Skip to content

README.md

Latest commit

 

History

History
102 lines (74 loc) · 3.69 KB

README.md

File metadata and controls

102 lines (74 loc) · 3.69 KB

Medical Chatbot with RAG Architecture

A medical question-answering system that leverages Retrieval-Augmented Generation (RAG) to provide accurate and context-aware medical information. This application combines the power of large language models with vector search to deliver precise answers to medical queries.

🚀 Key Features

  • Advanced RAG Pipeline: Implements a robust Retrieval-Augmented Generation system for accurate medical information retrieval
  • Multi-document Support: Processes and indexes multiple PDF documents from the data/ directory
  • Semantic Search: Utilizes Pinecone's vector database for efficient similarity search across medical documents
  • State-of-the-Art LLM: Powered by Google's Gemini model through LangChain for high-quality response generation
  • Web Interface: User-friendly Flask-based web interface for seamless interaction
  • Scalable Architecture: Designed for easy extension and integration with additional data sources
  • Customizable Prompts: Easily adjustable system prompts to tailor responses to medical domain requirements
  • Efficient Chunking: Smart text splitting to maintain context while processing large documents

📊 Technical Architecture

The application follows a modern microservices architecture with the following components:

  1. Frontend: Lightweight HTML/JS interface with responsive design
  2. Backend: Flask web server handling API requests
  3. Vector Database: Pinecone for efficient vector similarity search
  4. Embedding Model: all-MiniLM-L6-v2 for creating document embeddings
  5. LLM Integration: Google's Gemini model for generating human-like responses
  6. Document Processing: Automated pipeline for PDF ingestion and text extraction

Project Structure

  • app.py: Flask app and RAG pipeline
  • store_index.py: Builds Pinecone index from PDFs in data/
  • src/helper.py: Load PDF(s), split text, and create embeddings
  • src/prompt.py: System prompt for the assistant
  • templates/chat.html: Frontend chat page
  • static/style.css: Simple styles

Prerequisites

  • Python 3.10+
  • A Pinecone account and API key
  • A Google AI Studio API key (for Gemini)

1) Setup (Windows PowerShell)

python -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
python -m pip install -r requirements.txt

2) Environment Variables

Create a .env file in the project root:

PINECONE_API_KEY="your_pinecone_api_key"
GOOGLE_API_KEY="your_google_api_key"

Notes:

  • The code uses the Pinecone index name medical-catboot and expects embeddings of dimension 384 (all-MiniLM-L6-v2).
  • Default serverless spec: cloud aws, region us-east-1.

3) Add/Update Your Data

Place your PDFs in the data/ folder (the repo includes data/Medical_book.pdf). Rebuild the index after you change files.

4) Build/Refresh the Pinecone Index

.\.venv\Scripts\python store_index.py

This will:

  • Read PDFs from data/
  • Split into chunks
  • Create sentence-transformer embeddings (384 dims)
  • Create or reuse Pinecone index medical-catboot
  • Upsert embeddings

5) Run the App

.\.venv\Scripts\python app.py

Open http://localhost:8080 in your browser.

Troubleshooting

  • Ensure .env contains valid PINECONE_API_KEY and GOOGLE_API_KEY.
  • If you change PDFs, rerun store_index.py to refresh embeddings.
  • If the index doesn’t exist, the script will create it (serverless us-east-1).
  • On corporate networks, set proxy env vars for pip/downloads if needed.

Tech Stack

  • Python, Flask
  • LangChain (Retrieval chain)
  • Sentence Transformers (all-MiniLM-L6-v2)
  • Pinecone (Vector DB)
  • Google Gemini (via langchain-google-genai)