PDF RAG Chatbot API

A Retrieval-Augmented Generation (RAG) API that allows you to upload PDF documents and ask questions about their content. It uses FastAPI for the backend, Pinecone for vector storage, Sentence Transformers for embeddings, and Groq for fast LLM inference.

Architecture

graph TD
    subgraph Client
        User[User]
    end

    subgraph API["FastAPI Server"]
        UploadEndpoint["/upload_pdf"]
        AskEndpoint["/ask"]
        PDFProcessor["PDF Processor"]
        RAGEngine["RAG Engine"]
    end

    subgraph Services
        Embedder["Embedding Model<br/>(all-MiniLM-L6-v2)"]
        Pinecone["Pinecone Vector DB"]
        LLM["LLM<br/>(Groq)"]
    end

    %% Upload Flow
    User -- Upload PDF --> UploadEndpoint
    UploadEndpoint -- Extract & Chunk --> PDFProcessor
    PDFProcessor -- Text Chunks --> UploadEndpoint
    UploadEndpoint -- Generate Embeddings --> Embedder
    Embedder -- Vectors --> UploadEndpoint
    UploadEndpoint -- Upsert Vectors --> Pinecone

    %% Ask Flow
    User -- Ask Question --> AskEndpoint
    AskEndpoint -- Process Query --> RAGEngine
    RAGEngine -- Embed Query --> Embedder
    Embedder -- Query Vector --> RAGEngine
    RAGEngine -- Retrieve Context --> Pinecone
    Pinecone -- Relevant Chunks --> RAGEngine
    RAGEngine -- Prompt + Context --> LLM
    LLM -- Answer --> RAGEngine
    RAGEngine -- Response --> AskEndpoint
    AskEndpoint -- JSON Response --> User

Loading

Features

• PDF Ingestion: Upload PDF files to extract text and chunk it for processing.
• Vector Search: Uses Pinecone to store and retrieve relevant text chunks based on semantic similarity.
• Question Answering: Generates answers using Groq's cloud-hosted LLMs (default: Llama 3.3 70B), grounded in the retrieved context.
• Modular Design: Clean separation of concerns (API, Config, PDF Processing, RAG Engine).

Tech Stack

• Framework: FastAPI
• Vector Database: Pinecone
• Embeddings: sentence-transformers/all-MiniLM-L6-v2
• LLM: Groq API (supports Llama 3.3, Mixtral, and other models)
• PDF Processing: pypdf

Installation

1. Clone the repository:

   git clone <repository-url>
   cd <repository-directory>

2. Install dependencies:

   pip install -r requirements.txt
   pip install python-dotenv

3. Set up Environment Variables: Copy the .env.example file to .env and add your API keys:

   cp .env.example .env

   Then edit .env and add your keys:

   PINECONE_API_KEY=your_pinecone_api_key_here
   GROQ_API_KEY=your_groq_api_key_here

   Get your Groq API key from: https://console.groq.com/keys

Usage

1. Start the server:

   uvicorn main:app --reload

2. API Endpoints:

   • GET /: Health check. Returns a welcome message.
   • POST /upload_pdf: Upload a PDF file to be indexed.
     • Body: form-data with key pdf and file value.
   • POST /ask: Ask a question about the uploaded PDF.
     • Body: JSON object {"query": "Your question here"}

Docker Setup

You can also run this application using Docker.

1. Build the Docker image:

   docker build -t pdf-rag-app .

2. Run the Docker container: Make sure you have your .env file set up as described in the Installation section.

   docker run -p 8000:8000 --env-file .env pdf-rag-app

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

Project Structure

• main.py: FastAPI application and route definitions.
• config.py: Configuration and initialization of services (Pinecone, Models).
• pdf_processor.py: Logic for extracting and chunking text from PDFs.
• rag_engine.py: Core RAG logic (Retrieval + Generation).
• requirements.txt: Python dependencies.