Basic_Medical_Chatbot_Using_RAG

This project implements a Retrieval-Augmented Generation (RAG) based medical chatbot using a single Jupyter Notebook. The chatbot aims to provide accurate and contextually relevant responses to medical queries by combining retrieval from a vector database with generative capabilities.

Project Overview

Medical chatbots can play a crucial role in providing healthcare information, symptom assessment, and wellness advice, particularly for remote areas or during non-office hours. This project leverages a RAG-based approach to build a medical chatbot that retrieves relevant contexts and generates responses, enhancing user interaction and information accuracy.

Key Components of the Project

1. Data Gathering and Preparation:

   • Load and prepare a medical dataset from Hugging Face (e.g., PubMed QA) to train the chatbot.
   • Preprocess and chunk context data to fit within the model's embedding limitations.

2. Vector Database Creation:

   • Create dense and sparse embeddings of medical contexts using Sentence Transformers and SpladeEncoder models.
   • Store embeddings in a Pinecone vector database to enable fast and accurate retrieval.

3. Retrieval-Augmented Generation (RAG) Pipeline:

   • Build a RAG pipeline using LangChain to combine retrieval from Pinecone with text generation using OpenAI’s GPT model.
   • Queries are passed to the RAG pipeline, which retrieves relevant contexts and generates a response.

4. Evaluation:

   • Use the RAGAS evaluation metrics to assess the quality of responses based on context recall, context precision, faithfulness, answer relevancy, answer similarity, and answer correctness.
   • Visualize evaluation results for better insights into model performance.

Project Flow

1. Import Libraries: Import all necessary Python libraries such as Hugging Face datasets, Pinecone, Sentence Transformers, LangChain, and RAGAS.

2. Data Loading and Preprocessing:

   • Load a dataset from Hugging Face.
   • Chunk contexts into manageable sizes.

3. Vector Database Setup:

   • Generate dense and sparse embeddings for each context.
   • Upload embeddings to Pinecone for efficient retrieval.

4. RAG Pipeline Setup:

   • Set up LangChain with Pinecone for context retrieval.
   • Use OpenAI’s GPT model to generate answers based on retrieved contexts.

5. Evaluation:

   • Test the chatbot's responses on sample queries.
   • Calculate performance metrics using RAGAS.

Example Usage

Question and Answering

# Ask a question using the RAG pipeline
query = "What are the symptoms of diabetes?"
result = qa.invoke(query)
print("Generated Response:", result['result'])

Evaluation Example

# Evaluate RAG pipeline with sample queries and responses
eval_data = [...]  # Prepare sample evaluation data
result = evaluate(
    dataset=ragas_eval,
    metrics=[faithfulness, answer_relevancy, context_precision, context_recall]
)
print("Evaluation Results:", result)

Setup Instructions

1. Install Dependencies: Make sure to install all necessary Python libraries. Use the following command to install dependencies:

   pip install -r requirements.txt

2. API Keys:

   • Set up an API key for Pinecone in your environment variables.
   • Set up an API key for OpenAI in your environment variables if using GPT for generation.

3. Run the Notebook: Open the Jupyter Notebook and follow the code sections to execute each part of the workflow.

Dependencies

• Python 3.8+
• transformers
• datasets
• pinecone-client
• sentence-transformers
• langchain
• ragas
• matplotlib
• seaborn

Limitations

• The model's responses are limited to the quality and scope of the dataset provided.
• Requires access to Pinecone and OpenAI API for the retrieval and generation process.