Document Q&A RAG System with FastAPI

A powerful Document Question-Answering system built using Retrieval-Augmented Generation (RAG) architecture, FAISS Vector Search, and a FastAPI backend. It supports both Bengali and English for querying documents and provides context-aware answers. The system is also integrated with Streamlit for an interactive, user-friendly interface.

Project Demonostration in Youtube

Document Question Answering using RAG LLM | FAISS | FastAPI

Web Application

Streamlit App

Features

• 📄 PDF Document Processing
• 🔄 Multiple Text Extraction Methods (pdfplumber, pymupdf, langchain)
• 💾 Vector Database Storage using FAISS
• 🤖 RAG-based Question Answering
• 🌐 FastAPI Backend with Swagger Documentation
• 🖥️ Interactive Streamlit Frontend
• 📊 RAG System Evaluation Tools
• 🔤 Multi-language Support (Bengali & English)

Project Structure

├── app/
│   ├── config/           
│   ├── core/            
│   ├── data/             
│   │   ├── pdfs/       
│   │   ├── texts/       
│   │   └── vectorstores/ 
│   ├── processing/      
│   ├── routes/         
│   ├── schemas/        
│   └── services/       
├── docs/               
├── logs/              
├── requirements.txt   
├── run.sh            
└── streamlit_app.py  

Installation

1. Clone the repository:

git clone https://github.com/FaisalAhmedBijoy/Document-QA-RAG-System-FastAPI.git
cd Document-QA-RAG-System-FastAPI

2. Create and activate a conda environment:

conda create -n rag_llm python=3.12
conda activate rag_llm

3. Install dependencies:

pip install -r requirements.txt

4. Collect Groq API KEY from Groq and paste API KEY in the .env file.

GROQ_API_KEY="YOUR_API_KEY"

Key Dependencies

• FastAPI: Web framework for building APIs
• Streamlit: Frontend interface
• LangChain: Framework for developing LLM applications
• FAISS: Vector similarity search
• Pydantic: Data validation
• PDF Processing: pdf2image, pdfplumber, PyMuPDF

Usage

1. Start the FastAPI backend:

   python -m app.main

2. Launch the Streamlit frontend:

   streamlit run streamlit.py

3. Access the applications:

   • FastAPI Swagger UI: http://localhost:8000/docs
   • Streamlit Interface: http://localhost:8501

4. Run using Docker:

   docker compose -f docker-compose.yaml up --build
   docker compose -f docker-compose.yaml up

Processing Pipeline

1. PDF Processing

   • Convert PDF documents to text using multiple methods
   • Store extracted text in the app/data/texts directory

   python -m app.processing.pdf_to_text

2. Text Chunking

   • Split documents into manageable chunks
   • Optimize chunk size for better retrieval

   python -m app.processing.generate_text_chunks

3. Embedding Generation and Vector Store

   • Generate embeddings for text chunks. Embedding model used l3cube-pune/bengali-sentence-similarity-sbert from HuggingFace.
   • Store vectors in FAISS index

     python -m app.processing.generate_embeddings
     python -m app.processing.generate_vector_db

   • Vector stored into app/data/vectorstores/faiss_index

4. RAG Chain

   • Retrieve relevant context using vector similarity
   • Generate answers using LLM with retrieved context. Lllama-3.3-70b-versatile LLM model used from Groq.

     python -m app.processing.generate_rag_chain

   • prompt used for this task

     prompt_template = """
        You are an assistant that answers questions strictly based on the provided document text.
     
        Rules:
        - Only use the information from the given Context.
        - Do not use outside knowledge.
        - If the answer is not found in the Context, reply exactly: "Information not found in the document."
        - Provide only the answer, without repeating the question or the context.
     
        Context: {context}
        Question: {question}
        Answer:
        """

5. Single Query Inference

   • Run a query using LLM model

     python -m app.processing.single_query_inference

   • Perform question answering about the document

     "query": "What is the email address of the candidate?",
     "answer": "faisal.cse16.kuet@gmail.com"

6. API Endpoints

   

   • /rag/query: Process questions and generate answers

     • Method: POST
     • Params

       "query": "What is the name the candidate?",

     • Response

       "query": "What is the name the candidate?",
       "answer": "Faisal Ahmed",

   • /rag/query-with-reference: Process questions and generate answers

     • Method: POST
     • Params

       "query": "What is the job duration in Business Automation Limited?"
       "expected_answer": "Nov 2024 - Present",

     • Response

       "query": "What is the job duration in Business Automation Limited?",
       "expected_answer": "Nov 2024 - Present",
       "actual": "November, 2024 - Present",
       "cosine_similarity": 0.8834819187419994,
       "context": [
          "[Doc 1]: • Developed cluster-based remark suggestions in a product • Developed a time series forecasting model for a product registration count. • Data analysis on project cost estimation in the company data and perform EDA. • Developed a website backend service using FastAPI and PostgreSQL that includes SMS and email sending modules, payment gateway integration, and custom PDF generation. Next Solution Lab, Dhaka, Bangladesh February, 2024 - October, 2024 AI Engineer • Developed deep learning-based prod",
          "[Doc 2]: Sentence Punctuation Restoration [GitHub] A transformer-based Bangla model was used to build the sentence punctuation model. Llama 3.2 was also used to infer with non-punctuation sentence correction. FastAPI was used to prepare the API for deployment with Docker. Tech Stack: BanglaBERT, LLM, Llama 3.2, FastAPI Chat Bot using LLM with Gradio [GitHub] The chatbot is built with Flask for the backend and uses a pre-trained model from Hugging Face for generating responses. Tech Stack: LLM, Gen AI, Fl",
       ]

   • /rag/upload-document-pdf: Upload PDF for generate new FAISS vector

     • Method: POST
     • Params: PDF File
     • Response: FAISS vector saved Document ID.

   • /rag/query-by-document: Process questions and generate answers based on document id

     • Method: POST
     • Params

       "query": "What is the name the candidate?",
       "document_id": "1757268986878",

     • Response

       "query": "What is the name the candidate?",
       "answer": "Faisal Ahmed",

   • /rag/list-vector-storest: Show the current vector list

     • Method: GET
     • Response

       "vectors": {
          "vector_stores": [
             "faiss_index_1757267758132",
             "faiss_index_1757268986878",
          ],
          "document_ids": [
             "1757267758132",
             "1757268986878",
          ]
       }

   • /rag/pdf/{ddocument_id}: Show the PDF with document id

     • Method: GET

Evaluation

The system includes evaluation tools to measure:

• Answer relevancy
• Response accuracy
• Retrieval quality

Documentation

• API documentation available at http://127.0.0.1:8000/docs endpoint
• Sample outputs and UI screenshots in docs/ directory
• Detailed logging in logs/ directory

Acknowledgments

• FastAPI
• LangChain
• Streamlit
• FAISS