RAG Application with FastAPI and LangChain

This project implements a Retrieval-Augmented Generation (RAG) application using FastAPI, LangChain, and Mistral AI. It allows users to upload PDF or Word documents, store their text in a vector database, and perform similarity searches with language model-generated responses (e.g., translations). The application is accessible via a browser through FastAPI's Swagger UI.

Features

• Upload Documents: Upload PDF or Word files and store their extracted text in a Chroma vector store, associated with a user ID.
• Search and Generate: Perform similarity searches on stored documents filtered by user ID and generate responses (e.g., translations) using Mistral AI.
• Browser Access: Interact with the API through an interactive Swagger UI at http://localhost:8000/docs.

Prerequisites

• Python 3.8+
• Mistral AI API key (set as environment variable MISTRAL_API_KEY)

Installation

1. Clone the Repository (or create the files manually): Ensure the following files are in your project directory:

   • main.py: FastAPI application with endpoints.
   • vector_store.py: Vector store management (document storage and retrieval).
   • response.py: RAG logic with Mistral AI.

2. Install Dependencies: Install the required Python packages:

   pip install fastapi uvicorn PyPDF2 python-docx langchain_chroma langchain_huggingface langchain_core langchain_text_splitters sentence-transformers langchain_mistralai

3. Set Environment Variable: Set your Mistral AI API key:

   export MISTRAL_API_KEY='your-api-key'

   On Windows, use:

   set MISTRAL_API_KEY=your-api-key

Project Structure

• main.py: Defines FastAPI endpoints for uploading documents and performing RAG searches.
• vector_store.py: Handles document storage and similarity search using Chroma and HuggingFace embeddings.
• response.py: Implements RAG by combining document retrieval with Mistral AI response generation.

Usage

1. Run the FastAPI Server: Navigate to the project directory and run:

   python main.py

   The server will start at http://localhost:8000. If port 8000 is in use, edit main.py to change the port (e.g., port=8001).

2. Access in Browser: Open a browser and navigate to:

   http://localhost:8000/docs
   

   This opens the Swagger UI, where you can test the API endpoints interactively.

3. API Endpoints:

   • POST /upload-document/:
     • Purpose: Upload a PDF or Word document and store its text in the vector store.
     • Parameters:
       • file: PDF or Word file (.pdf or .docx).
       • user_id: Unique identifier (e.g., UUID like 123e4567-e89b-12d3-a456-426614174000).
     • Example:
       • In Swagger UI, click "Try it out", upload a file, enter a user_id, and execute.
       • Response:

         {
           "message": "Document processed and stored successfully",
           "user_id": "123e4567-e89b-12d3-a456-426614174000"
         }

   • POST /search/:
     • Purpose: Perform a similarity search and generate a response (e.g., translation) using retrieved documents.
     • Parameters (JSON body):
       • query: Search query string.
       • user_id: Same user ID used in document upload.
       • input_language: Input language (default: "English").
       • output_language: Desired output language (default: "German").
     • Example:
       • In Swagger UI, enter:

         {
           "query": "LangChain provides abstractions to make working with LLMs easy",
           "user_id": "123e4567-e89b-12d3-a456-426614174000",
           "input_language": "English",
           "output_language": "German"
         }

       • Response:

         {
           "message": "Search and generation completed successfully",
           "user_id": "123e4567-e89b-12d3-a456-426614174000",
           "query": "LangChain provides abstractions to make working with LLMs easy",
           "response": "Ich liebe programmieren.",
           "documents": [
             {
               "content": "This is a sample text to be embedded.",
               "metadata": {"user_id": "123e4567-e89b-12d3-a456-426614174000"}
             }
           ]
         }

4. Testing with cURL (Alternative):

   • Upload a document:

     curl -X POST -F "file=@document.pdf" -F "user_id=123e4567-e89b-12d3-a456-426614174000" http://localhost:8000/upload-document/

   • Perform a search:

     curl -X POST http://localhost:8000/search/ -H "Content-Type: application/json" -d '{"query":"LangChain provides abstractions to make working with LLMs easy","user_id":"123e4567-e89b-12d3-a456-426614174000","input_language":"English","output_language":"German"}'

Notes

• Mistral AI API Key: Required for response.py. Ensure it’s set in the environment or configured in response.py.
• Document Requirements: Uploaded files must be valid PDFs or Word documents with extractable text (e.g., not scanned PDFs without OCR).
• User ID: The user_id in the /search/ endpoint must match the one used in /upload-document/ to retrieve relevant documents.
• Customization: Modify the prompt in response.py to change the RAG behavior (e.g., for question answering instead of translation).
• Vector Store: Documents are stored in ./chroma_langchain_db. Ensure write permissions in the directory.
• Troubleshooting:
  • No Results: Ensure documents are uploaded with the same user_id before searching.
  • Port Conflicts: Change the port in main.py if 8000 is in use.
  • API Errors: Check the server console for detailed error messages.

Dependencies

• fastapi: API framework
• uvicorn: ASGI server
• PyPDF2: PDF text extraction
• python-docx: Word document text extraction
• langchain_chroma: Chroma vector store
• langchain_huggingface: HuggingFace embeddings
• langchain_core: LangChain core components
• langchain_text_splitters: Text splitting utilities
• sentence-transformers: Embedding models
• langchain_mistralai: Mistral AI integration