🧠 MediBot – AI-Powered Medical Chatbot with RAG using LangChain and Hugging Face

MediBot is an AI-powered chatbot built using LangChain, Hugging Face, FAISS, and Streamlit. It leverages Retrieval-Augmented Generation (RAG) to answer medical queries based only on uploaded medical PDFs, avoiding hallucinations and ensuring responses are grounded in real, verified content.

#Team members

1. Rahul kumar
2. Bipeen Raj
3. Piyush Ranjan
4. Sourav

---

📌 Features

• 💬 Ask medical questions in natural language
• 📚 Answers are retrieved from your uploaded PDF documents
• 🔎 Uses FAISS for fast vector-based semantic search
• 🤖 Uses Hugging Face Transformers for LLM responses
• 📄 Preprocesses large PDFs into chunks for accurate retrieval
• 🧠 Embeds chunks using Sentence Transformers

---

🛠️ Tech Stack

Tool/Library	Purpose
LangChain	Building RAG pipelines
FAISS	Storing and querying vector DB
HuggingFace	Hosting LLMs and embedding models
Streamlit	Web UI for interaction
Python + venv	Core development & environment mgmt

---

📂 Project Structure

MediBot/ ├── data/ # Folder containing input PDFs ├── vectorstore/ # FAISS vector database │ └── db_faiss/ # Stored embeddings ├── create_memory_for_llm.py # Script to load PDFs & store vector DB ├── connect_memory_with_llm.py # Query LLM with vector DB (RAG) ├── .env # Hugging Face API key └── README.md # This file

⚙️ Setup Instructions

1. Create and Activate a Virtual Environment

python -m venv venv venv\Scripts\activate # Windows

2. Install Dependencies

pip install -r requirements.txt

3. Add Your Hugging Face Token Create a .env file in the root directory and add:

HF_TOKEN=your_huggingface_api_token_here

You can get a free token at: https://huggingface.co/settings/tokens

🧠 Step 1: Create the Vector Memory (FAISS Index)

Run this script to load your PDFs, chunk text, generate embeddings, and save to FAISS:

python create_memory_for_llm.py

it will- ✅ Loads PDFs from data/ ✅ Splits text into chunks ✅ Embeds each chunk ✅ Saves embeddings into FAISS (vectorstore/db_faiss)

🔍 Step 2: Interact with the Chatbot (RAG + LLM)

Once the FAISS vector store is ready, run:

python connect_memory_with_llm.py

Ask medical questions and get LLM responses grounded in your PDFs.

🌐 Optional: Run the Web App Launch the chatbot interface via Streamlit:

streamlit run app.py

Make sure app.py is configured to load the correct FAISS DB and embedding model.

🧪 Sample Query Input:

What are the symptoms of malaria?

Response:

Malaria symptoms typically include high fever, chills, sweating, headache, nausea, and muscle pain. These symptoms often appear 10–15 days after the bite of an infected mosquito.

📦 requirements.txt

langchain>=0.2.0 langchain-community>=0.2.0 langchain-huggingface>=0.0.1 huggingface_hub>=0.21.4 faiss-cpu>=1.7.4 sentence-transformers>=2.6.1 python-dotenv>=1.0.1 streamlit>=1.35.0

Install using:

pip install -r requirements.txt

📄 Dataset Notes

Source: The Gale Encyclopedia of Medicine

Place your PDF files in the data/ folder.

Ensure PDFs are well-formatted (not scanned images) for accurate text extraction and chunking.

⚠️ Important Notes

-FAISS DB is local and persistent -Responses are limited to your PDF data — no external knowledge used -Works offline after embedding stage (Hugging Face API still required during query) -Use from a trusted environment — deserialization involves pickle