RAG Document Application

An end-to-end RAG application (from scratch) based on FastAPI that processes PDFs, images, and web pages to obtain OCR data, generates embeddings using OpenAI's embedding models, and utilizes Pinecone as a vector database for search. It answers questions based on search results using OpenAI Chat Completion!

Contents

• Checklist
• Installation
• Configuration
• Usage
• Endpoints

Checklist

• Basic API authentication
• End-2-End API authentication/authorization (coming)
• Rate limiting to prevent abuse.
• Asynchronous processing.
• File uploading (AWS S3) including validation and sanitization.
• OCR processing (from mock files), Embedding generation using OpenAI.
• End-2-End OCR (Model Serving) and Google APIs, Mathpix API Integrations
• processes PDFs, Images
• processes Web Pages
• Storage and search of embeddings with Pinecone.
• Search from the document
• Applied Tokenizers for better search performance.
• Talk to your data! Chat and generate the answer from the search results using OpenAI's chat completions.
• OpenAPI chat completion streaming support
• Dockerize App.
• Implemented Redis for caching and frequent queries.
• The detailed comments, docstrings, lint checks, pip8.
• CD (GitHub actions)
• CI & PyTest codes
• Chat Interface

Prerequisites

• Python 3.11
• Docker
• Docker Compose (v2.20.2)
• AWS S3 Credentials (API-Keys)
• OpenAI API-Key
• Pinecone API-Key
• Prepare your .env file, learn in Configuration

Installation

git clone https://github.com/teamunitlab/rag-document-app.git
cd rag-document-app/
docker-compose up -d --build

Configuration

1. Fill a .env file in the root directory and add the following environment variables:

   API_KEY=your_api_key_here
   OPENAI_API_KEY=your_openai_api_key_here
   PINECONE_API_KEY=your_pinecone_api_key_here
   PINECONE_INDEX_NAME=your_pinecone_index_name
   AWS_ACCESS_KEY_ID=your_aws_access_key_id
   AWS_SECRET_ACCESS_KEY=your_aws_secret_access_key
   AWS_DEFAULT_REGION=your_aws_region
   BUCKET_NAME=your_s3_bucket_name

2. Ensure AWS credentials and bucket policies are correctly configured to allow S3 access.

Usage

1. Run the FastAPI application:

   docker-compose up -d

2. Access the API documentation at http://localhost:8000/docs.

Endpoints

Upload File

• URL: /upload
• Method: POST
• Description: Upload files to S3. Limited to 10 requests per minute.
• Request:
  • files: List of files to upload.
  • API-Key: Header for API key authentication.
• Response: JSON containing file IDs and URLs.

Process OCR

• URL: /ocr
• Method: POST
• Description: Process OCR for a given file URL. Limited to 10 requests per minute.
• Request:
  • url: URL of the file to process, it is obtained during file (ducument) uploading.
  • API-Key: Header for API key authentication.
• Response: JSON containing information about the processing status.

Extract Data

• URL: /extract
• Method: POST
• Description: Reply to a query using OpenAI chat completions and search based on the given File ID from Pinecone, Limited to 10 requests per minute.
• Request:
  • file_id: The file ID, it is obtained during file (ducument) uploading.
  • query: Ask a question from your document!
  • API-Key: Header for API key authentication.
• Response: JSON containing information about a reply to the question and the three top search results.

References

1. http://unitlab.ai/
2. https://blog.unitlab.ai/unitlab-ai-data-collection-and-annotation-for-llms-and-generative-ai/
3. https://docs.unitlab.ai/ai-models/model-integration
4. https://blog.unitlab.ai/