Medical Chatbot

This repository contains a Python application that implements a conversational chatbot for medical assistance. The chatbot is built using the LangChain library and is powered by a pre-trained LLaMA (Large Language Model Adapted) model. The application ingests and processes medical documents, creates a vector store, and provides a Streamlit-based user interface for interacting with the chatbot.

Features

• Ingests and processes PDF documents from a specified directory
• Splits the documents into smaller text chunks for efficient processing
• Creates embeddings for the text chunks using a HuggingFace sentence embeddings model
• Stores the text chunks and their embeddings in a FAISS vector store
• Utilizes a pre-trained LLaMA model for generating responses
• Maintains conversation context using a memory component
• Provides a text input field and an audio recorder for user queries
• Displays the chat history in the Streamlit application

Directory Structure

• app.py: Contains the Streamlit application code for the chatbot interface.

• ingest.py: Handles the ingestion and preprocessing of PDF documents.

• model.py: Defines the main components of the chatbot, including the language model, embeddings, and the retrieval-based question-answering chain.

• langchain.ipynb: A Jupyter Notebook that demonstrates the core functionality of the chatbot in an interactive manner.

• data/: Directory containing the PDF documents to be ingested (not included in the repository).

• vectorestores/: Directory where the FAISS vector store is saved (created during runtime).

  Code Details

app.py

This file sets up the Streamlit application for the chatbot.

• Loads the FAISS vector store and embeddings from the local path.
• Creates an instance of the LLaMA language model using CTransformers.
• Initializes a ConversationBufferMemory to keep track of the chat history.
• Sets up the ConversationalRetrievalChain to handle the conversational flow, combining the language model, vector store retriever, and memory.
• Defines the Streamlit application with a text input field and an audio recorder for users to ask questions.
• When a question is submitted, calls the conversation_chat function to generate an answer using the ConversationalRetrievalChain.
• Displays the chat history in the Streamlit app using the streamlit_chat package.

ingest.py

This file handles the ingestion and preprocessing of the data (PDF documents in the data/ directory).

• Loads the PDF documents using the DirectoryLoader and PyPDFLoader from LangChain.
• Splits the loaded documents into smaller text chunks using the RecursiveCharacterTextSplitter.
• Generates embeddings for the text chunks using the HuggingFace embedding model.
• Creates a new FAISS vector store from the text chunks and their embeddings.
• Saves the vector store to the local path vectorestores/db_faiss.

model.py

This file defines the main components of the chatbot, including the language model, embeddings, and the retrieval-based question-answering chain.

• Loads a pre-trained LLaMA model from a .bin file and uses the CTransformers class from LangChain to interface with the model.
• Creates embeddings using the HuggingFace sentence embeddings model all-MiniLM-L6-v2.
• Loads the FAISS vector store from the local path vectorestores/db_faiss.
• Defines a custom prompt template to format the query and context before passing them to the language model.
• Sets up the RetrievalQA chain to perform the question-answering task by retrieving relevant document chunks from the vector store and generating an answer using the language model.

langchain.ipynb

This Jupyter Notebook demonstrates the core functionality of the chatbot in an interactive manner.

• Loads PDF documents from the data/ directory using the DirectoryLoader and PyPDFLoader.
• Splits the loaded documents into smaller text chunks using the RecursiveCharacterTextSplitter.
• Generates embeddings for the text chunks using the HuggingFaceEmbeddings model.
• Creates a FAISS vector store from the text chunks and their embeddings, and saves it to the local path vectorestores/db_faiss.
• Defines a custom prompt template to format the query and context.
• Creates an instance of the LLaMA language model using CTransformers.
• Initializes a ConversationBufferMemory to keep track of the chat history.
• Sets up the ConversationalRetrievalChain using the language model, vector store retriever, and memory.
• Defines the conversation_chat function to handle user input and generate responses.
• Provides an interactive loop for users to ask questions and receive responses from the chatbot.

Prerequisites

Before running the application, ensure that you have the following prerequisites installed:

• Python (version 3.7 or later)
• Required Python packages (listed in requirements.txt)
• LLaMA model files (e.g., llama-2-7b-chat.ggmlv3.q8_0.bin)

Installation

1. Clone the repository:

git clone https://github.com/Idk507/health_bot

2. Navigate to the project directory:

cd health_bot

3.Install Requirements.txt

pip install -r requirements.txt

Download the LLaMA model files (e.g., llama-2-7b-chat.ggmlv3.q8_0.bin) and place them in the project directory. Place the PDF documents you want to ingest in the data/ directory.

Usage

Run the ingestion script to preprocess the documents and create the vector store:

python ingest.py

Start the Streamlit application:

streamlit run app.py

The application will open in your default web browser, where you can interact with the chatbot by typing or recording your questions.