Swift OCR: LLM Powered Fast OCR ⚡

🌟 Features

• Flexible Input Options: Accepts PDF files via direct upload or by specifying a URL.
• Advanced OCR Processing: Utilizes OpenAI’s GPT-4 Turbo with Vision model for accurate text extraction.
• Performance Optimizations:
  • Parallel PDF Conversion: Converts PDF pages to images concurrently using multiprocessing.
  • Batch Processing: Processes multiple images in batches to maximize throughput.
  • Retry Mechanism with Exponential Backoff: Ensures resilience against transient failures and API rate limits.
• Structured Output: Extracted text is formatted using Markdown for readability and consistency.
• Robust Error Handling: Comprehensive logging and exception handling for reliable operations.
• Scalable Architecture: Asynchronous processing enables handling multiple requests efficiently.

📹 Demo

https://github.com/user-attachments/assets/6b39f3ea-248e-4c29-ac2e-b57de64d5d65

Demo video showcasing the conversion of NASA’s Apollo 17 flight documents, which include unorganized, horizontally and vertically oriented pages, into well-structured Markdown format without any issues.

Here’s a single, comprehensive section on cost comparison for your README:

Cost Comparison and Value Proposition

Our solution offers an optimal balance of affordability, accuracy, and advanced features:

Cost Breakdown

• Average token usage per image: ~1200
• Total tokens per page (including prompt): ~1500
• [GPT4O] Input token cost: $5 per million tokens
• [GPT4O] Output token cost: $15 per million tokens

For 1000 documents:

• Estimated total cost: $15

Cost Optimization Options

1. Utilizing GPT4 mini: Reduces cost to ~$8 per 1000 documents
2. Implementing batch API: Further reduces cost to ~$4 per 1000 documents

Market Comparison

This solution is significantly more affordable than alternatives:

• Our cost: $15 per 1000 documents
• CloudConvert: ~$30 per 1000 documents (PDFTron mode, 4 credits required)

While cost-effectiveness is a major advantage, our solution also provides:

• Superior accuracy and consistency
• Precise table generation
• Output in easily editable markdown format

This combination of affordability and advanced features makes solution stand out in the document processing market. It’s not just about being cheaper; it’s about providing excellent value through reliability, flexibility, and high-quality output.

🛠️ Installation

Prerequisites

• Python 3.8+
• Git
• Virtualenv (optional but recommended)

Steps

1. Clone the Repository

   git clone https://github.com/yigitkonur/llm-openai-ocr.git
   cd llm-openai-ocr
   

2. Create a Virtual Environment

   python3 -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate
   

3. Install Dependencies

   pip install -r requirements.txt
   

4. Configure Environment Variables

   Create a .env file in the root directory and add the following variables:

   OPENAI_API_KEY=your_openai_api_key
   AZURE_OPENAI_ENDPOINT=your_azure_openai_endpoint
   OPENAI_DEPLOYMENT_ID=your_openai_deployment_id
   OPENAI_API_VERSION=your_openai_api_version  # Default is "gpt-4o"
   BATCH_SIZE=10  # Optional: Default is 1
   MAX_CONCURRENT_OCR_REQUESTS=5  # Optional: Default is 5
   MAX_CONCURRENT_PDF_CONVERSION=4  # Optional: Default is 4
   

   Note: Replace your_openai_api_key, your_azure_openai_endpoint, and your_openai_deployment_id with your actual OpenAI credentials.

5. Run the Application

   uvicorn main:app --reload
   

   The API will be available at http://127.0.0.1:8000.

🎯 Usage

API Endpoint

POST /ocr

Request Parameters

• file: (Optional) Upload a PDF file.
• ocr_request.url: (Optional) URL of the PDF to process.

You must provide either a file or a URL, not both.

Example Using curl

Uploading a PDF File:

curl -X POST "http://127.0.0.1:8000/ocr" -F "file=@/path/to/your/document.pdf"

Providing a PDF URL:

curl -X POST "http://127.0.0.1:8000/ocr" -F "ocr_request={\"url\": \"https://example.com/document.pdf\"}" -H "Content-Type: application/json"

Response

• 200 OK

  {
    "text": "Extracted and formatted text from the PDF."
  }
  

• Error Responses

  • 400 Bad Request: Invalid input parameters.
  • 422 Unprocessable Entity: Validation errors.
  • 500 Internal Server Error: Processing errors.

🧰 Configuration

All configurations are managed via environment variables. Ensure you have a .env file set up with the necessary variables as described in the Installation section.

Key Configuration Variables

• OPENAI_API_KEY: Your OpenAI API key.
• AZURE_OPENAI_ENDPOINT: The endpoint for Azure OpenAI service.
• OPENAI_DEPLOYMENT_ID: Deployment ID for the OpenAI model.
• OPENAI_API_VERSION: API version for OpenAI (default: “gpt-4o”).
• BATCH_SIZE: Number of images to process per OCR request (default: 1).
• MAX_CONCURRENT_OCR_REQUESTS: Maximum number of concurrent OCR requests (default: 5).
• MAX_CONCURRENT_PDF_CONVERSION: Maximum number of concurrent PDF page conversions (default: 4).
  Here’s the revised license section with the requested changes:

📜 License (thx for issue)

Please note that PyMuPDF requires changing the license to GNU AGPL v3.0. You can fork this project, implement pdf2image, and use it freely. While I don’t have any particular interest in licensing, I am legally obligated to add this information.

GNU AFFERO GENERAL PUBLIC LICENSE
Version 3, 19 November 2007

Copyright © 2024 Yiğit Konur

This program is free software: you can redistribute it and/or modify
it under the terms of the GNU Affero General Public License as published
by the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.

This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU Affero General Public License for more details.

You should have received a copy of the GNU Affero General Public License
along with this program. If not, see https://www.gnu.org/licenses/.