🏦 Ai Bank Policy Comparator

Discover. Compare. Secure. Your Perfect Home Loan

Born from my parents' struggle with home loan documents - an AI-powered platform built with React, TypeScript, and Python that uses NLP and document parsing to simplify complex policy comparisons for everyday people.

🎯 Overview

The AI Bank Policy Comparator empowers customers and analysts to:

• 📤 Upload Documents: Ingest 2-4 bank home loan MITC documents (PDF/TXT/DOCX)
• 🤖 AI Analysis: Extract key terms using Google Gemini LLM with confidence scoring
• ⚖️ Smart Comparison: Side-by-side analysis with human-readable explanations
• 🔍 Evidence Tracing: Click any cell to see normalized value, explanation, and source evidence
• 📊 Export Results: Generate CSV/JSON reports for decision making
• 🎨 Butter Money Branding: Beautiful blue gradient theme matching Butter Money's website

🎯 User Workflow

For Customers:

1. Upload 2-4 home loan MITC documents from different banks
2. Wait for AI processing (30-60 seconds per document)
3. Compare policies side-by-side with clear explanations
4. Click any cell to understand differences and see evidence
5. Export comparison report for decision making

For Analysts:

1. Batch process multiple customer document sets
2. Review confidence scores and flag low-confidence extractions
3. Validate AI extractions against source documents
4. Generate reports for customer consultations

📁 Project Structure

bank-comparator/
├── app/
│   ├── models/              # Pydantic data models
│   ├── services/            # Business logic
│   │   ├── document_service.py    # Document processing
│   │   ├── extraction_service.py  # AI fact extraction
│   │   ├── comparison_service.py  # Policy comparison
│   │   └── export_service.py      # Report generation
│   ├── api/                 # FastAPI endpoints
│   │   ├── documents.py           # Document upload/management
│   │   ├── comparison.py          # Comparison endpoints
│   │   └── main.py               # API application
│   └── ui/                  # Streamlit interface
│       ├── main.py               # Main UI application
│       ├── beautiful_comparison.py # Comparison display
│       └── components.py         # Reusable UI components
├── tests/                   # Test suite
├── data/                    # Sample documents and uploads
├── logs/                    # Application logs
├── requirements.txt         # Python dependencies
├── run_api.py              # API server launcher
├── run_ui.py               # UI server launcher
└── README.md               # This documentation

🚀 Quick Start

1. Environment Setup

# Create virtual environment
python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

2. Configuration

# Set up environment variables
cp .env.example .env
# Add your Google Gemini API key to .env

3. Run the Application

# Terminal 1: Start FastAPI backend
python run_api.py
# Or: uvicorn app.api.main:app --reload --port 8000

# Terminal 2: Start Streamlit UI  
python run_ui.py
# Or: streamlit run app/ui/main.py

4. Access the App

• Streamlit UI: http://localhost:8501
• API Documentation: http://localhost:8000/docs
• Health Check: http://localhost:8000/health

✅ Current Status

✅ Completed Features

• Document Processing: PDF/TXT/DOCX upload and text extraction
• AI Integration: Google Gemini LLM for fact extraction with confidence scoring
• Smart Extraction: Rule-based + LLM hybrid approach for accuracy
• Comparison Engine: Side-by-side analysis with status indicators
• Beautiful UI: Butter Money branded interface with blue gradient theme
• Evidence Tracing: Click cells to see normalized values and source evidence
• Export Functionality: CSV/JSON report generation
• API Architecture: FastAPI backend with comprehensive endpoints

🔄 In Progress

• Performance optimization for large documents
• Enhanced error handling and validation
• Comprehensive test coverage
• Docker containerization

🎯 Planned Enhancements

• Batch processing for multiple document sets
• Advanced analytics and insights
• Integration with Butter Money's existing systems
• Mobile-responsive design improvements

🌟 Key Features

� Bank-Specific Intelligence

• Auto-detects major Indian banks (HDFC, ICICI, SBI, Axis, Kotak)
• Clean bank names in comparison headers (no more confusing filenames)
• Contextual explanations for each policy term
• Indian financial notation support (₹, Rs, Lakhs, Crores)

🤖 Advanced AI Processing

• Hybrid extraction: Rule-based + Google Gemini LLM
• Confidence scoring with color-coded indicators
• Smart gap filling for missing information
• Conflict detection for contradictory data

📊 Beautiful Comparison Interface

• Expandable cells showing full values (no truncation)
• Human-readable explanations for each difference
• Evidence tracing with clean source text display
• Status indicators: Same/Different/Missing with clear badges

🔍 Evidence & Transparency

When you click any cell, see:

1. Normalized value with clear formatting
2. Brief explanation in plain English
3. Confidence score with reliability indicator
4. Exact evidence from source document
5. Document reference for verification

🎨 User Experience

• No truncated text - see complete information
• Clear bank identification - HDFC Bank vs ICICI Bank
• Intuitive navigation - click to explore
• Responsive layout for different screen sizes

📈 Export & Analysis

• CSV reports for spreadsheet analysis
• JSON exports for system integration
• Comparison summaries with key metrics
• Evidence documentation for audit trails

🛠 Technology Stack

Backend

• FastAPI - Modern Python web framework
• Google Gemini - LLM for document analysis
• PyMuPDF & PDFPlumber - PDF text extraction
• Pandas - Data processing and analysis
• Pydantic - Data validation and serialization

Frontend

• Streamlit - Interactive web interface
• Plotly - Data visualization
• Custom CSS - Butter Money branding

Development

• pytest - Testing framework
• uvicorn - ASGI server
• python-dotenv - Environment management

🔧 Configuration

Environment Variables (.env)

# Required
GOOGLE_API_KEY=your_gemini_api_key_here

# Optional
LOG_LEVEL=INFO
MAX_FILE_SIZE_MB=10
UPLOAD_DIR=./data/uploads

API Configuration

• Port: 8000 (configurable)
• Max file size: 10MB per document
• Supported formats: PDF, TXT, DOCX
• Concurrent uploads: Up to 4 documents

🐛 Troubleshooting

Common Issues

1. "No module named 'app'" Error

# Make sure you're in the bank-comparator directory
cd bank-comparator
python run_api.py

2. "Google API Key not found" Error

# Check your .env file has the correct API key
echo $GOOGLE_API_KEY  # Should show your key

3. "Port already in use" Error

# Kill existing processes
pkill -f "uvicorn"
pkill -f "streamlit"
# Or use different ports
uvicorn app.api.main:app --port 8001
streamlit run app/ui/main.py --server.port 8502

4. PDF Processing Issues

• Ensure PDFs are text-based (not scanned images)
• Check file size is under 10MB
• Try converting to TXT if extraction fails

5. Slow Processing

• Large documents take longer to process
• Check internet connection for Gemini API calls
• Monitor logs for processing status

Performance Tips

• Smaller files process faster
• Text files are processed quicker than PDFs
• Clear document structure improves extraction accuracy
• Good internet connection reduces API latency

📞 Support

For issues or questions:

1. Check the troubleshooting section above
2. Review API logs in logs/ directory
3. Test with sample documents in data/ folder
4. Verify all dependencies are installed correctly

🚀 Deployment

Local Development

# Development mode with auto-reload
python run_api.py --reload
python run_ui.py --debug

Production Deployment

# Production mode
uvicorn app.api.main:app --host 0.0.0.0 --port 8000
streamlit run app/ui/main.py --server.port 8501 --server.address 0.0.0.0

---

Helping customers discover, compare, and secure their perfect home loan