OCR LLM Experience Letter Parser

A FastAPI-based service that automates the extraction and parsing of information from experience letters and relieving letters using OCR and Large Language Models.

Features

• Document Processing: Supports both PDF and image formats
• OCR Integration: Uses doctr for accurate text extraction
• LLM Processing: Leverages Predibase's Mistral-7B model for intelligent information extraction
• Structured Output: Returns parsed information in consistent JSON format
• Secure API: Includes basic authentication for API endpoints
• Comprehensive Logging: Built-in logging system for debugging and monitoring

Architecture

The service follows a modular architecture:

• FastAPI for RESTful API endpoints
• Document OCR processing using doctr
• LLM-based text parsing using Predibase
• JSON response formatting

Technical Requirements

• Python 3.8+
• CUDA-compatible GPU (for OCR processing)
• Required Python packages:
  • fastapi
  • uvicorn
  • python-multipart
  • doctr
  • predibase
  • python-jose[cryptography]

Installation

1. Clone the repository:

git clone https://github.com/rohandhupar1996/paperface-llm.git
cd OCR_PREDIBASE

2. Install dependencies:

pip install -r requirements.txt

3. Configure the application:

• Update src/configs/config.json with your Predibase API token and model preferences
• Modify logging configuration if needed

Usage

1. Start the server:

python main.py

The server will start on http://0.0.0.0:9001

2. API Endpoints:

• Health Check:

GET /health

• Document Upload and Processing:

POST /upload/

3. Authentication:

• Default credentials:
  • Username: admin
  • Password: secret

API Documentation

POST /upload/

Upload one or more experience/relieving letters for processing.

Request:

• Method: POST
• Authentication: Basic Auth
• Content-Type: multipart/form-data
• Body: List of files (PDF/Images)

Response:

[
    {
        "filename": "example.pdf",
        "file_output": {
            "FullName": "John Doe",
            "Employer": "Example Corp",
            "EmployeeId": "EMP123",
            "Designation": "Software Engineer",
            "Location": "New York",
            "DateOfJoining": "01/01/20",
            "LastDateOfWorking": "31/12/22"
        }
    }
]

Error Handling

The service includes comprehensive error handling for:

• File type validation
• OCR processing errors
• LLM processing failures
• Authentication failures

Logging

Logs are written to app.log and include:

• Request processing details
• OCR operation status
• LLM processing results
• Error traces for debugging

Security Considerations

• Basic authentication required for API endpoints
• Configuration stored in separate config file
• Sensitive information should be managed through environment variables
• File type validation before processing

Contributing

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