel veri korumaBelge ve görüşmelerde kişisel veri tespiti, maskeleme ve denetime hazır veri haritası.Vatandaş hizmetleri ve dosya ön incelemeBaşvuru doğru birime, dosya eksiksiz, karar memurda: mevzuat maddesiyle gerekçeli.Kamu alımı ve ihale uyumuTeklif dosyaları şartnameyle karşılaştırılır; komisyon hazır bulgu listesi üzerinde karar verir. ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Sağlık URL: https://www.aginies.com/tr/sektorler/healthcare Language: tr Summary: Klinik karar hekimde kalır; ön yetkilendirme, kabul ve randevu operasyonunu ajanlar taşır. Sağlık # Klinik karar hekimde kalır; ön yetkilendirme, kabul ve randevu operasyonunu ajanlar taşır. Hastane ve sağlık gruplarında zamanın büyük kısmı belge, form ve telefonla geçer: ödeyici ön yetkilendirmesi, hasta ön kabulü, randevu değişiklikleri. Aginies bu operasyonel yükü HBYS, ödeyici portalları ve iletişim kanallarına bağlı ajanlarla üstlenir; klinik nitelikteki her karar hekimde ve hemşirede kalır, hasta verisi kurum sınırını terk etmez. Kısaca Aginies, Sağlık sektöründe 4 canlı ajan senaryosu ve 3 çözüm paketi sunar: Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuç, Ön yetkilendirme (prior authorization), Hasta ön kabul ve triyaj, Randevu planlama ve no-show azaltma, Çağrı merkezi kalite izleme; çözümler: Sesli AI ile hasar ve talep kaydı, Sağlık operasyonları ve klinik yönlendirme, KVKK kişisel veri koruma. Üretime geçişi engelleyen başlıklar Ön yetkilendirme gecikmesi, Kabul ve triyaj yükü, Randevu kayıpları, Veri hassasiyeti; uyum tarafında 3 gereklilik (Özel nitelikli sağlık verisi kurum içinde; modeller on-prem, dış bağlantı yok dahil) kurulumun parçasıdır. Neden üretime ulaşamıyor 01 ### Ön yetkilendirme gecikmesi Ödeyici kuralları dağınık; her dosya için belge toplama ve gerekçe yazma saatler alıyor. 02 ### Kabul ve triyaj yükü Semptom bilgisi telefonda alınıyor, forma geçiriliyor, sonra tekrar soruluyor. 03 ### Randevu kayıpları Gelmeyen hasta oranı yüksek; boşalan slot yeniden dolmuyor. 04 ### Veri hassasiyeti Sağlık verisi özel nitelikli; bulut tabanlı asistanlar uyum ekibinden geçemiyor. Nereden başlanır: ilk 90 gün Üç başlangıç noktası; her biri mevcut sistemlerle kısa bir pilotta ölçülebilir. - 01 ajan senaryosuHasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuç → Randevu ve hazırlık süreci klinik karar içermez; hızlı başlanır, gelmeme oranı hemen ölçülür. - 02 ajan senaryosuÖn yetkilendirme (prior authorization) → Sigorta ön onayı belge ağırlıklı ve kurallı; geri ödeme süresini kısaltır. - 03 ajan senaryosuHasta ön kabul ve triyaj → Kayıt ve ön değerlendirme hemşire onayıyla çalışır; acil yükünü düzenler. Senaryolar ## Bu sektörde canlı izleyebileceğiniz ajanlar SağlıkHITL ### Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuç Hasta yazar ya da arar; randevu, ön hazırlık talimatı, sigorta ön onayı ve sonuç bildirimi tek koordinatörde birleşir. 58.000 çalıştırma / ay %95,1 başarı 5 ajan SağlıkHITL ### Ön yetkilendirme (prior authorization) İstem düşer düşmez ödeyici kuralları taranır, klinik kanıt toplanır, dosya dakikalar içinde gönderime hazır. 6.200 çalıştırma / ay %94,1 başarı 4 ajan SağlıkHITL ### Hasta ön kabul ve triyaj Hasta yazar ya da anlatır; anamnez ve aciliyet skoru saniyeler sürer, son kararı her zaman hemşire verir. 24.500 çalıştırma / ay %95,8 başarı 5 ajan Sağlık ### Randevu planlama ve no-show azaltma Gelmeme riski yüksek hastalar WhatsApp'tan önceden aranır, boşalan slotlar bekleme listesinden anında dolar. 62.000 çalıştırma / ay %98,2 başarı 4 ajan Kurumsal FonksiyonlarHITL ### Çağrı merkezi kalite izleme Örneklem değil, tüm çağrılar rubriğe göre puanlanır; uyum ihlali ve koçluk fırsatı sabah liderin önünde. 90.000 çalıştırma / ay %98,8 başarı 4 ajan Sektör mimarisi Sektördeki 4 senaryodan türetilmiştir: süpervizörler, uzman ajanlar, Hub model aileleri ve dokunulan sistemler. Süpervizörler Hasta Yolculuğu SüpervizörüÖn Yetki SüpervizörüTriyaj SüpervizörüRandevu Süpervizörü Uzman ajanlar Diyalog AjanıRandevu AjanıKlinik Hazırlık AjanıSigorta Ajanıİstem AjanıÖdeyici Kural AjanıGönderim AjanıSes AjanıSemptom AjanıAciliyet AjanıYönlendirme AjanıGelmeme Riski AjanıRezervasyon Ajanı Modeller (Hub kataloğu) aginies-talay-2aginies-sunkar-1aginies-tulpar-4aginies-konrul-1aginies-voice-tr-v3 Dokunulan sistemler WhatsApp BusinessHospital Information SystemSchedulingInsurer PortalLab / ImagingEHR (HL7/FHIR)Payer Policy Vector StorePayer PortalExchange / M365 MailHIS SchedulingTriage Protocol Vector StoreWhatsApp Business APIHospital Information System (HIS)SMS Gateway Kanallar whatsappvoiceappapiemailwebsmsbatch Simülasyon hacimleri 150.700çalıştırma / ay %96,4başarı oranı 7,3 smedyan süre $0,0023tahmini maliyet / çalıştırma Sektör senaryolarının simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Uyum ve egemenlik - Özel nitelikli sağlık verisi kurum içinde; modeller on-prem, dış bağlantı yok - Klinik karar niteliğindeki her adım hekim ya da hemşire onayında - KVKK ve sağlık mevzuatına uygun maskeleme, erişim kaydı ve denetim izi Uyum çerçevesi - Kişisel Sağlık Verileri Yönetmeliği ve KVKK özel nitelikli veri: sağlık verisi hastane sınırında, erişim rol ve amaç bazlı, her erişim kayıtlı - Sağlık Bakanlığı HBYS ve e-Nabız entegrasyon kuralları: ajan HBYS üzerinden işlem yapar, ayrı veri kopyası tutmaz - Klinik karar hekimde: ajan hazırlık, sorgu ve organizasyon yapar; şüphede hekim onay kartı Güvenlik → Önerilen kurulum Hastane veri merkezinde on-prem; modeller kurum içinde, dış bağlantı yok. Hastane grupları için grup veri merkezinde dedike tenant. Onay kuralları Sektör senaryolarının insan onayı eşikleri. - Hasta yolculuğu koordinatörü: randevu, hazırlık, onay ve sonuçİlaç etkileşimi, kontrendikasyon şüphesi veya ek tetkik ihtiyacı → hekim onayı - Ön yetkilendirme (prior authorization)Cerrahi işlem veya maliyet > 25.000 TL → kullanım yönetimi hemşiresi onayı - Hasta ön kabul ve triyajHer vakada aciliyet seviyesini triyaj hemşiresi onaylar; "acil durum" vakaları doğrudan 112'ye yönlendirilir Ölçülen hedefler | Randevuya gelmeme oranı | %30–40 azalma | | Ön onay tamamlanma süresi | Günlerden saatlere | | Çağrı merkezi ortalama işlem süresi | %35 kısalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef pilotun ilk haftasında birlikte sabitlenir. İlgili çözüm paketleri Sesli AI ile hasar ve talep kaydıÇağrı merkezine gelen hasar bildirimini sesli ajan alır, kaydı sistemde açar.Sağlık operasyonları ve klinik yönlendirmeAkıllı triyaj ve hasta yönlendirme; klinik karar her zaman hekimde kalır.KVKK kişisel veri korumaBelge ve görüşmelerde kişisel veri tespiti, maskeleme ve denetime hazır veri haritası. ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Sigorta URL: https://www.aginies.com/tr/sektorler/insurance Language: tr Summary: Hasar bildiriminden tahsilata: her adımı izlenen, onay noktaları belli süreçler. Sigorta # Hasar bildiriminden tahsilata: her adımı izlenen, onay noktaları belli süreçler. Sigortada hacim dalgalı, belgeler dağınık ve her karar düzenleyiciye açıklanabilir olmak zorunda. Aginies, hasar kaydı, poliçe yenileme, underwriting doküman işleme ve rücu gibi süreçleri ortak bir orkestrasyon katmanında çalıştırır; eksper ataması ve ödeme onayı gibi kritik noktalar insanda kalır. Kısaca Aginies, Sigorta sektöründe 6 canlı ajan senaryosu ve 3 çözüm paketi sunar: Uçtan uca kasko hasarı: ihbardan ödemeye, Sesli hasar ihbarı (FNOL), Poliçe yenileme teklifi, Underwriting belge çıkarımı, Fotoğrafla kasko hasar tespiti, Rücu ve tahsilat; çözümler: Sesli AI ile hasar ve talep kaydı, Hasar operasyonları uçtan uca, Underwriting ve yenileme operasyonları. Üretime geçişi engelleyen başlıklar Fırtına günü kapasitesi, Belge kaosu, Açıklanamayan karar, Kaçan rücu; uyum tarafında 4 gereklilik (KVKK ve GDPR uyumlu veri işleme; hassas veri maskeleme dahil) kurulumun parçasıdır. Neden üretime ulaşamıyor 01 ### Fırtına günü kapasitesi Hasar bildirimleri çağrı merkezinde yığılıyor; bekleme süresi memnuniyeti düşürüyor. 02 ### Belge kaosu Underwriting ve hasar dosyaları farklı formatlarda geliyor; eksik belge süreci durduruyor. 03 ### Açıklanamayan karar Model çıktısı var, gerekçesi yok; düzenleyici ve müşteri için savunulamıyor. 04 ### Kaçan rücu Üçüncü taraf sorumluluğu dosyalarda saklı kalıyor; tahsilat fırsatı kaçıyor. Nereden başlanır: ilk 90 gün Üç başlangıç noktası; her biri mevcut sistemlerle kısa bir pilotta ölçülebilir. - 01 ajan senaryosuUçtan uca kasko hasarı: ihbardan ödemeye → Kasko hasarı en yüksek hacimli ve en görünür süreç; fotoğraf ekspertizi eksper kapasitesini hemen açar. - 02 ajan senaryosuSesli hasar ihbarı (FNOL) → Çağrı merkezi ihbarları temsilcisiz alınır; mevcut SIP altyapısıyla pilot kısa sürer. - 03 çözüm paketiUnderwriting ve yenileme operasyonları → Yenileme dönemi takvimle gelir; otomatik teklif yenileme oranını ölçülebilir biçimde etkiler. Senaryolar ## Bu sektörde canlı izleyebileceğiniz ajanlar SigortaHITL ### Uçtan uca kasko hasarı: ihbardan ödemeye Sesli ihbar, fotoğraf ekspertizi, anlaşmalı servis ataması ve ödeme talimatı tek akışta; eksper yalnızca eşik üstünde devreye girer. 21.000 çalıştırma / ay %91,2 başarı 5 ajan SigortaHITL ### Sesli hasar ihbarı (FNOL) Müşteri arar ve anlatır; poliçe plakadan bulunur, teminat kontrol edilir, dosya açılır ve eksper atanır. 34.000 çalıştırma / ay %95,8 başarı 5 ajan SigortaHITL ### Poliçe yenileme teklifi Poliçeler yeniden fiyatlanır, sert artışlar onaya gider; müşteri tek dokunuşla onaylar, poliçe anında kesilir. 22.000 çalıştırma / ay %94,4 başarı 4 ajan SigortaHITL ### Underwriting belge çıkarımı Sağlık raporu ve tahliller okunur, kılavuza göre değerlendirilir; underwriter yalnızca eşik üstünü görür. 4.800 çalıştırma / ay %94,9 başarı 4 ajan SigortaHITL ### Fotoğrafla kasko hasar tespiti Fotoğraflardan hasar tespit edilir, parça/işçilikle tahmin çıkar, eksper onaylar; servis randevusu alınır. 15.500 çalıştırma / ay %93,3 başarı 4 ajan SigortaHITL ### Rücu ve tahsilat Kapanmış dosyalar rücu için taranır, kusurlu tarafın sigortacısı bulunur, talep avukat onayıyla gönderilir. 3.400 çalıştırma / ay %95,5 başarı 4 ajan Sektör mimarisi Sektördeki 6 senaryodan türetilmiştir: süpervizörler, uzman ajanlar, Hub model aileleri ve dokunulan sistemler. Süpervizörler Hasar SüpervizörüYenileme SüpervizörüUnderwriting SüpervizörüRücu Süpervizörü Uzman ajanlar İhbar AjanıFoto Ekspertiz AjanıDosya Ajanıİletişim AjanıKonuşma AjanıPoliçe AjanıHasar Kayıt AjanıFiyatlama AjanıTeklif AjanıTanzim AjanıBelge AjanıTıbbi Değerlendirme AjanıUnderwriting AjanıGörüntü AjanıEkspertiz AjanıServis AjanıTarama AjanıKanıt AjanıTahsilat Ajanı Modeller (Hub kataloğu) aginies-voice-tr-v3aginies-konrul-1aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Dokunulan sistemler Call Centre SIPPolicy AdminClaims SystemRepair NetworkPayment HubGenesys SIPPolicy Admin SystemGuidewire ClaimCenterWhatsApp Business APIRating EngineEmail (Exchange)Underwriting WorkbenchDocument Store (S3)Vector StoreParts & Labour Catalogue (Audatex)Repair Network PortalInsurance Registry (SBM) Kanallar voiceappwhatsappbatchemailwebapi Simülasyon hacimleri 100.700çalıştırma / ay %94,1başarı oranı 7,8 smedyan süre $0,0042tahmini maliyet / çalıştırma Sektör senaryolarının simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Uyum ve egemenlik - KVKK ve GDPR uyumlu veri işleme; hassas veri maskeleme - Ödeme ve red kararlarında zorunlu insan onayı - Her hasar kararı için gerekçeli, zaman damgalı iz - On-prem ya da Türkiye içi dedike bölge seçenekleri Uyum çerçevesi - SEDDK düzenlemeleri ve sigorta bilgi sistemleri gereklilikleri: hasar ve poliçe kararları gerekçeli, zaman damgalı, denetime açık - KVKK özel nitelikli veri: sağlık ve hasar fotoğrafları kurum sınırında işlenir, maskeleme ve erişim kaydı zorunlu - Ödeme ve red kararı her zaman yetkili eksper veya underwriter’da; ajan dosyayı hazırlar Güvenlik → Önerilen kurulum Türkiye içi dedike bölge veya egemen paylaşımlı kurulum; hasar fotoğrafı ve sağlık verisi için görüntü modelleri ülke içi GPU’da. Onay kuralları Sektör senaryolarının insan onayı eşikleri. - Uçtan uca kasko hasarı: ihbardan ödemeyeTahmin > 150.000 TL, kusur oranı belirsiz veya önceki hasarla çakışma → eksper onayı - Sesli hasar ihbarı (FNOL)Yaralanma veya rezerv > 250.000 TL → hasar uzmanına sıcak aktarım - Poliçe yenileme teklifiPrim artışı > %20 veya indirim > %10 → yenileme ekibi lideri onayı - Underwriting belge çıkarımıEk prim ≥ %50 veya teminat > 5.000.000 TL → kıdemli underwriter onayı - Fotoğrafla kasko hasar tespitiTahmin > 15.000 TL veya model güveni < %80 → eksper onayı - Rücu ve tahsilatRücu talebi > 50.000 TL → rücu avukatı onayı Ölçülen hedefler | Basit hasar dosyası kapanış süresi | Haftalardan günlere | | Eksper gerektiren dosya payı | %40 azalma | | Otomatik yenilenen poliçe | %60–75 | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef pilotun ilk haftasında birlikte sabitlenir. İlgili çözüm paketleri Sesli AI ile hasar ve talep kaydıÇağrı merkezine gelen hasar bildirimini sesli ajan alır, kaydı sistemde açar.Hasar operasyonları uçtan ucaİhbardan ödemeye: sesli kayıt, fotoğraf ekspertizi, servis ataması ve rücu tek akışta.Underwriting ve yenileme operasyonlarıTeklif belgeleri okunur, risk değişkenleri çıkarılır, yenileme teklifi kurallar içinde hazırlanır. ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # IoT ve Bağlı Cihazlar URL: https://www.aginies.com/tr/sektorler/iot Language: tr Summary: Binlerce cihazdan gelen sinyali tek olaya, tek karara indiren ajanlar: anomali, OTA, soğuk zincir. IoT ve Bağlı Cihazlar # Binlerce cihazdan gelen sinyali tek olaya, tek karara indiren ajanlar: anomali, OTA, soğuk zincir. Bağlı cihaz filoları büyük hacimde veri üretir; karar yükü hâlâ ekiplerdedir. Anomali alarmları kümelenmeden ekibe düşer, firmware dağıtımı korkuyla ertelenir, sensör sapmasında kimin ne yapacağı belirsizdir. Aginies IoT platformu, zaman serisi veritabanı, cihaz yönetimi ve saha servis sistemlerine bağlı ajanlarla sinyali olaya, olayı karara çevirir; toplu cihaz komutları ve kritik segmentler her zaman insan onayından geçer. Kısaca Aginies, IoT ve Bağlı Cihazlar sektöründe 4 canlı ajan senaryosu ve 3 çözüm paketi sunar: Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyen, Cihaz filosu anomali triyajı, Kademeli OTA güncelleme dağıtımı, Soğuk zincir sapması müdahalesi, Bakım talebi triyajı, Kesinti bildirimi yönetimi; çözümler: Operasyonel otomasyon ve talep yönetimi, Saha servis ve sevk otomasyonu, IoT cihaz filosu operasyonu. Üretime geçişi engelleyen başlıklar Alarm gürültüsü, Riskli OTA dağıtımı, Sensörden aksiyona boşluk, Veri hacmi; uyum tarafında 3 gereklilik (Toplu cihaz komutları ve kritik filo segmentlerinde insan onayı dahil) kurulumun parçasıdır. Neden üretime ulaşamıyor 01 ### Alarm gürültüsü Aynı kök nedenden yüzlerce alarm; ekip gerçek olayı kaçırıyor. 02 ### Riskli OTA dağıtımı Tek seferde tüm filoya güncelleme; geri alma planı yok. 03 ### Sensörden aksiyona boşluk Eşik ihlali görülüyor ama servis emri, müşteri bildirimi ve kalite kararı ayrı ayrı elle yapılıyor. 04 ### Veri hacmi Telemetri hacmi bulut model maliyetini patlatıyor; küçük on-prem modeller gerekiyor. Nereden başlanır: ilk 90 gün Üç başlangıç noktası; her biri mevcut sistemlerle kısa bir pilotta ölçülebilir. - 01 ajan senaryosuSaha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyen → Uzaktan çözülen arıza doğrudan saha maliyetini düşürür; cihaz bulutu bağlantısıyla pilot kısa. - 02 ajan senaryosuCihaz filosu anomali triyajı → Alarm gürültüsü ekipleri yoruyor; kümeleme ilk haftada görünür. - 03 ajan senaryosuKademeli OTA güncelleme dağıtımı → Aşamalı OTA en çok korkulan süreç; durdurma ve geri alma kriterleriyle güven kazanır. Senaryolar ## Bu sektörde canlı izleyebileceğiniz ajanlar IoT ve Bağlı CihazlarHITL ### Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyen Müşteri "cihaz çalışmıyor" der; telemetri okunur, uzaktan tanı konur, güncelleme gönderilir, gerekirse teknisyen doğru parçayla atanır. 37.000 çalıştırma / ay %94,1 başarı 5 ajan IoT ve Bağlı CihazlarHITL ### Cihaz filosu anomali triyajı Binlerce cihazdan gelen telemetri anomalileri kümelenir, kök neden aday listesi çıkar, doğru ekibe tek olay olarak düşer. 18.000 çalıştırma / ay %96,9 başarı 4 ajan IoT ve Bağlı CihazlarHITL ### Kademeli OTA güncelleme dağıtımı Firmware dağıtımı %1 → %10 → %100 kademelerinde ilerler; her kademede sağlık kapısı, kritik filolarda insan onayı. 320 çalıştırma / ay %98,1 başarı 4 ajan IoT ve Bağlı CihazlarHITL ### Soğuk zincir sapması müdahalesi Sıcaklık sapması anında değerlendirilir, ürün riski hesaplanır, saha ekibi sevk edilir, kalite kararı kayıt altına alınır. 2.700 çalıştırma / ay %97,4 başarı 4 ajan Üretim & OtomotivHITL ### Bakım talebi triyajı Sensör alarmı ve teknisyen sesli notu tek iş emrine dönüşür; yedek parça ve güvenli duruş penceresi hazırdır. 3.400 çalıştırma / ay %95,2 başarı 5 ajan EnerjiHITL ### Kesinti bildirimi yönetimi Bildirimler fider bazında kümelenir, tek OSS kaydı açılır, etkilenen herkes ekip yola çıkmadan bilgilenir. 38.000 çalıştırma / ay %97,7 başarı 5 ajan Sektör mimarisi Sektördeki 4 senaryodan türetilmiştir: süpervizörler, uzman ajanlar, Hub model aileleri ve dokunulan sistemler. Süpervizörler Cihaz Destek SüpervizörüFilo Sağlık SüpervizörüDağıtım SüpervizörüSoğuk Zincir Süpervizörü Uzman ajanlar Diyalog AjanıTelemetri AjanıTanı AjanıSaha AjanıSinyal AjanıOperasyon AjanıPlan AjanıSağlık AjanıDağıtım AjanıSensör AjanıKalite Ajanı Modeller (Hub kataloğu) aginies-talay-2aginies-sunkar-1aginies-tulpar-4 Dokunulan sistemler WhatsApp BusinessDevice Cloud (MQTT)Fleet RegistryOTA ServiceField ServiceIoT Platform (MQTT / Digital Twin)Time-series DBKnown Issues (vector)ITSM / IncidentSlackCI/CD (signed artefacts)Device Management / OTAObservability (metrics)IoT PlatformWMS / Lot TrackingStability Rules (vector)WhatsApp Business API Kanallar whatsappwebapibatchapp Simülasyon hacimleri 58.020çalıştırma / ay %95,1başarı oranı 6,9 smedyan süre $0,0025tahmini maliyet / çalıştırma Sektör senaryolarının simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Uyum ve egemenlik - Toplu cihaz komutları ve kritik filo segmentlerinde insan onayı - Sağlık kapısı düşerse otomatik geri alma; her kademe kayıtlı - Telemetri kurum içinde; küçük on-prem modellerle düşük maliyet Uyum çerçevesi - KVKK ve cihaz verisi: telemetri kurum içinde, kullanıcıya ait veri amaç sınırlı - Ürün güvenliği ve siber dayanıklılık gereklilikleri (ihracatta AB CRA dahil): OTA paketleri imzalı, dağıtım kayıtlı, geri alma hazır - Emniyet kritik cihazlar: yalnızca okuma; komut ve toplu işlem ürün sahibi onayında Güvenlik → Önerilen kurulum Hibrit: cihaz bulutu ve zaman serisi verisi mevcut altyapıda, ajanlar ve küçük modeller Türkiye’deki veri merkezimizde veya üreticinin veri merkezinde. Onay kuralları Sektör senaryolarının insan onayı eşikleri. - Saha cihaz destek asistanı: telemetri, uzaktan tanı, OTA ve teknisyenToplu OTA (>100 cihaz) veya garanti dışı cihaz değişimi → ürün sahibi onayı - Cihaz filosu anomali triyajıToplu cihaz komutu > 100 cihaz → filo operasyon lideri onayı - Kademeli OTA güncelleme dağıtımıKritik filo segmentine dağıtım → ürün sahibi onayı; sağlık kapısı düşerse otomatik geri alma - Soğuk zincir sapması müdahalesiÜrün taşıma ya da imha kararı → kalite yöneticisi onayı Ölçülen hedefler | Saha ziyareti gerektiren arıza | %35 azalma | | Alarmdan olaya gürültü | %80 azalma | | OTA geri alma oranı | %1 altı | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef pilotun ilk haftasında birlikte sabitlenir. İlgili çözüm paketleri Operasyonel otomasyon ve talep yönetimiTalepleri sınıflandıran, yönlendiren ve cevaplayan ajanlar; hızlı süreç, düşük maliyet.Saha servis ve sevk otomasyonuSensör alarmı ya da müşteri bildirimi → doğru teknisyen, doğru parça, doğru zaman.IoT cihaz filosu operasyonuAnomali kümeleme, kök neden ve sağlık kapılı OTA dağıtımı: filo tek olay, tek karar. ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Lojistik URL: https://www.aginies.com/tr/sektorler/logistics Language: tr Summary: Sevkiyat istisnaları, gümrük belgeleri ve soğuk zincir: olay olduğu anda karar veren ajanlar. Lojistik # Sevkiyat istisnaları, gümrük belgeleri ve soğuk zincir: olay olduğu anda karar veren ajanlar. Lojistikte değer, istisnanın ne kadar hızlı fark edilip çözüldüğünde saklıdır: geciken araç, eksik gümrük belgesi, sıcaklık sapması. Aginies TMS, WMS, gümrük ve IoT sistemlerine bağlı ajanlarla istisnayı algılar, seçenekleri hesaplar, müşteriyi bilgilendirir; maliyetli kararlar operasyon yöneticisinde kalır. Kısaca Aginies, Lojistik sektöründe 3 canlı ajan senaryosu ve 3 çözüm paketi sunar: Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA, Sevkiyat istisna yönetimi, Gümrük belgesi hazırlama, Soğuk zincir sapması müdahalesi, Siparişim nerede? (WISMO); çözümler: Operasyonel otomasyon ve talep yönetimi, Saha servis ve sevk otomasyonu, Lojistik kontrol kulesi. Üretime geçişi engelleyen başlıklar İstisna geç fark edilir, Gümrük belgesi hataları, Soğuk zincir görünürlüğü, Müşteri iletişimi; uyum tarafında 3 gereklilik (Gümrük beyanı ve yüksek maliyetli rota değişikliklerinde insan onayı dahil) kurulumun parçasıdır. Neden üretime ulaşamıyor 01 ### İstisna geç fark edilir Gecikme müşteri aradığında öğreniliyor; alternatif plan yok. 02 ### Gümrük belgesi hataları Belge hazırlığı elle; bir tarife kodu hatası günlerce bekletiyor. 03 ### Soğuk zincir görünürlüğü Sensör var, karar süreci yok; sapmada kim ne yapacak belirsiz. 04 ### Müşteri iletişimi Aynı bilgi üç kanaldan ayrı ayrı isteniyor; proaktif bildirim yok. Nereden başlanır: ilk 90 gün Üç başlangıç noktası; her biri mevcut sistemlerle kısa bir pilotta ölçülebilir. - 01 ajan senaryosuKontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA → Gecikme yönetimi müşteri memnuniyetinin ve SLA maliyetinin merkezinde; telematik varsa pilot hızlı. - 02 ajan senaryosuSevkiyat istisna yönetimi → İstisna masası günlük hacmi yüksek; TMS entegrasyonu tek başına yeterli. - 03 ajan senaryosuGümrük belgesi hazırlama → Belge hazırlığı kurallı ve zaman baskılı; beyan kararı gümrük müşavirinde kalır. Senaryolar ## Bu sektörde canlı izleyebileceğiniz ajanlar LojistikHITL ### Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLA Telematik sinyali gecikmeyi erken yakalar; alternatif rota ve taşıyıcı seçilir, müşteri bilgilendirilir, SLA tazmini hesaplanır. 96.000 çalıştırma / ay %93,4 başarı 5 ajan LojistikHITL ### Sevkiyat istisna yönetimi Gecikme ya da hasar geldiğinde etki hesaplanır, taşıyıcı yeniden planlanır, müşteri daha sormadan bilgilenir. 12.800 çalıştırma / ay %95,5 başarı 4 ajan LojistikHITL ### Gümrük belgesi hazırlama GTİP kodu, fatura ve çeki listesi dakikalarda hazırlanır; düşük güvenli satırlar gümrük uzmanına gider. 4.300 çalıştırma / ay %94,7 başarı 4 ajan IoT ve Bağlı CihazlarHITL ### Soğuk zincir sapması müdahalesi Sıcaklık sapması anında değerlendirilir, ürün riski hesaplanır, saha ekibi sevk edilir, kalite kararı kayıt altına alınır. 2.700 çalıştırma / ay %97,4 başarı 4 ajan Perakende & E-ticaret ### Siparişim nerede? (WISMO) Müşteri sorar, sipariş ve kargo olayları anında birleşir; gecikmede yeni tarih verilir, kargo araştırılır. 110.000 çalıştırma / ay %98,1 başarı 4 ajan Sektör mimarisi Sektördeki 3 senaryodan türetilmiştir: süpervizörler, uzman ajanlar, Hub model aileleri ve dokunulan sistemler. Süpervizörler Kontrol Kulesi SüpervizörüSevkiyat SüpervizörüGümrük Süpervizörü Uzman ajanlar Tahmin AjanıRota AjanıMüşteri AjanıFinans Ajanıİstisna AjanıEtki AjanıAksiyon AjanıTarife AjanıBelge AjanıMüşavir Ajanı Modeller (Hub kataloğu) aginies-tulpar-4aginies-talay-2aginies-sunkar-1 Dokunulan sistemler TMSTelematicsCarrier APICustomer PortalERP BillingTMS (Transport Management)Salesforce CRMWhatsApp Business APISAP S/4HANAWMSTariff & Regulation Vector StoreCustoms Broker PortalDocument Store (S3) Kanallar apiemailwhatsappbatch Simülasyon hacimleri 113.100çalıştırma / ay %93,7başarı oranı 6,9 smedyan süre $0,0032tahmini maliyet / çalıştırma Sektör senaryolarının simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Uyum ve egemenlik - Gümrük beyanı ve yüksek maliyetli rota değişikliklerinde insan onayı - GDP uyumlu soğuk zincir kaydı ve sapma dokümantasyonu - Müşteri ve sevkiyat verisi kurum içinde; her mesajın kaydı Uyum çerçevesi - Gümrük mevzuatı: beyanname ve belge hazırlığı ajanla, beyan ve imza yetkili müşavirde - GDP ve soğuk zincir kuralları: sıcaklık sapmaları dokümante edilir, karar kalite sorumlusunda - KVKK ve müşteri sözleşmeleri: sevkiyat ve alıcı verisi kurum içinde, her müşteri mesajı kayıtlı Güvenlik → Önerilen kurulum Egemen paylaşımlı kurulum; TMS, telematik ve taşıyıcı API’leri allowlist ile bağlanır, sevkiyat verisi Türkiye’de kalır. Onay kuralları Sektör senaryolarının insan onayı eşikleri. - Kontrol kulesi: gecikme tahmini, rota değişikliği, müşteri bilgilendirme, SLARota maliyet artışı > %15 veya tazmin > 10.000 TL → operasyon müdürü onayı - Sevkiyat istisna yönetimiEk maliyet > 1.000 € → sevk şefi onayı - Gümrük belgesi hazırlamaGTİP güven skoru < 0,90 veya vergi farkı > 2.000 € → gümrük uzmanı onayı Ölçülen hedefler | Gecikme tespit öncülü | Saatler önce | | Zamanında teslimat | +5–10 puan | | “Kargom nerede” çağrıları | %50 azalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef pilotun ilk haftasında birlikte sabitlenir. İlgili çözüm paketleri Operasyonel otomasyon ve talep yönetimiTalepleri sınıflandıran, yönlendiren ve cevaplayan ajanlar; hızlı süreç, düşük maliyet.Saha servis ve sevk otomasyonuSensör alarmı ya da müşteri bildirimi → doğru teknisyen, doğru parça, doğru zaman.Lojistik kontrol kulesiSapmayı erken yakalayan, rotayı değiştiren, müşteriyi bilgilendiren ve SLA’yı hesaplayan operasyon ajanları. ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Üretim ve Otomotiv URL: https://www.aginies.com/tr/sektorler/manufacturing Language: tr Summary: Tedarikçi faturasından bakım talebine: ERP ve MES’in üstünde karar veren, iş emri açan ajanlar. Üretim ve Otomotiv # Tedarikçi faturasından bakım talebine: ERP ve MES’in üstünde karar veren, iş emri açan ajanlar. Üretim tesislerinde süreçler ERP, MES, bakım ve kalite sistemleri arasında bölünmüştür; her geçiş bir e-posta, bir form, bir bekleme demektir. Aginies bu sistemlere bağlı ajanlarla üçlü eşleştirmeyi, bakım triyajını ve garanti taleplerini uçtan uca yürütür; tolerans dışı her karar ilgili mühendis ya da yöneticiye düşer. Kısaca Aginies, Üretim ve Otomotiv sektöründe 4 canlı ajan senaryosu ve 3 çözüm paketi sunar: Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna, Tedarikçi faturası üçlü eşleştirme, Bakım talebi triyajı, Garanti talebi işleme, Sevkiyat istisna yönetimi, Cihaz filosu anomali triyajı; çözümler: Fatura ve doküman veri çıkarımı, Ajan kiralama (Agent Outsourcing), Üretim ve bakım operasyonları. Üretime geçişi engelleyen başlıklar Sistemler arası kopukluk, Bakım talebi yığını, Fatura ve mal kabul farkları, Garanti maliyeti; uyum tarafında 3 gereklilik (OT / IT ayrımı korunur; ajanlar üretim ağına yalnızca okuma ile bağlanır dahil) kurulumun parçasıdır. Neden üretime ulaşamıyor 01 ### Sistemler arası kopukluk ERP, MES ve bakım sistemi aynı olayı üç kez, üç formatta görüyor. 02 ### Bakım talebi yığını Talepler öncelik verilmeden sıraya giriyor; kritik arıza sıradan taleple aynı kuyrukta. 03 ### Fatura ve mal kabul farkları Üçlü eşleştirme elle yapılıyor; toleranslar kişiye göre değişiyor. 04 ### Garanti maliyeti Talepler geç ve tutarsız değerlendiriliyor; bayi ve müşteri bekliyor. Nereden başlanır: ilk 90 gün Üç başlangıç noktası; her biri mevcut sistemlerle kısa bir pilotta ölçülebilir. - 01 ajan senaryosuHat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna → Plansız duruş en pahalı olay; MES ve historian bağlantısıyla ilk hatta pilot. - 02 ajan senaryosuBakım talebi triyajı → Bakım talepleri hacimli ve kurallı; CMMS entegrasyonu hızlı sonuç verir. - 03 ajan senaryosuTedarikçi faturası üçlü eşleştirme → Finans tarafında ölçülebilir kazanç; üretim ağına dokunmadan başlar. Senaryolar ## Bu sektörde canlı izleyebileceğiniz ajanlar Üretim & OtomotivHITL ### Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporuna MES alarmı düşer; kök neden tarihsel veriyle bulunur, yedek parça kontrol edilir, iş emri ve plan güncellemesi dakikalar içinde hazırdır. 3.100 çalıştırma / ay %92,2 başarı 5 ajan Üretim & OtomotivHITL ### Tedarikçi faturası üçlü eşleştirme Fatura, sipariş ve mal kabul SAP'de otomatik eşleşir; yalnızca tolerans dışı kalemler satın almacıya gider. 21.000 çalıştırma / ay %97,1 başarı 4 ajan Üretim & OtomotivHITL ### Bakım talebi triyajı Sensör alarmı ve teknisyen sesli notu tek iş emrine dönüşür; yedek parça ve güvenli duruş penceresi hazırdır. 3.400 çalıştırma / ay %95,2 başarı 5 ajan Üretim & OtomotivHITL ### Garanti talebi işleme Bayi talebi dakikalarda kapsam, parça ve işçilik yönünden doğrulanır; eşik üstü tutar bölge müdürüne gider. 8.600 çalıştırma / ay %96,3 başarı 4 ajan LojistikHITL ### Sevkiyat istisna yönetimi Gecikme ya da hasar geldiğinde etki hesaplanır, taşıyıcı yeniden planlanır, müşteri daha sormadan bilgilenir. 12.800 çalıştırma / ay %95,5 başarı 4 ajan IoT ve Bağlı CihazlarHITL ### Cihaz filosu anomali triyajı Binlerce cihazdan gelen telemetri anomalileri kümelenir, kök neden aday listesi çıkar, doğru ekibe tek olay olarak düşer. 18.000 çalıştırma / ay %96,9 başarı 4 ajan Sektör mimarisi Sektördeki 4 senaryodan türetilmiştir: süpervizörler, uzman ajanlar, Hub model aileleri ve dokunulan sistemler. Süpervizörler Üretim SüpervizörüFatura SüpervizörüBakım SüpervizörüGaranti Süpervizörü Uzman ajanlar Tanı AjanıParça AjanıPlan AjanıRapor AjanıYakalama AjanıEşleştirme AjanıKayıt AjanıSinyal AjanıSaha Notu AjanıTeşhis AjanıPlanlama AjanıTalep AjanıDoğrulama AjanıÖdeme Ajanı Modeller (Hub kataloğu) aginies-tulpar-4aginies-sunkar-1aginies-talay-2aginies-konrul-1aginies-voice-tr-v3 Dokunulan sistemler MESSCADA HistorianCMMSERP Spare PartsShift Channel (Teams)SAP S/4HANAExchange / M365 MailTolerance Policy Vector StoreSCADA / HistorianCMMS (Maximo)SAP S/4HANA MMManuals & Work Order Vector StoreDealer PortalWarranty Policy Vector StoreDocument Store (S3) Kanallar apiappemailbatchvoiceweb Simülasyon hacimleri 36.100çalıştırma / ay %96,3başarı oranı 7,1 smedyan süre $0,005tahmini maliyet / çalıştırma Sektör senaryolarının simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Uyum ve egemenlik - OT / IT ayrımı korunur; ajanlar üretim ağına yalnızca okuma ile bağlanır - Tolerans dışı fatura, garanti ve bakım kararlarında insan onayı - Tedarikçi ve müşteri verisi kurum içinde; on-prem model seçeneği Uyum çerçevesi - OT güvenliği (IEC 62443 yaklaşımı): ajanlar üretim ağına yalnızca okuma ile bağlanır, komut vermez; OT/IT ayrımı korunur - Ürün güvenliği ve garanti mevzuatı: garanti kararları kurallı, tolerans dışı dosya insana - KVKK ve tedarikçi sözleşmeleri: tedarikçi ve müşteri verisi kurum içinde, paylaşım sözleşme kapsamıyla sınırlı Güvenlik → Önerilen kurulum Hibrit: üretim ağına yakın okuma ajanları ve historian erişimi tesiste, karar ve doküman ajanları Türkiye’deki veri merkezimizde veya kurumun veri merkezinde. Onay kuralları Sektör senaryolarının insan onayı eşikleri. - Hat duruşu müdahalesi: alarmdan iş emrine ve vardiya raporunaPlan değişikliği > 2 saat veya 50.000 TL üzeri acil satın alma → vardiya amiri onayı - Tedarikçi faturası üçlü eşleştirmeFiyat sapması > %2 veya > 1.000 TL → satın alma sorumlusu onayı - Bakım talebi triyajıPlanlı duruş > 4 saat → vardiya amiri onayı - Garanti talebi işlemeGaranti talebi > 5.000 € → bölge garanti müdürü onayı Ölçülen hedefler | Plansız duruş süresi | %20–30 azalma | | Alarmdan iş emrine süre | Dakikalar | | Dokunmadan işlenen tedarikçi faturası | %70–85 | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef pilotun ilk haftasında birlikte sabitlenir. İlgili çözüm paketleri Fatura ve doküman veri çıkarımıOCR → doğrulama → ERP: elle veri girişi olmadan temiz kayıt.Ajan kiralama (Agent Outsourcing)Dış kaynak insan yerine dış kaynak ajan: iOS geliştirici, iş analisti, muhasebeci, testçi; iş kaydıyla.Üretim ve bakım operasyonlarıHat duruşundan iş emrine, bakım talebinden garanti dosyasına: MES, CMMS ve ERP üzerinde karar veren ajanlar. ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Perakende & E-ticaret URL: https://www.aginies.com/tr/sektorler/retail Language: tr Summary: Her müşteri sorusunun sonunda bir işlem bekler; ajanlar o işlemi sipariş sisteminde yapar. Perakende & E-ticaret # Her müşteri sorusunun sonunda bir işlem bekler; ajanlar o işlemi sipariş sisteminde yapar. Sezon zirvelerinde destek hacmi katlanıyor; “siparişim nerede”, iade, teslimat değişikliği ve ürün soruları ekipleri boğuyor. Aginies, katalog ve politika bilgisine bağlı ajanlarla soruyu cevaplar, sipariş sisteminde işlemi yapar; katalog zenginleştirme ve yorum yönetimi gibi arka plan işlerini de aynı katmanda çalıştırır. Kısaca Aginies, Perakende & E-ticaret sektöründe 6 canlı ajan senaryosu ve 3 çözüm paketi sunar: Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin, Siparişim nerede? (WISMO), İade ve para iadesi, Ürün soru-cevap asistanı, Katalog zenginleştirme (toplu), Yorum yanıtlama ve kalite eskalasyonu; çözümler: E-ticaret müşteri desteği ve aksiyon, E-posta sınıflandırma ve ek doğrulama, Çağrı merkezi kalite yönetimi. Üretime geçişi engelleyen başlıklar Sezon zirvesi, Cevap veren, işlem yapmayan bot, Katalog borcu, Yorum ve şikâyet gürültüsü; uyum tarafında 3 gereklilik (KVKK/GDPR uyumlu müşteri verisi işleme dahil) kurulumun parçasıdır. Neden üretime ulaşamıyor 01 ### Sezon zirvesi Kasım–Aralık hacmi yılın kalanının üç katı; ekip esnemiyor. 02 ### Cevap veren, işlem yapmayan bot Müşteri iptal istiyor, bot “müşteri hizmetlerine yönlendiriyorum” diyor. 03 ### Katalog borcu Tedarikçi belgelerinden ürün özelliklerini elle girmek haftalar alıyor. 04 ### Yorum ve şikâyet gürültüsü Binlerce yorum arasında ürün kusuru sinyali kayboluyor. Nereden başlanır: ilk 90 gün Üç başlangıç noktası; her biri mevcut sistemlerle kısa bir pilotta ölçülebilir. - 01 ajan senaryosuKombin asistanı: beden, stok, bütçe ve stile göre tam kombin → Gelir tarafında doğrudan etki: sepet büyüklüğü ve iade oranı aynı anda ölçülür. - 02 ajan senaryosuSiparişim nerede? (WISMO) → “Kargom nerede” çağrıları destek hacminin en büyük kalemi; sipariş sistemine bağlı ajan ilk hafta sonuç verir. - 03 ajan senaryosuİade ve para iadesi → İade politikası netse otomasyon hızlıdır; istisnalar insana kalır. Senaryolar ## Bu sektörde canlı izleyebileceğiniz ajanlar Perakende & E-ticaretHITL ### Kombin asistanı: beden, stok, bütçe ve stile göre tam kombin Müşteri bir fotoğraf ve bir cümle gönderir; ajan stokta olan, bedenine uyan ve bütçeye sığan üç kombin kurar, seçileni sepete ekler. 142.000 çalıştırma / ay %96,1 başarı 5 ajan Perakende & E-ticaret ### Siparişim nerede? (WISMO) Müşteri sorar, sipariş ve kargo olayları anında birleşir; gecikmede yeni tarih verilir, kargo araştırılır. 110.000 çalıştırma / ay %98,1 başarı 4 ajan Perakende & E-ticaretHITL ### İade ve para iadesi Politika ve müşteri geçmişi saniyede kontrol edilir, etiket üretilir; iade tutarı kargo taramasıyla ödenir. 48.000 çalıştırma / ay %96,7 başarı 4 ajan Perakende & E-ticaret ### Ürün soru-cevap asistanı Beden, uyumluluk ve özellik soruları katalog ve iade verisiyle yanıtlanır; stok ve fiyat doğrulanır. 120.000 çalıştırma / ay %98,7 başarı 4 ajan Perakende & E-ticaretHITL ### Katalog zenginleştirme (toplu) Tedarikçi PDF ve görsellerinden öznitelik çıkarılır, TR/EN içerik yazılır; yüzlerce SKU dakikalarda hazır. 900 çalıştırma / ay %95,8 başarı 4 ajan Perakende & E-ticaretHITL ### Yorum yanıtlama ve kalite eskalasyonu Yorumlar sınıflanır, marka tonunda yanıtlanır; tekrarlayan kusurlar iade verisiyle QA'e Jira kaydı olur. 26.000 çalıştırma / ay %97,2 başarı 4 ajan Sektör mimarisi Sektördeki 6 senaryodan türetilmiştir: süpervizörler, uzman ajanlar, Hub model aileleri ve dokunulan sistemler. Süpervizörler Stil SüpervizörüSipariş Süpervizörüİade SüpervizörüAlışveriş SüpervizörüKatalog SüpervizörüMüşteri Sesi Süpervizörü Uzman ajanlar Görsel AjanıStilist AjanıKatalog AjanıSepet AjanıNiyet AjanıTakip AjanıYanıt AjanıKabul AjanıPolitika Ajanıİşlem AjanıBelge AjanıNormalizasyon Ajanıİçerik AjanıSınıflandırma AjanıKalite Ajanı Modeller (Hub kataloğu) aginies-konrul-1aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Dokunulan sistemler WhatsApp BusinessProduct Catalogue (PIM)InventoryCustomer ProfileCart & Checkout APIShopify / OMSCarrier API (Yurtiçi, DHL)WhatsApp Business APIZendeskPayment Gateway (iyzico)Email (Exchange)Vector StorePIM / CatalogueWeb Chat WidgetDocument Store (S3)SAP ERPReview Platform APIJira (QA) Kanallar appwebwhatsappvoiceemailbatchapi Simülasyon hacimleri 446.900çalıştırma / ay %97,4başarı oranı 5,9 smedyan süre $0,001tahmini maliyet / çalıştırma Sektör senaryolarının simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Uyum ve egemenlik - KVKK/GDPR uyumlu müşteri verisi işleme - Politika dışı iade ve kampanya istisnalarında insan onayı - Sipariş sistemine yazılan her aksiyon için iz Uyum çerçevesi - Tüketici mevzuatı ve mesafeli satış kuralları: cayma, iade ve teslimat süreleri ajan kurallarına kodlanır - KVKK ve pazarlama izni: temas yalnızca izinli kanalda, öneriler amaç sınırlı veriyle - Kampanya ve fiyat kuralları: indirim yetkisi matrisle sınırlı, sınır dışı teklif müşteri hizmetlerine Güvenlik → Önerilen kurulum Egemen paylaşımlı kurulum; sezon tepe yüklerinde GPU kapasitesi kullanım bazlı ölçeklenir, müşteri verisi Türkiye’de kalır. Onay kuralları Sektör senaryolarının insan onayı eşikleri. - Kombin asistanı: beden, stok, bütçe ve stile göre tam kombinKampanya kuralı dışı indirim veya kişiye özel fiyat talebi → müşteri hizmetleri - İade ve para iadesiİade tutarı > 2.500 TL veya 12 ayda > 5 iade → ekip lideri onayı - Katalog zenginleştirme (toplu)Yeni kategori veya tedarikçi fiyatı sapması > %15 → kategori yöneticisi onayı - Yorum yanıtlama ve kalite eskalasyonu1 yıldızlı yorum yanıtları ve kusur eskalasyonları → topluluk yöneticisi onayı Ölçülen hedefler | İlk temasta çözüm oranı | +15–25 puan | | Sezon tepe ek temsilci ihtiyacı | %40–60 azalma | | Kombin önerisiyle sepet büyüklüğü | +10–20% | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef pilotun ilk haftasında birlikte sabitlenir. İlgili çözüm paketleri E-ticaret müşteri desteği ve aksiyonSipariş nerede, iade, teslimat değişikliği: ajan işlemi sipariş sisteminde tamamlar.E-posta sınıflandırma ve ek doğrulamaGelen kutusu doğru ekibe, eksik ek daha iş başlamadan tespit edilir.Çağrı merkezi kalite yönetimiGörüşmelerin %100’ü objektif kriterlerle puanlanır, örnekleme biter. ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Yazılım Geliştirme URL: https://www.aginies.com/tr/sektorler/software Language: tr Summary: PM, analist, mimar, geliştirici ve test ajanları: özellik talebinden üretime giden hattın tamamı. Yazılım Geliştirme # PM, analist, mimar, geliştirici ve test ajanları: özellik talebinden üretime giden hattın tamamı. Yazılım ekiplerinde darboğaz çoğunlukla hattın geri kalanındadır: gereksinim netleştirme, tasarım kararı, inceleme, sürüm, olay müdahalesi. Aginies bu hattı Jira, Git ve CI/CD’ye bağlı bir SDLC ajan takımıyla çalıştırır; Code Aginies IDE içinde kurumun kendi modelleriyle tamamlama sağlar. Üretim kararı ve mimari onay her zaman insanda kalır. Kısaca Aginies, Yazılım Geliştirme sektöründe 8 canlı ajan senaryosu ve 3 çözüm paketi sunar: Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR, Özellik teslimat hattı, CI/CD sürüm hattı, Olay müdahalesi ve acil düzeltme, Kod incelemesi ve güvenlik kapısı, Backlog düzenleme ve tahminleme, Kademeli OTA güncelleme dağıtımı; çözümler: SDLC ajan ekibi, IoT cihaz filosu operasyonu, Ajan kiralama (Agent Outsourcing). Üretime geçişi engelleyen başlıklar Gereksinim belirsizliği, İnceleme darboğazı, Sürüm riski, Kod sızıntısı endişesi; uyum tarafında 3 gereklilik (Kod ve tamamlama modelleri kurum içinde; Code Aginies on-prem dahil) kurulumun parçasıdır. Neden üretime ulaşamıyor 01 ### Gereksinim belirsizliği Talep, hikâyeye dönüşene kadar üç toplantı; kabul kriterleri sonradan yazılıyor. 