This is a high-stakes executive proof-of-concept designed to demonstrate the power of Generative AI in the insurance industry. The application synthesizes structured system data with unstructured policy documents to provide instant, transparent, and accurate coverage determinations.
- Dynamic File Discovery: No hardcoded file paths in the database. The system discovers PDS and Policy Schedules directly from the claim's directory.
- Executive Demo Mode: A caching toggle for zero-latency presentations.
- Single-Column UI: Optimized for a clean, top-to-bottom reading experience on mobile or desktop.
- AI Analysis Pipeline: A 4-step reasoning breakdown (Extraction, PDS Analysis, Limits Verification, Synthesis).
/
├── claims_data.csv # System of Record (Structured Data)
├── data/
│ └── {Claim_ID}/ # Claim-specific directory
│ ├── pds/ # Product Disclosure Statement (PDF)
│ ├── policy_schedule/# Customer's specific coverage (PDF)
│ └── claim_documents/# Additional supporting evidence
├── src/
│ ├── app.py # Streamlit Frontend
│ ├── data_extraction.py # Structured data retrieval
│ ├── pds_analysis.py # LLM analysis of PDS
│ ├── limits_verification.py # LLM verification of schedule limits
│ └── synthesis.py # Final decision logic
├── tests/ # Automated test suite
└── cache/ # Local storage for pre-approved results
- Python 3.10+
- uv (recommended)
- A Google Gemini API Key
The project uses uv for high-performance dependency management.
# Install dependencies and create a virtual environment
uv syncCreate a .env file in the root directory and add your configuration:
GOOGLE_API_KEY=your_api_key_here
GEMINI_MODEL=gemini-2.0-flash # Or your preferred modelTo start the Streamlit application using uv:
uv run streamlit run src/app.pyThe application will be available in your browser at http://localhost:8501.
To run the test suite:
uv run pytest