🧠 Suicide Mitigation Chatbot

An AI-powered chatbot designed to identify emotional distress and engage users in empathetic, human-like conversations. Built to support mental health, especially in critical situations, it uses an ensemble of Microsoft's DialoGPT and a custom Seq2Seq LSTM model.

---

📁 Project Directory (as on your system)

D:\Git\Sucide_Mitigation_Chatbot
├── .ipynb_checkpoints/
├── .venv/                         ← Python virtual environment
├── Data.csv                       ← Dataset used for training
├── Ensembled_Model.ipynb          ← Notebook for ensembling models
├── model/                         ← Folder containing the trained model files
├── model.zip                      ← Zipped model archive (836 MB)
├── Review_1.pptx                  ← Project presentation
├── UI/                            ← React frontend + Flask backend
├── Untitled-1.ipynb               ← Miscellaneous notebook
├── Zeroth_Review.pptx            ← Initial review slides

---

🚀 Getting Started

1. Start the Flask Server

Navigate to the backend directory:

cd UI/public/flask-server

Run the server:

python server.py

Ensure all required packages are installed:

pip install -r requirements.txt

---

2. Start the React Frontend

In a new terminal, run:

cd UI
yarn
yarn dev

The app will start on http://localhost:3000.

---

🤖 Model Overview

This chatbot uses an ensemble of two models for generating emotionally rich and context-aware responses:

Model	Description
DialoGPT	A pretrained transformer-based conversational model by Microsoft
Custom Seq2Seq	A bidirectional LSTM encoder-decoder trained on emotion-focused dialogue

---

🧱 Custom Seq2Seq Architecture

Component	Configuration
encoder_embedding	Embedding layer — 30,522 tokens, 256-dim, padding_idx=0
encoder_lstm	BiLSTM — input: 256, hidden: 512, batch_first=True
decoder_embedding	Embedding layer — 30,522 tokens, 256-dim, padding_idx=0
decoder_lstm	LSTM — input: 256, hidden: 1024, dropout: 0.4, batch_first=True
output_layer	Linear — in_features=1024, out_features=30522
dropout	Dropout with p=0.4

🔁 Bidirectional encoding captures rich context; decoder uses higher capacity for expressive responses.

---

💡 Features

• 💬 Deep conversations powered by LSTM + DialoGPT
• 🌐 Flask API for model inference
• ⚛️ React-based modern UI
• 🧠 Focus on empathy and mental health

---

📊 Resources

• 📁 Dataset: Data.csv
• 📄 Presentations: Review_1.pptx, Zeroth_Review.pptx
• 📚 Notebooks: Ensembled_Model.ipynb

---

A detailed explanation is published here

---

🤝 Contributing

Pull requests and feedback are welcome. Together, we can build technology that cares. ❤️