AI Underwriting Assistant

An advanced document processing system for commercial real estate underwriting, capable of extracting and analyzing information from various document types including rent rolls, P&L statements, operating statements, and lease documents.

• Multi-Format Document Support

  • PDF (with OCR capabilities)
  • Excel spreadsheets
  • Word documents

• Specialized Document Extractors

  • Rent Roll Extractor
    • Tenant information
    • Lease terms
    • Occupancy data
    • Rent analysis
  • P&L Statement Extractor
    • Revenue items
    • Expense categories
    • NOI calculations
    • Historical comparisons
  • Operating Statement Extractor
    • Combined financial metrics
    • Budget variance analysis
    • Period comparisons
  • Lease Document Extractor
    • Key lease terms
    • Financial obligations
    • Special provisions
    • Key dates

• Advanced Analysis

  • Confidence scoring for extracted data
  • Automated data validation
  • Risk factor identification
  • Financial metrics calculation

• API Features

  • Document upload and processing
  • Extraction results retrieval
  • Status monitoring
  • Statistical analysis

• Python 3.9+
• MongoDB 4.4+
• Tesseract OCR
• Poppler (for PDF processing)
• Azure OpenAI API access

1. Clone the repository:

git clone https://github.com/yourusername/ai-underwriting.git
cd ai-underwriting

2. Create and activate a virtual environment:

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

3. Install dependencies:

pip install -r requirements.txt

4. Install Tesseract OCR:

• Windows: Download and install from GitHub
• Mac: brew install tesseract
• Linux: sudo apt-get install tesseract-ocr

5. Install Poppler:

• Windows: Download from Poppler Releases
• Mac: brew install poppler
• Linux: sudo apt-get install poppler-utils

6. Configure environment variables:

cp backend/.env.example backend/.env

Edit .env with your settings.

Key environment variables:

# MongoDB Configuration
MONGODB_URL=mongodb://localhost:27017
MONGODB_DB_NAME=ai_underwriting

# OCR Configuration
POPPLER_PATH=/path/to/poppler
TESSERACT_PATH=/path/to/tesseract

# Azure OpenAI Configuration
AZURE_OPENAI_ENDPOINT=your-endpoint
AZURE_OPENAI_API_KEY=your-key
AZURE_OPENAI_DEPLOYMENT_NAME=your-deployment
AZURE_OPENAI_EMBEDDING_DEPLOYMENT_NAME=your-embedding-deployment

1. Start the server:

cd backend
uvicorn main:app --reload

2. Access the API documentation:

http://localhost:8000/api/v1/docs

• POST /api/v1/documents/upload

  • Upload and process a document
  • Supports PDF, DOCX, XLSX

• GET /api/v1/documents/{document_id}/status

  • Check processing status
  • Get confidence scores

• GET /api/v1/documents/{document_id}/content

  • Get extracted content
  • Access all extractions

• GET /api/v1/documents/{document_id}/extraction/{extractor_type}

  • Get specific extraction results
  • Filter by extractor type

• GET /api/v1/documents/types

  • List supported document types
  • Get extractor descriptions

backend/
├── api/
│   └── documents.py
├── services/
│   ├── extractors/
│   │   ├── base.py
│   │   ├── rent_roll.py
│   │   ├── pl_statement.py
│   │   ├── operating_statement.py
│   │   └── lease.py
│   └── ocr.py
├── db/
│   └── mongodb.py
├── config/
│   └── settings.py
└── main.py

1. Create a new extractor class in services/extractors/
2. Inherit from BaseExtractor
3. Implement required methods:
   • can_handle()
   • extract()
   • validate()

Example:

from .base import BaseExtractor

class NewDocumentExtractor(BaseExtractor):
    def can_handle(self, content: str, filename: str) -> bool:
        # Implement document type detection
        pass

    def extract(self, content: str) -> Dict[str, Any]:
        # Implement data extraction
        pass

    def validate(self) -> bool:
        # Implement validation rules
        pass

Run tests:

pytest

1. Fork the repository
2. Create a feature branch
3. Commit changes
4. Push to the branch
5. Create a Pull Request

This project is licensed under the MIT License - see the LICENSE file for details.