RAG QA Bot

This application allows users to upload PDF files, create a vector database from the document using open-source HuggingFace embeddings, and ask questions related to the PDF content using a Retrieval-Augmented Generation approach. The app integrates with LangChain Framework, OpenAI's LLM and HuggingFace embeddings.

Features

• Upload a PDF file and save it locally. Later we can create the API to delete the old files.
• Create a vector database from the PDF's content using HuggingFace model sentence-transformers/all-mpnet-base-v2
• Ask questions about the PDF content.
• View the context used for answering the questions that is toggleable via a checkbox.
• The POC of RAG pipeline is tested in using rag_pipeline.ipynb

Screenshots of the RAG app

Installation

Clone the Repository

git clone https://github.com/yourusername/rag-qa-bot.git
cd rag-qa-bot

Set up a Virtual Environment (optional but recommended)

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

Install Dependencies

pip install -r requirements.txt

The main packages are:

• streamlit: For the web UI.
• PyPDFLoader: To extract content from PDF files.
• langchain: For embeddings, document chunking, and question-answering.
• faiss-cpu: For vector store creation and retrieval.
• openai: To integrate with OpenAI's language models.

Configuration

We provide a config.json file in the root directory, this will allow you to select the models at your choice, with the following details:

{
  "embedding_model": "sentence-transformers/all-MiniLM-L6-v2",
  "openai_api_key": "your_openai_api_key",
  "openai_model": "gpt-3.5-turbo",
  "vector_db_path": "./vector_store"
}

Usage

Running the App

To launch the Streamlit web app, run the following command in your terminal:

streamlit run main.py

browse the url http://localhost:8501/

How It Works

1. PDF Upload: The user uploads a PDF file using the Streamlit file uploader.
2. Document Chunking: The PDF content is split into manageable chunks using the RecursiveCharacterTextSplitter api fo LangChain.
3. Embeddings Generation: The chunks are passed through a HuggingFace embedding model to generate embeddings.
4. Vector Store Creation: The embeddings are stored in a FAISS-based vector store, which is then saved locally.
5. Question Answering: When a user asks a question, the system retrieves the relevant context from the vector store and generates an answer using OpenAI's LLM.

Project Structure

|-- src/
|   |-- utils.py             # Helper functions such as file-saving logic
    |-- rag_application.py   # Class to implement the RAG pipeline
|-- main.py                  # Main Streamlit app file
|-- requirements.txt         # List of required dependencies
|-- config.json              # Configuration file
|-- rag_pipeline.ipynb       # Test and POC the RAG pipeline