AI Medical Consultant

Overview

The AI Medical Consultant is an advanced chatbot designed to provide accurate medical information by querying a robust knowledge base derived from "The GALE Encyclopedia of Medicine." This project utilizes cutting-edge NLP techniques and frameworks to deliver reliable and instant medical guidance.

Technical Architecture

Data Ingestion

• Vector Database Creation: The textual data from "The GALE Encyclopedia of Medicine" PDF is transformed into a searchable vector database using FAISS. This process is facilitated by the Hugging Face's sentence-transformers/all-MiniLM-L6-v2 model, optimized for running on CPU environments.
• Storage: The generated embeddings are stored in a structured directory for efficient retrieval.

Model Configuration

• Language Model: The project employs the quantized version of LLaMA-2, specifically llama-2-7b-chat.ggmlv3.q3_K_M.bin, configured with a reduced temperature setting for generating coherent and contextually appropriate responses.
• Custom Prompting: A tailored prompt template enhances the query handling by structuring user inputs and model responses effectively.

Interface and Interaction

• Data Chunking: Utilizing RecursiveCharacterTextSplitter from the Langchain framework, the data is segmented into manageable chunks that ensure comprehensive coverage of the medical encyclopedia content.
• Retrieval-Augmented Generation: A RetrievalQA chain enhances the chatbot's response accuracy by dynamically sourcing relevant information from the vector database.

Application

• Web Interface: The chatbot is accessible via a web-based application built using Chainlit, offering users a seamless and interactive platform to inquire about medical knowledge.

Setup and Deployment

Requirements

• Python 3.8+
• Libraries: Langchain, CTransformers, FAISS, Chainlit, sentence_transformers
• Download the suitable quantized LLM based on your RAM capacities from Hugging Face: Llama-2-7B-Chat-GGML

Installation

1. Clone the repository:

   git clone [repository-link]
   

2. Install dependencies:

   pip install -r requirements.txt
   

3. Run the ingest.py file first: to create vector database

   python ingest.py
   

4. Launch the application on localhost:

   chainlit run model.py -w