Medical AI Assistant

This project implements a Medical AI Assistant using a combination of Natural Language Processing (NLP) techniques and deep learning models. The assistant is designed to answer medical-related questions based on user input and a context-aware retrieval system. The core functionalities include Named Entity Recognition (NER), Retrieval-Augmented Generation (RAG), and summarization.

Project Overview

The assistant leverages several pre-trained and fine-tuned models, including:

• RoBERTa NER Model: A fine-tuned version of RoBERTa Large for detecting medical entities such as age, height, and weight.
• Gemma-2 Model: A fine-tuned Gemma-2 language model, using QLoRA (4-bit quantization with LoRA parameters) for efficient inference.
• T5-Large Model: A summarization model used to generate concise summaries of the conversation between the user and the assistant.

Features

• Named Entity Recognition (NER): Detects and extracts medical entities such as age, height, and weight from the user input.
• Retrieval-Augmented Generation (RAG): Searches a FAISS-based index of medical documents to provide relevant context for answering user queries.
• Summarization: Provides a summary of the interaction, focusing on the key points discussed.

Project Structure

• Data Loading and Indexing:

  • The project uses PyPDFDirectoryLoader to load documents from a specified directory.
  • Documents are split into manageable chunks using RecursiveCharacterTextSplitter and indexed using FAISS with embeddings generated by the HuggingFaceEmbeddings model.

• NER Model:

  • The NER model is based on RobertaForTokenClassification, fine-tuned on a custom dataset of around 1000 samples.
  • The model is designed to recognize entities like weight, height, and age.

• Language Models:

  • The assistant uses a fine-tuned Gemma-2 model, implemented with 4-bit quantization for efficient performance on GPUs.
  • The T5-Large model is used for summarization of interactions.

Installation

Prerequisites

• Python 3.8 or higher
• PyTorch
• Hugging Face Transformers
• LangChain
• FAISS
• PEFT
• Hugging Face Hub

Clone the repository

git clone https://github.com/yourusername/medical-ai-assistant.git
cd medical-ai-assistant