EyHacks AI Claims Processing System

A comprehensive, AI-driven insurance claims processing architecture featuring real-time conversational voice agents, automated document verification via multi-agent AI crews, and dynamic frontend real-time tracking.

🚀 Architecture Overview

This project consists of several microservices orchestrated via Docker Compose:

1. Frontend (React/Vite)
   • Agent dashboard for tracking claims, listening to live AI conversation transcripts, and viewing AI document analysis.
2. Backend Server (Node.js/Express)
   • REST API handling authentication (JWT), MongoDB schemas for Claims, and file uploads.
   • Pushes document processing tasks to RabbitMQ.
3. Voice AI Engine (Python/LangGraph)
   • Uses Twilio Media Streams to stream raw phone call audio via WebSockets.
   • Uses Deepgram for ultra-low latency real-time transcription (STT).
   • Powered by a LangGraph AI agent equipped with external tools (MongoDB querying, meeting scheduling, etc.).
   • Emits real-time transcripts and AI states back to the Node backend via Socket.IO.
4. Document Verification Worker (Python/CrewAI)
   • A dedicated Python worker consuming the document_processing RabbitMQ queue.
   • Utilizes a CrewAI multi-agent setup (Fraud Analyst & Underwriter) to thoroughly analyze uploaded claim documents, verify authenticity, and push detailed analytical reports back into MongoDB.
5. Auto-Tunnel Daemon (Python)
   • Automatically provisions temporary cloudflared edge tunnels for local development.
   • Dynamically updates Twilio Phone Number webhooks on the fly.

🛠 Features

• Real-time Conversational Voice AI: Call the Twilio number and converse with an AI agent capable of scheduling tasks and querying claim databases in real-time.
• Live Dashboard Syncing: As the voice AI talks to the client on the phone, the conversation transcript is dynamically streamed to the React dashboard in real-time using Socket.IO.
• Automated Document Underwriting: Any document uploaded to a claim is securely processed by autonomous CrewAI agents in the background, and their full analysis is seamlessly integrated into the Voice AI's knowledge base.
• Intelligent Network Routing: The auto_tunnel.py daemon ensures that Twilio can seamlessly hit your local services without you ever needing to configure manual NAT/Firewall port forwarding.

⚙️ Project Setup

1. Prerequisites

• Docker & Docker Compose
• Node.js (v18+)
• Python 3.12+
• Twilio Account (Phone Number, Account SID, Auth Token)
• Deepgram API Key
• LLM API Keys (Groq, OpenAI, etc.)

2. Environment Variables

Create a .env file in the root directory (where docker-compose.yml is).

# Twilio
TWILIO_ACCOUNT_SID=your_sid
TWILIO_AUTH_TOKEN=your_token
TWILIO_PHONE_NUMBER=+1234567890
YOUR_PHONE_NUMBER=+0987654321

# Deepgram
DEEPGRAM_API_KEY=your_key

# LLMs
GROQ_API_KEY=your_groq_key
OPENAI_API_KEY=your_openai_key

# Database / Auth
JWT_SECRET=super_secret_key
SMTP_HOST=smtp.gmail.com
SMTP_PORT=587
SMTP_USER=your_email
SMTP_PASS=your_app_password

3. Start the Backend Infrastructure

The system uses Docker Compose to orchestrate MongoDB, RabbitMQ, Redis, Node, and the Python services.

docker compose up -d

4. Enable External Voice Traffic (Twilio Webhooks)

Because Twilio needs to hit your machine from the public internet to deliver phone calls, run the auto-tunneling script. This script will automatically spin up Cloudflare edge tunnels and configure your Twilio account for you:

# Optional: Setup virtual environment
python3.12 -m venv .venv
source .venv/bin/activate
pip install -r backend/python/requirements.txt

# Run the tunnel daemon
python3 auto_tunnel.py

Note: Leave this script running in a separate terminal window.

5. Run the Frontend

To boot the React Dashboard for the human agents:

npm install
npm run dev

Navigate to http://localhost:5173 to access the dashboard.

🚀 Triggering a Test Call

Ensure all services are running and the auto_tunnel.py script has successfully updated your webhooks. Then simply run:

python3 trigger_call.py

Your phone will ring in a few seconds. Answer it, and you'll be speaking directly to the AI agent!