🩺 Medical RAG Chatbot (LangChain + Pinecone)

A production-ready, full-stack medical question-answering conversational AI built with LangChain, Pinecone, and Flask. This application reads medical PDF documents and answers user queries using Retrieval-Augmented Generation (RAG), drastically reducing hallucinations by grounding the AI in factual medical documents.

---

✨ Features (Refactored & Upgraded)

• Conversational Memory: Remembers past interactions per session using SQLite and LangChain's RunnableWithMessageHistory.
• Real-Time Streaming: Uses Server-Sent Events (SSE) to stream GPT-4o-mini responses to the UI just like ChatGPT.
• Dynamic PDF Uploads: Directly upload new medical PDFs from the UI. Documents are chunked and ingested into Pinecone on the fly.
• Source Citations: Every AI response tracks and displays the exact document sources and page numbers used to generate the answer.
• Chat History & Sessions: Creates and manages multiple chat sessions using a robust SQLite database backend (SQLAlchemy).
• Markdown & Code UI: A modern Light/Dark themed UI with typing indicators, markdown parsing (marked.js), and smooth auto-scrolling.
• Export to PDF: Single-click export of any chat session into a formatted PDF using html2pdf.js.
• Voice Support: Integrated Web Speech API for voice-to-text dictation mapping directly into the chat input.
• Rate Limiting: Integrated Flask-Limiter to protect API endpoints from spam and abuse.

---

🛠 Tech Stack

• Backend Framework: Python / Flask / Flask-SQLAlchemy / Flask-Limiter
• AI & RAG Pipeline: LangChain Classic / OpenAI API (gpt-4o-mini)
• Vector Database: Pinecone Serverless
• Embeddings: HuggingFace (sentence-transformers/all-MiniLM-L6-v2)
• Frontend: Vanilla JS (ES6), HTML5, CSS3, FontAwesome
• Deployment: Docker

---

📂 Project Structure

medical-rag-chatbot/
├── .env                     # Environment variables (Ignored in Git)
├── app.py                   # Flask Application API & Controller
├── config.py                # App Configuration & Env Validations
├── database.py              # SQLite Database configuration
├── models.py                # SQL Models (ChatSession, ChatMessage)
├── Dockerfile               # Containerization configuration
├── requirements.txt         # Python dependencies
├── src/
│   └── services/
│       ├── doc_service.py   # Handles PDF upload, chunking, and indexing
│       ├── llm_service.py   # RAG pipeline, LLM connections, memory
│       └── vector_service.py# Pinecone embedding wrappers
├── utils/
│   ├── rate_limiter.py      # Flask-Limiter definitions
│   └── prompts.py           # System prompts with hardcoded medical disclaimers
├── static/
│   ├── css/style.css        # Responsive, themable styling
│   └── js/chat.js           # Frontend interactions, SSE streaming, Web Speech
└── templates/
    └── chat.html            # Core UI layout

---

⚙️ Setup Instructions

1️⃣ Clone the repository

git clone https://github.com/Shehjad2019/medical-rag-chatbot.git
cd medical-rag-chatbot

2️⃣ Environment Variables

Create a .env file in the root directory:

PINECONE_API_KEY=your_pinecone_api_key_here
OPENAI_API_KEY=your_openai_api_key_here
PINECONE_INDEX_NAME=medicalbot

(Need keys? Get them at Pinecone and OpenAI.)

3️⃣ Run Locally (Virtual Environment)

# Create venv and activate
python -m venv venv
source venv/bin/activate   # Mac/Linux
# venv\Scripts\activate    # Windows

# Install dependencies
pip install -r requirements.txt

# Run the application
python app.py

Access the app: http://localhost:8080

4️⃣ Run via Docker (Recommended for Production)

# Build the image
docker build -t medical-rag:latest .

# Run the container (pass env variables or use --env-file)
docker run -p 8080:8080 \
  -e OPENAI_API_KEY="your_openai_key" \
  -e PINECONE_API_KEY="your_pinecone_key" \
  medical-rag:latest

---

⚠️ Disclaimer & Notes

• Educational Purposes Only: This project is built for learning, prototyping, and portfolio purposes.
• Not Medical Advice: The AI includes strict system prompt disclaimers warning against using its outputs to replace professional medical diagnosis or treatment.
• Data Dependency: The quality of the RAG responses heavily relies upon the quality of the PDFs uploaded to the Pinecone index.

---

👨‍💻 Author

Shehjad Patel
Computer Engineering | GenAI & LangChain Enthusiast

⭐ If you like this project, please consider giving it a star on GitHub! 😊