EL_Metabtabaty: Mental Health Chatbot

EL_Metabtabaty is a mental health chatbot designed to provide support, resources, and empathetic conversations. Developed as part of a graduation project for the Samsung Innovation Campus, this chatbot uses advanced AI techniques to engage with users on mental health topics. EL_Metabtabaty operates with two core approaches: Retrieval-Augmented Generation (RAG) and Fine-Tuned Llama 3.2. The app is deployed on Streamlit, offering users an intuitive interface to access its services.

Table of Contents

• Overview
• Features
• Data Preprocessing
• Project Structure
• Installation
• Usage
• Contributing
• License

Overview

EL_Metabtabaty is designed to assist users by providing information, support, and empathetic responses related to mental health topics. This chatbot uses two main approaches:

• RAG Model: Integrates document retrieval and generation, allowing for contextually rich, accurate answers based on embedded documents.
• Fine-Tuned Llama 3.2: A fine-tuned conversational model, optimized with techniques like LoRA, QLoRA, and Gradient Checkpointing for high-quality and efficient dialogue generation. dataset link : https://www.kaggle.com/datasets/thedevastator/nlp-mental-health-conversations

Features

RAG Model (Deployed on Streamlit)

The deployed RAG model includes the following features:

• Session History: Maintains chat history across sessions for a personalized experience.
• Voice Input: Supports voice input, making interactions accessible.
• User-Friendly Interface: Streamlit-based UI that’s smooth and easy to use.

Fine-Tuned Llama 3.2

The alternate configuration uses a fine-tuned version of Llama 3.2, optimized with memory-efficient techniques, making it suitable for more resource-limited environments.

Project Structure

• El_Metabtabaty.py
• LICENSE
• README.md
• data
• Llama3.2_Finetuning_Chatbot.ipynb

Data Preprocessing:

• Data Ingestion: CSV file reading for mental health conversations
• Text Splitting: Splits conversations into manageable chunks for processing
• Embedding: Hugging Face all-MiniLM-L6-v2 embeddings for creating vector representations
• VectorStore DB: FAISS, used to store and query embeddings efficiently

Installation

• Clone this repository:

git clone https://github.com/ahmedomer13218/Mental-Health-Chatbot

cd Chatbot_mentalHealth_directory

Usage

• Run the Streamlit app to interact with the chatbot :

streamlit run EL_Metabtabaty.py

Contributing

We welcome contributions! If you'd like to contribute, please fork the repository and make your changes. After testing, submit a pull request for review.

License

This project is licensed under the MIT License - see the LICENSE file for details.

EL_Metabtabaty aims to be a valuable resource in mental health support by leveraging advanced AI. We are committed to making this chatbot accessible, reliable, and effective in promoting mental well-being.