ClaimsIQ — Insurance Claims Triage Agent Demo

An AI agent that processes insurance claims through a multi-step workflow. The agent starts as a "confused rookie" and becomes a "seasoned expert" as Hindsight memories accumulate.

Watch the agent learn coverage rules, adjuster assignments, and escalation patterns in real-time through a pipeline dashboard.

Quick Start​

1. Start Hindsight API (port 8888)​

docker run -p 8888:8888 ghcr.io/anthropics/hindsight:latest

2. Start Backend (port 8000)​

cd backend
pip install -r requirements.txt
./run.sh

3. Start Frontend (port 5173)​

cd frontend
npm install
npm run dev

4. Open Browser​

Navigate to http://localhost:5173.

How It Works​

The agent processes insurance claims using 6 tools:

1. Classify the claim category (auto, property, flood, etc.)
2. Look up the policy details
3. Check coverage rules for the policy type
4. Check fraud indicators
5. Assign the right adjuster
6. Submit a decision for validation

The system validates each decision against ground truth. If the agent makes a mistake (wrong adjuster, incorrect coverage call), the decision is rejected with feedback — creating learning signal for Hindsight.

Agent Modes​

Key Learning Challenges​

• Water damage vs Flood: Gold policies cover water damage (burst pipe) but NOT flood damage (rain/river). The agent must learn this subtle distinction.
• Adjuster routing: 8 adjusters with different specialties and regions. The agent must learn who handles what.
• Escalation thresholds: Claims over $50K need a senior adjuster; over $100K need manager review.
• Fraud detection: Multiple indicators (near-limit claims, repeated address) route to the fraud specialist.

Environment Variables​