RAG_Playground

A proof-of-concept Retrieval-Augmented Generation (RAG) playground built with Flask, LangChain, and OpenAI/Tavily. This project demonstrates how to combine PDF retrieval, web search, and conversational agents into a simple chat interface.

Features

• AgenticRAG: loads PDFs, applies adaptive chunking, builds a FAISS vector store, and uses a LangChain agent with tool support.
• PDF search tool for retrieving relevant document chunks.
• Web search tool powered by Tavily for up-to-date answers when PDFs do not contain enough information.
• Conversation memory for chat history and context.
• Simple Flask frontend with a chat UI at /.
• Additional hybrid retrieval utilities for BM25 and metrics evaluation.

Repository Structure

• app.py - Flask web app exposing chat, clear memory, and history endpoints.
• agenticrag.py - Main RAG implementation using LangChain agents and tools.
• adaptivechunking.py - Adaptive text chunking utility for document splitting.
• hybridrag.py - Hybrid retrieval support combining FAISS and BM25.
• ragmetrics.py - Evaluation metrics for RAG retrieval performance.
• templates/chat.html - Web UI for chat interaction.
• requirements.txt - Python dependencies.

Requirements

• Python 3.10+ recommended
• pip for package installation
• OpenAI API key
• Tavily API key

Setup

1. Create a Python virtual environment:

python3 -m venv .venv
source .venv/bin/activate

2. Install dependencies:

pip install -r requirements.txt

3. Add your API keys inside app.py or via environment variables.

4. Place your PDF documents in the repository root or update the pdf_files list in app.py.

Running the App

Start the Flask server:

python app.py

Then open http://127.0.0.1:5000 in your browser to use the chat UI.

Usage

• Ask questions in the chat box.
• The agent will search uploaded PDFs first and fallback to web search if needed.
• Use the /clear endpoint to reset conversation memory.
• The /history endpoint returns chat history as JSON.

Example

Update the PDF files and keys in app.py:

rag = AgenticRAG(
    pdf_files=["./document1.pdf", "./document2.pdf"],
    openai_api_key="YOUR_OPENAI_API_KEY",
    tavily_api_key="YOUR_TAVILY_API_KEY",
    verbose=False,
)

Then run the app and ask questions about the uploaded documents.

Notes

• The app currently expects PDF files in the root folder.
• This project is designed for experimentation and demonstration, not production deployment.
• hybridrag.py and ragmetrics.py provide additional retrieval and evaluation tools for advanced RAG experimentation.

License

This repository does not include a license file. Add one if you intend to open source or share the project.