Claims FNOL assistant on Azure

A motor-insurance claims agent on Azure, built on the same harness as the support agent.

Active · join meAI EngineeringForward-deployed ready

7 of 15 tasks done

1 · Purpose

This is a motor insurer's first-notice-of-loss (FNOL) assistant: the moment a policyholder first says "a bus hit my car". It registers the claim, gives a reference, answers status questions and pays small payouts; anything larger, or anyone hurt, goes to a person. It runs on Azure and reuses the support agent's harness library, which supplies about 84% of the code it runs. It proves the harness carries to a second domain and a second cloud, and it is the running example in the agent harness course.

For: Insurance and financial-services technology teams, Azure architects, course participants.

2 · Business benefits

Small claims settled fasterClaims are registered at once and payouts up to the limit are paid without waiting for a handler.
Payouts stay within authorityAny payout above the limit waits for a handler's approval, enforced where the payment executes.
Injuries always reach a personA claim with someone hurt is handed to a handler, never decided by the agent.
Lower build costAbout 84% of the code is reused from the support agent's library, unchanged.
Runs on free tiersThe whole Azure deployment fits a free trial with a budget alert set first.

3 · Status and how to join

Active · join me. Under way: pick a task below, or tell me what you'd like to learn and I'll suggest one. Lead: Basant Choudhary.

Good first tasks this week:

  • Run the agent on your laptop and report what was unclear
  • Draft one exercise for the guided Azure tour
  • Add a scenario for withdrawing a claim

4 · Requirements

What it must do

How well it must do it

5 · Architecture, very high level

Chat page and handler des…Where policyholders chat andhandlers approve or take over.Sign-inProves who the policyholder orhandler is.AgentThe harness, the claims rulesand the model client.Model gatewayOne door to the model, with anallow-list, token limits andClaims systemRegisters claims and payoutsover MCP, checking the caller'sDurable waitsHold a payout or handoff untila person decides.State storeThe agent's conversations,approvals and the claims data.Telemetry and alertsTraces every turn and raisesalerts on the agent's signals.sign inmodel call
Components and what each is responsible for, with no tool names: the tools come in the next section.

Also: Chat page and handler desk → Agent: message · Agent → Claims system: tool call with token · Agent → Durable waits: payout over limit · Durable waits → Chat page and handler desk: approval request · Agent → State store: turn saved · Agent → Telemetry and alerts: trace

6 · Tech stack, per platform

ComponentOpen stackAzureAWS
Chat page and handler deskown web pages ✓Azure Container Apps ✓—
Sign-inlocal test sign-in ✓Microsoft Entra ID ✓—
AgentPython harness library, Pydantic AI ✓Azure Container Apps, App Configuration, Key Vault ✓not built
Model gatewayGroq called directly ✓API Management in front of Groq, Content Safety ✓—
Claims systemown MCP server ✓Azure Container Apps, on-behalf-of tokens ✓—
Durable waitsDBOS ✓DBOS on Azure Database for PostgreSQL ✓—
State storePostgreSQL ✓Azure Database for PostgreSQL with pgvector ✓—
Telemetry and alertslocal trace log ✓Application Insights, Log Analytics, log alerts ✓—

Bold with ✓ = built and running today. Everything else is the planned or possible binding: the components stay the same, only the tools change.

7 · What needs to be done

Phase 1: on the laptop 4 of 4 done

TaskStatusSizeSkill
Specification, simulated world and 29 scenariosdoneMSpecification design
All four gates pass with a scripted modeldoneMAI testing
Chat with it locally with the real modeldoneMPython
Checks placed by YAML as plug-insdoneMPython

Phase 1: in Azure 3 of 5 done

TaskStatusSizeSkill
All Azure resources created from Bicep and azddoneLAzure infrastructure
Security fixes for money and identity pathsdoneMSecurity
Live sign-in, on-behalf-of tokens and email alertsdoneMAzure
Report a claim through the gateway and see it in Application InsightsdoingMAzure
Deploy from GitHub Actions with a federated logintodoSCI/CD

Evaluation loop and guided tour 0 of 3 done

TaskStatusSizeSkill
Turn the watch rules into online checks placed by YAMLtodoMPython
A job that reads Application Insights and writes findingstodoMAzure, KQL
Write the guided tour of every Azure resource, with exercisestodoMTechnical writing

Phase 2 0 of 3 done

TaskStatusSizeSkill
Label about 500 motor-claims messages for a local routertodoMData labelling
Answer policy-wording questions from retrieved clauses, with citationstodoLRAG
Measure cost per conversation and per outcome from real tracestodoMCost analysis

Size: S = an evening or two, M = about a week part-time, L = a few weeks.

8 · Roles

RoleHow manyHours / weekWhat they do
Azure engineer13-5 hours/weekOwns the Bicep, the deploy pipeline and the Azure checks.
Agent engineer13-5 hours/weekBuilds claims features and fixes findings, spec first.
Insurance domain reviewer11-2 hours/weekChecks the claims rules and wording against real practice.

9 · What you'll learn

Level-up questions

10 · Links