A proof-of-concept showing how specialized AI agents can work together to process insurance claims — from damage assessment to fraud detection to customer communication.

Five agents collaborate through a supervisor pattern built on LangGraph: a Claim Assessor analyzes damage photos and validates repair costs, a Policy Checker searches policy documents using semantic similarity (supports English and Dutch), a Risk Analyst detects fraud patterns and scores risk, a Communication Agent drafts personalized emails and document requests, and a Supervisor orchestrates the whole workflow.

The interesting part is watching agents reason in real-time. Every decision is explainable — you can trace exactly why a claim was flagged or approved. Processing drops from hours to minutes while keeping a human in the loop for final decisions.

Built with FastAPI, Next.js 15, Azure OpenAI (GPT-4.1 with vision), and FAISS for vector search. Runs on Azure Container Apps.