MediBot: A Private AI Medical Chatbot

MediBot is a 100% private and secure medical chatbot. It uses a locally-hosted Llama 2 model to answer questions only based on a set of trusted medical documents.

This project uses Retrieval-Augmented Generation (RAG) to prevent the AI from "hallucinating" or providing dangerous, non-factual medical advice. The AI is forced to find the answer in your documents and then summarize it.

Demo

Key Features

• 100% Local LLM: Your questions are processed on your own computer by Llama 2. Nothing is sent to an external API (like OpenAI or Google).
• RAG Pipeline: Ensures all answers are based only on the trusted medical PDFs you provide, building in safety and accuracy.
• Professional Vector DB: Uses Pinecone as a cloud-based, scalable vector store for managing the knowledge base.
• Source-Backed Answers: The chatbot shows you exactly which documents it used to find the answer, building user trust.
• Professional UI: A custom-built, responsive dark-mode UI with a "popping" hover effect, built with Flask, HTML, CSS, and JavaScript.

Architecture

The application is a hybrid model, combining the best of local privacy and cloud scalability.

1. Frontend (HTML/CSS/JS): A user-friendly web interface that sends user questions to the backend.
2. Backend (Flask): A Python server that:
   • Receives the question.
   • Queries the Pinecone vector database to find relevant facts.
   • Injects those facts into a prompt for the Local Llama 2 model.
   • Returns the final, sourced answer to the frontend.

Tech Stack

• Backend: Python, Flask
• LLM: Llama 2 (running locally via CTransformers)
• Vector DB: Pinecone
• Embeddings: Sentence-Transformers (all-MiniLM-L6-v2)
• Document Loading: LangChain, PyPDFLoader
• Frontend: HTML, CSS, JavaScript

Getting Started

Follow these steps to set up and run the project on your local machine.

1. Prerequisites

• Python 3.9 or newer.
• A Llama 2 GGUF model file. This project runs the LLM locally and requires you to download the model file (approx. 4GB).
• A free Pinecone account.

2. Installation

1. Clone the repository:

   git clone [https://github.com/YOUR_USERNAME/YOUR_REPOSITORY_NAME.git](https://github.com/YOUR_USERNAME/YOUR_REPOSITORY_NAME.git)
   cd YOUR_REPOSITORY_NAME

2. Create and activate a virtual environment:

   # Create the venv
   python -m venv venv
   
   # Activate on Windows
   .\venv\Scripts\activate

3. Install all required packages:

   pip install -r requirements.txt

3. Configuration

1. Download the Llama 2 Model:

   • You must download a Llama 2 GGUF model file.
   • A compatible model can be downloaded here: Llama-2-7B-Chat-GGUF (Q4_K_M)
   • Save this .gguf file to a permanent location on your computer (e.g., C:\Users\YourUser\models\).

2. Set up Pinecone:

   • Log in to your Pinecone account.
   • Create a new Index.
   • Give it an Index name (e.g., medical-bot).
   • Crucial: Set the Dimensions to 384 (to match our embedding model).
   • Set the Metric to cosine.

3. Set up Environment File:

   • Create a file named .env in the project root.
   • Go to "API Keys" in Pinecone and copy your key.
   • Add your key to the .env file:

     PINECONE_API_KEY=YOUR_SECRET_API_KEY_HERE
     

4. Update Config Files:

   • ingest.py: Change the INDEX_NAME variable to match your Pinecone index name.

   • app.py:

     • Change the INDEX_NAME variable to match your Pinecone index name.
     • Crucial: Change the MODEL_PATH variable to the full, absolute path of your downloaded Llama 2 .gguf file from Step 1.

     # Example of what to change in app.py:
     MODEL_PATH = r"C:\Users\YourUser\models\llama-2-7b-chat.Q4_K_M.gguf"

Step-by-Step Execution

Step 1: Ingest Your Data (One-Time Setup)

This script "teaches" the bot by reading your documents and uploading them to Pinecone.

1. Place all your medical .pdf files into the /data folder.
2. Run the ingestion script from your terminal:

   python ingest.py

3. Wait for the script to finish. You will see a "Successfully uploaded" message.

Step 2: Run the Chatbot

1. Run the main Flask application:

   python app.py

2. Wait for the LLM to load. You will see a message like: * Running on http://127.0.0.1:5000
3. Open that URL (http://127.0.0.1:5000) in your web browser.
4. You can now start asking questions!