GIC Insurance Analytics RAG

data used is public data

Overview

This system provides analytics over General Insurance Company (GIC) premium data (FY24 & FY25, Apr-Oct). It uses Retrieval-Augmented Generation (RAG) to deliver grounded, hallucination free insights.

Key Features

• Zero Hallucinations: All answers grounded in actual data with citations
• Advanced Analytics: Growth metrics, volatility analysis, risk classifications
• Persistent Storage: ChromaDB vector database with disk persistence

Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Configure Groq API

Edit .env file and add your Groq API key:

GROQ_API_KEY=your_actual_key_here

Get your free API key at: https://console.groq.com/keys

3. Generate Knowledge Base

cd Rag
python document_generator.py

This creates 49 semantic documents from your insurance data:

• 34 company summaries
• 9 segment overviews
• 4 risk classifications
• 1 industry overview
• 1 growth insights document

4. Launch the App

streamlit run app.py

The app will open at http://localhost:8501

Sample Queries

Try asking:

• "Which insurers have risky growth?"
• "What is the total industry premium in FY25?"
• "Compare health and motor segments"
• "Which companies are exposed to crop insurance risk?"
• "Show me companies with high YoY growth"
• "What are the key trends in FY25?"

Project Structure

rag star/
├── app.py                      # Streamlit web interface
├── requirements.txt            # Python dependencies
├── .env                        # API keys (gitignored)
│
├── Rag/                        # Core RAG system
│   ├── __init__.py            # Package initialization
│   ├── analytics.py           # Data analysis logic
│   ├── document_generator.py  # Semantic document creation
│   └── rag.py                 # RAG engine (Groq + ChromaDB)
│
├── data/
│   ├── processed/
│   │   ├── gic_ytd_master_apr_oct.csv      # Main dataset
│   │   ├── rag_knowledge_base.csv           # Generated documents
│   │   └── rag_metadata.json               # Document metadata
│   ├── cleaned/               # Monthly Excel files
│   └── raw/                   # Original data
│
├── analysis/
│   └── gic_ytd_premium_analysis_report.ipynb
│
└── chroma_db/                 # Vector database (auto-created)

Architecture

1. Analytics Layer (analytics.py)

• Computes growth metrics at industry/segment/company levels
• Calculates volatility and risk classifications
• Portfolio concentration analysis
• Company ranking by growth/stability

2. Document Generator (document_generator.py)

• Transforms analytics into semantic text chunks
• Creates 5 document types optimized for retrieval
• Saves to CSV with metadata

3. RAG Engine (rag.py)

• Embeddings: sentence-transformers/all-MiniLM-L6-v2
• Vector DB: ChromaDB with persistent storage
• LLM: Groq Llama 3.3 70B (ultra-fast inference)
• Retrieval: Top-k semantic search
• Generation: Grounded answers with citations

4. Web Interface (app.py)

• Streamlit chat interface
• Real-time query processing
• Knowledge base management
• System status monitoring

System Requirements

• Python 3.8+
• 8GB RAM minimum
• Internet connection (for Groq API)
• macOS, Linux, or Windows

Usage Guide

Basic Workflow

1. Ask a Question: Type your query in the chat input
2. RAG Retrieval: System finds top-4 relevant documents
3. Groq Generation: Llama 3.3 generates grounded answer
4. Citations: All facts include document IDs

Knowledge Base Management

• Click " Rebuild Knowledge Base" in sidebar to regenerate documents
• Useful after updating the source CSV data
• Automatically re-ingests into vector database

API Key Management

If Groq API key is missing:

• System falls back to template-based responses
• Still functional but less sophisticated
• Add key to .env and restart app

Testing

Test Document Generation

cd Rag
python document_generator.py

Expected output: Saved 49 documents to data/processed

Test RAG Engine

cd Rag
python rag.py

Runs 5 test queries and displays answers.

Test Analytics

cd Rag
python analytics.py

Troubleshooting

"Knowledge base not found"

Run: cd Rag && python document_generator.py

"GROQ_API_KEY not set"

Edit .env file and add your API key. Restart the app.

Dependency conflicts

Some packages may show version conflicts. These are non-critical and can be ignored unless functionality is broken.

ChromaDB errors

Delete chroma_db/ folder and restart app to regenerate the vector database.

Data Coverage

Time Period: FY24 & FY25 (April-October)

Segments:

• Health Insurance
• Motor (Total)
• Miscellaneous (incl. Crop)
• Fire & Property
• Personal Accident
• Engineering
• Liability
• Marine
• Aviation

Companies: 34 insurance companies tracked

Metrics:

• Premium volumes (YTD)
• YoY growth rates
• Monthly premium trends
• Portfolio concentration
• Risk classifications

Key Insights Generated

1. Company Summaries: Premium, top segment, volatility, risk notes
2. Segment Analysis: Market share, YoY growth, characteristics, risks
3. Risk Classifications: Crop risk, health strategy types, concentration
4. Industry Overview: Total premium, trends, strategic insights
5. Growth Patterns: Momentum analysis, quality indicators

Security

• API keys stored in .env (gitignored)
• No data sent outside Groq API
• All data processing happens locally
• ChromaDB runs locally (no cloud)

Performance

• Query Response: 1-3 seconds
• Knowledge Base Gen: ~5 seconds for 49 documents
• Vector Ingestion: ~10 seconds for full corpus
• Embedding Model: Runs on CPU, very efficient

License

This project uses:

• Groq API (subject to Groq's terms)
• ChromaDB (Apache 2.0)
• Streamlit (Apache 2.0)
• Sentence Transformers (Apache 2.0)

For issues or questions:

1. Check the Troubleshooting section
2. Verify Groq API key is valid
3. Ensure all dependencies are installed
4. Review data paths in code

---