How alisoliman/insurance-multi-agent works
Project Type and Purpose
• This is a full-stack, multi-agent AI claims demo.
• Frontend: Next.js 15 (React 19) with shadcn/ui for interactive UIs and live agent dashboards.
• Backend: FastAPI (Python 3.12) orchestrating a LangGraph multi-agent workflow backed by Azure OpenAI (GPT-4.1) and FAISS vector search.
• Purpose: Accept insurance claims (text + PDFs + images), run them through specialized AI agents (Claim Assessor, Policy Checker, Risk Analyst, Communication Agent) under a Supervisor Agent, and present an explainable recommendation to the user.
File Structure
• Top-level directories:
– backend/: FastAPI app, endpoints (app/api/v1/endpoints), core config, models, services, workflow (agents, tools, supervisor).
– frontend/: Next.js app router (app/agents, app/documents, demo), global components, UI library (components/ui), API clients (lib).
– infra/: Azure Bicep templates for Container App environment, registry, roles.
– Dockerfiles for both frontend and backend, and azure.yaml for CI/CD.
• Architectural patterns:
– MVC-style separation in backend (endpoints ← services ← models).
– Supervisor pattern in LangGraph for orchestrating parallel agents.
– Infrastructure as code (IaC) with Bicep.
Extract Key Insights from README
• Multi-Agent Workflow: Supervisor dispatches Claim Assessor, Policy Checker, Risk Analyst in parallel; optionally triggers Communication Agent; returns synthesized summary.
• AI Providers: Azure OpenAI for LLM and embeddings; FAISS for local vector DB (planned Azure AI Search).
• Deployment: Azure Container Apps hosting two containers (frontend & backend) behind managed identities, ACR, Log Analytics, HTTPS endpoints.
Diagramming Conventions
– Blue rectangles for UI and frontend;
– Green for backend services and orchestration;
– Orange/Yellow for AI calls and vector DB;
Detailed Capture of Architectural Elements
• Explicitly show all five agents and PDF/policy tools.
• Vector store separately from raw policy files.
• Build pipeline (Dockerfiles + azure.yaml + Bicep) in a side note or as a CI/CD component.