02 ### İnceleme darboğazı PR’lar günlerce bekliyor; güvenlik taraması ayrı bir kuyruk. 03 ### Sürüm riski Sürüm notu, kapsam ve geri alma planı elle; her sürüm bir gece mesaisi. 04 ### Kod sızıntısı endişesi Bulut kod asistanları kurumun kaynak kodunu dışarı taşıyor. Nereden başlanır: ilk 90 gün Üç başlangıç noktası; her biri mevcut sistemlerle kısa bir pilotta ölçülebilir. - 01 ajan senaryosuÖzellik teslimat hattı → Gereksinimden PR’a hat en görünür darboğaz; tek takımla pilot 4–6 hafta. - 02 ajan senaryosuKod incelemesi ve güvenlik kapısı → Her PR’da güvenlik ve kalite kapısı düşük riskli, hemen ölçülür. - 03 ajan senaryosuLegacy modül modernizasyonu: analiz, test, yeniden yazım, PR → Modernizasyon programları yıllarca sürer; karakterizasyon testleriyle güvenli adımlarla gidilir. Senaryolar ## Bu sektörde canlı izleyebileceğiniz ajanlar Yazılım GeliştirmeHITL ### Legacy modül modernizasyonu: analiz, test, yeniden yazım, PR Jira'daki bir epic'ten başlar; eski modül okunur, davranışı testlerle sabitlenir, yeniden yazılır ve kanıtlı bir PR açılır. 1.900 çalıştırma / ay %90,4 başarı 5 ajan Yazılım GeliştirmeHITL ### Özellik teslimat hattı Bir Jira epic'i ajanlarla kriter, tasarım, kod ve teste dönüşür; insan yalnızca PR'ı inceler ve onaylar. 1.200 çalıştırma / ay %91,2 başarı 5 ajan Yazılım GeliştirmeHITL ### CI/CD sürüm hattı Derleme, test, tarama ve staging otomatik; üretime yalnızca sürüm yöneticisi onayıyla ArgoCD ile çıkılır. 950 çalıştırma / ay %94,7 başarı 5 ajan Yazılım GeliştirmeHITL ### Olay müdahalesi ve acil düzeltme Alarm gelir, hata yeniden üretilir, yama ve regresyon hazır; nöbetçi onaylar, post-mortem taslağı bekler. 320 çalıştırma / ay %90,5 başarı 5 ajan Yazılım GeliştirmeHITL ### Kod incelemesi ve güvenlik kapısı Her PR ajanlarca incelenir, taranır, kapsamı ölçülür; geliştirici düzeltir, insan yalnızca temiz PR'ı onaylar. 4.200 çalıştırma / ay %93,6 başarı 5 ajan Yazılım GeliştirmeHITL ### Backlog düzenleme ve tahminleme Talepler triyajlanır, netleştirilir, karmaşıklığı tahminlenir; sprint önerisi ürün sahibinin onayına gider. 340 çalıştırma / ay %92,8 başarı 4 ajan IoT ve Bağlı CihazlarHITL ### Kademeli OTA güncelleme dağıtımı Firmware dağıtımı %1 → %10 → %100 kademelerinde ilerler; her kademede sağlık kapısı, kritik filolarda insan onayı. 320 çalıştırma / ay %98,1 başarı 4 ajan Sektör mimarisi Sektördeki 8 senaryodan türetilmiştir: süpervizörler, uzman ajanlar, Hub model aileleri ve dokunulan sistemler. Süpervizörler Modernizasyon SüpervizörüTeslimat SüpervizörüSürüm SüpervizörüOlay Süpervizörüİnceleme SüpervizörüPlanlama SüpervizörüiOS Geliştirici SüpervizörüTest Süpervizörü Uzman ajanlar Analiz AjanıTest AjanıGeliştirici Ajanıİnceleme AjanıAnalist AjanıMimar AjanıDerleme AjanıGüvenlik AjanıSürüm AjanıPost-mortem Ajanıİnceleyici AjanıProje Yöneticisi AjanıGeliştirme AjanıTest Plan AjanıKoşum AjanıHata Ajanı Modeller (Hub kataloğu) aginies-tulpar-4aginies-sunkar-1aginies-talay-2 Dokunulan sistemler JiraGit (self-hosted)CI PipelineStatic AnalysisGitHubCI Runner (GitHub Actions)ConfluenceSonarQube + SnykKubernetes / ArgoCDSlackObservability (Grafana / Datadog)SonarQubeDependency Scanner (Snyk)Slack / TeamsXcode CloudFigmaWorkforce Work LogCI/CDDevice / Browser GridJira / Xray Kanallar apibatchwebapp Simülasyon hacimleri 9.550çalıştırma / ay %92,7başarı oranı 15,8 smedyan süre $0,0148tahmini maliyet / çalıştırma Sektör senaryolarının simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Uyum ve egemenlik - Kod ve tamamlama modelleri kurum içinde; Code Aginies on-prem - Üretim dağıtımı ve mimari kararlarda insan onayı - SAST, bağımlılık ve gizli anahtar taraması her PR’da; kanıt sürümle saklanır Uyum çerçevesi - Kaynak kod ve fikri mülkiyet: kod kurum dışına çıkmaz; Code Aginies ve tamamlama modelleri on-prem - Güvenli yazılım geliştirme gereklilikleri (regüle sektörlerde BDDK/SEDDK beklentileri dahil): SAST, bağımlılık ve sır taraması her PR’da, kanıt sürümle saklanır - Üretim dağıtımı ve mimari kararlar insanda; ajan yalnızca atanan depo ve dallarda yazar Güvenlik → Önerilen kurulum Code Aginies ve kodlama modelleri kurumun veri merkezinde veya kapalı devre kurulumla Türkiye’deki veri merkezimizde; SDLC ajanları Jira, Git ve CI/CD’ye servis hesaplarıyla bağlı. Onay kuralları Sektör senaryolarının insan onayı eşikleri. - Legacy modül modernizasyonu: analiz, test, yeniden yazım, PRÜretim şeması değişikliği veya kapsam dışı dosya → mimar onayı; PR birleştirme her zaman insanda - Özellik teslimat hattıHer PR → en az 1 insan kod incelemesi onayı; ajan main'e doğrudan yazamaz - CI/CD sürüm hattıHer üretim dağıtımı → sürüm yöneticisi onayı zorunlu; kritik CVE varsa kapı otomatik kapanır - Olay müdahalesi ve acil düzeltmeHer acil üretim dağıtımı → nöbetçi mühendis onayı; yama PR'ı da insan incelemesinden geçer - Kod incelemesi ve güvenlik kapısıHer PR → CODEOWNERS insan onayı; yüksek SAST bulgusu veya kapsam < %80 kapıyı kapatır - Backlog düzenleme ve tahminlemeSprint kapsamı → ürün sahibi onayı; ajan sprint'e talep atayamaz, yalnızca önerir - iOS özellik geliştirme (kiralık geliştirici ajanı)Her PR bir insan tarafından incelenir; üretim dağıtımı ekip liderinin onayında - Regresyon test döngüsü (kiralık test ajanı)Sürüm “go / no-go” kararı sürüm yöneticisinde Ölçülen hedefler | İlk incelemede kabul edilen PR | %60–75 | | Gereksinimden PR’a süre | %40–50 kısalma | | Test kapsamı | +15–25 puan | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef pilotun ilk haftasında birlikte sabitlenir. İlgili çözüm paketleri SDLC ajan ekibiPM, analist, mimar, geliştirici ve test ajanları CI/CD hattında birlikte çalışır.IoT cihaz filosu operasyonuAnomali kümeleme, kök neden ve sağlık kapılı OTA dağıtımı: filo tek olay, tek karar.Ajan kiralama (Agent Outsourcing)Dış kaynak insan yerine dış kaynak ajan: iOS geliştirici, iş analisti, muhasebeci, testçi; iş kaydıyla. ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası --- # Telekom URL: https://www.aginies.com/tr/sektorler/telecom Language: tr Summary: Milyonlarca abone, binlerce senaryo: fatura, arıza, dolandırıcılık ve teklif tek katmanda. Telekom # Milyonlarca abone, binlerce senaryo: fatura, arıza, dolandırıcılık ve teklif tek katmanda. Telekom operasyonu hacim ve çeşitlilik demek: fatura itirazı, SIM değişikliği, arıza bildirimi, paket yükseltme. Aginies, faturalama, provizyon ve ağ sistemlerine bağlı ajanlarla bu senaryoları uçtan uca çalıştırır; dolandırıcılık riski taşıyan işlemler risk skoruna göre insana düşer. Kısaca Aginies, Telekom sektöründe 5 canlı ajan senaryosu ve 3 çözüm paketi sunar: Abone hizmet asistanı: fatura, paket, arıza ve taahhüt, Fatura itirazı çözümü, SIM değişimi dolandırıcılık kontrolü, Şebeke kesintisi iletişimi, Tarife yükseltme danışmanı; çözümler: Çağrı merkezi kalite yönetimi, Ses analizi ve kalite, Abone hizmetleri ve şebeke desteği. Üretime geçişi engelleyen başlıklar Fatura itirazı hacmi, SIM değişikliği dolandırıcılığı, Arıza iletişimi, Teklif zamanlaması; uyum tarafında 3 gereklilik (Abone verisi Türkiye içinde; kişisel veri maskeleme dahil) kurulumun parçasıdır. Neden üretime ulaşamıyor 01 ### Fatura itirazı hacmi Her fatura döneminde çağrı merkezi tıkanıyor; itirazların çoğu birkaç kalıba giriyor. 02 ### SIM değişikliği dolandırıcılığı Hız ile güvenlik arasında sıkışan bir süreç; her iki yönde de maliyet var. 03 ### Arıza iletişimi Etkilenen aboneler bilgilendirilmeden çağrı merkezini arıyor. 04 ### Teklif zamanlaması Kullanım verisi var, doğru anda doğru teklif yok. Nereden başlanır: ilk 90 gün Üç başlangıç noktası; her biri mevcut sistemlerle kısa bir pilotta ölçülebilir. - 01 ajan senaryosuAbone hizmet asistanı: fatura, paket, arıza ve taahhüt → Çok konulu çağrılar en pahalı çağrılar; tek oturumda çözüm ortalama işlem süresini doğrudan düşürür. - 02 ajan senaryosuFatura itirazı çözümü → Fatura itirazları hacimli ve kurallı; BSS entegrasyonu ilk pilot için yeterli. - 03 ajan senaryosuSIM değişimi dolandırıcılık kontrolü → Dolandırıcılık riski yüksek, kural seti net; insan onayı korunarak hızlı devreye alınır. Senaryolar ## Bu sektörde canlı izleyebileceğiniz ajanlar TelekomHITL ### Abone hizmet asistanı: fatura, paket, arıza ve taahhüt Abone arar ya da yazar; fatura itirazı, paket değişikliği ve hat arızası aynı oturumda BSS ve şebeke sistemlerinde çözülür. 410.000 çalıştırma / ay %94,3 başarı 5 ajan TelekomHITL ### Fatura itirazı çözümü Fatura kalemleri ve kullanım kayıtları saniyede açıklanır; iyi niyet kredisi politikaya göre uygulanır. 88.000 çalıştırma / ay %96,3 başarı 4 ajan TelekomHITL ### SIM değişimi dolandırıcılık kontrolü Her SIM değişimi provizyondan önce risk skorlanır; şüpheli talepler bekletilir ve bankalara sinyal iletilir. 41.000 çalıştırma / ay %97,7 başarı 4 ajan TelekomHITL ### Şebeke kesintisi iletişimi Kesinti kaydı açılınca etkilenen aboneler bulunur, mesaj yazılıp onaylanır; müşteri aramadan bilgilendirilir. 1.400 çalıştırma / ay %98,5 başarı 4 ajan TelekomHITL ### Tarife yükseltme danışmanı Kullanım analiz edilir, doğru tarife karşılaştırmalı sunulur; onayla değişiklik anında provizyonlanır. 52.000 çalıştırma / ay %95,8 başarı 4 ajan Sektör mimarisi Sektördeki 5 senaryodan türetilmiştir: süpervizörler, uzman ajanlar, Hub model aileleri ve dokunulan sistemler. Süpervizörler Abone Hizmet SüpervizörüFatura SüpervizörüSIM Güvenlik SüpervizörüKesinti İletişim SüpervizörüTarife Süpervizörü Uzman ajanlar Diyalog AjanıHesap AjanıŞebeke AjanıTaahhüt AjanıNiyet AjanıFatura Analiz AjanıDüzeltme AjanıRisk AjanıDoğrulama AjanıProvizyon AjanıTespit AjanıEtki Ajanıİletişim AjanıKullanım AjanıTeklif Ajanı Modeller (Hub kataloğu) aginies-talay-2aginies-sunkar-1aginies-tulpar-4 Dokunulan sistemler IVR / SIPBSS / BillingOSS NetworkCRMSMS GatewayBilling (Amdocs)Usage Mediation / CDR StoreSalesforce CRMHLR / ProvisioningBank SIM-Swap Signal APIServiceNowNetwork OSSProduct Catalogue / Offer EngineWhatsApp Business API Kanallar voiceappwhatsappbranchapismsbatch Simülasyon hacimleri 592.400çalıştırma / ay %95,0başarı oranı 7,3 smedyan süre $0,0016tahmini maliyet / çalıştırma Sektör senaryolarının simülasyon tanımlarından toplanmıştır; kurum ölçeğine göre keşif oturumunda yeniden hesaplanır. Uyum ve egemenlik - Abone verisi Türkiye içinde; kişisel veri maskeleme - SIM değişikliği ve yüksek tutarlı düzeltmelerde insan onayı - Faturalama ve provizyon sistemlerine yazılan her işlem için iz Uyum çerçevesi - BTK düzenlemeleri ve Elektronik Haberleşme Kanunu: abonelik, taahhüt ve numara taşıma işlemleri kayıtlı ve geri izlenebilir - KVKK ve trafik verisi: abone verisi Türkiye içinde, ses kayıtları ve transkriptler maskelenir - SIM değişimi ve kimlik işlemleri: ek doğrulama ve insan onayı; şüpheli talep dolandırıcılık ekibine Güvenlik → Önerilen kurulum Egemen paylaşımlı kurulum veya operatörün kendi veri merkezinde dedike tenant; ses modelleri ülke içi GPU’da, SIP ile mevcut santrale bağlı. Onay kuralları Sektör senaryolarının insan onayı eşikleri. - Abone hizmet asistanı: fatura, paket, arıza ve taahhütTaahhüt iptali, 500 TL üzeri fatura düzeltmesi veya teknisyen önceliği → temsilci onayı - Fatura itirazı çözümüİyi niyet kredisi > 75 TL → ekip lideri onayı - SIM değişimi dolandırıcılık kontrolüRisk skoru ≥ 70 veya 72 saat içinde kimlik bilgisi değişikliği → dolandırıcılık masası onayı - Şebeke kesintisi iletişimiEtkilenen abone > 10.000 veya P1 kesinti → şebeke operasyon yöneticisi onayı - Tarife yükseltme danışmanıTaahhütlü veya indirimli özel teklif → kampanya yöneticisi onayı Ölçülen hedefler | Temsilcisiz çözülen talep payı | %60–70 | | Gereksiz teknisyen ziyareti | %35–45 azalma | | Ortalama işlem süresi | %40 kısalma | Benzer kurulumlarda hedeflenen aralıklar; başlangıç değeri ve hedef pilotun ilk haftasında birlikte sabitlenir. İlgili çözüm paketleri Çağrı merkezi kalite yönetimiGörüşmelerin %100’ü objektif kriterlerle puanlanır, örnekleme biter.Ses analizi ve kaliteTürkçe STT ile görüşmelerin %100’ü yazıya dökülür, puanlanır; modeller kurum içinde.Abone hizmetleri ve şebeke desteğiFatura, paket, taahhüt ve arıza işlemleri sesli ve yazılı kanalda, BSS ve şebeke sistemlerinde. ## Yapay zekâyı denemekten, yapay zekâyla dönüşüme geçme zamanı. 30 dakikalık keşif oturumunda önceliklendirdiğiniz 2–3 iş problemini ele alalım, benzer bir senaryonun canlı demosunu görün, Değer katmanı analiziyle başlayan bir yol haritası taslağı çıkaralım. 30 dakikalık keşif oturumu → dahi@aginies.com - Öncelikli 2–3 iş problemi - Benzer senaryonun canlı demosu - Değer analiziyle başlayan yol haritası