Insurance Triage Agent

An AI-powered insurance claim processing system that automatically classifies, extracts information, scores, and detects fraud in insurance claims using LangGraph and IBM Watsonx.

🚀 Features

• Web Interface: Modern Streamlit-based dashboard for easy interaction
• Multi-type Classification: Automatically classifies claims as auto, medical, or home insurance
• Advanced Information Extraction: Extracts key details including:
  • Insured name and policy ID
  • Claim value and type
  • Document summaries and metadata
• Intelligent Priority Scoring: Scores claims 1-1000 based on:
  • Catastrophic nature (40% weight)
  • Deadline proximity (30% weight)
  • Claim value (20% weight)
  • Document completeness (10% weight)
• Fraud Detection: AI-powered fraud risk assessment with:
  • Risk scoring (0-100%)
  • Anomaly detection
  • Historical pattern analysis
• Database Integration: SQLite database for claim storage and retrieval
• Document Processing: Supports both text and PDF documents

🏗️ Project Structure

Insurance_Priage/
├── streamlit_app.py       # Main web application
├── main.py                # CLI interface
├── workflow.py            # LangGraph workflow orchestration
├── agents.py              # AI agents (classification, extraction, scoring, fraud detection)
├── models.py              # Pydantic data models
├── database.py            # Database operations
├── config.py              # Configuration settings
├── import_past_claims.py  # Utility for importing historical claims
├── requirements.txt       # Python dependencies
├── .env.example          # Environment variables template
└── test_input/           # Sample claim documents
    ├── medical_receipt.txt
    ├── profile.txt
    └── referral.txt

🛠️ Setup

1. Create virtual environment:

   python -m venv venv
   venv\Scripts\activate  # Windows
   source venv/bin/activate  # Linux/Mac

2. Install dependencies:

   pip install -r requirements.txt

3. Configure environment:

   cp .env.example .env
   # Edit .env and add your Watsonx API credentials

🚦 Usage

Web Interface (Recommended)

streamlit run streamlit_app.py

Command Line Interface

# Process sample documents
python main.py

# Process specific documents
python main.py document1.txt document2.pdf

# View stored claims
python main.py --view-claims

🤖 AI Workflow

1. Document Ingestion: Load and process .txt or .pdf files
2. Classification: AI determines claim type (auto/medical/home)
3. Information Extraction: AI extracts structured data from documents
4. Priority Scoring: Claims scored 1-1000 based on multiple factors
5. Fraud Detection: AI analyzes for potential fraud indicators
6. Database Storage: Processed claims stored for review and analysis

🎯 Priority Scoring

• Catastrophic (800-1000): Life-threatening, total loss, fatalities
• High (600-799): Significant damage, urgent medical needs
• Medium (300-599): Moderate damage, non-emergency medical
• Low (1-299): Minor damage, routine claims

🕵️ Fraud Detection

• Risk Scoring: 0-100% fraud likelihood
• Key Indicators:
  • Unusual claim patterns
  • Document inconsistencies
  • Historical claim analysis
  • Value anomalies
• AI Reasoning: Detailed explanation of fraud risk factors

🌐 Web Interface Features

• Document Upload: Drag-and-drop support for multiple files
• Interactive Dashboard: Real-time claim processing status
• Visual Analytics: Charts and graphs for claim analysis
• Search & Filter: Easy navigation through processed claims
• Responsive Design: Works on desktop and tablet

📊 Database Schema

The SQLite database (insurance_claims.db) includes tables for:

• Claims (ID, insured info, claim details)
• Documents (content, metadata)
• Processing results (scores, timestamps)
• Fraud detection analysis

⚙️ Requirements

• Python 3.10+
• IBM Watsonx API credentials
• Streamlit
• LangGraph and LangChain
• SQLite3

📝 Example Output

🔍 Classifying insurance claim type...
📋 Extracting claim information...
⚖️ Scoring claim priority...
🕵️ Analyzing for potential fraud...
💾 Saving claim to database...
✅ Claim processed successfully!

📊 CLAIM PROCESSING RESULTS
==================================================
🆔 Claim ID: 42
👤 Insured: John Smith
📋 Policy ID: MED-2024-123456
🏷️ Insurance Type: medical
💰 Claim Value: $8,500.00
⭐ Priority Score: 675/1000
🛡️ Fraud Risk: 15% (Low Risk)
📅 Processed: 2024-08-14 15:30:00