HamChat — Domain-Specific AI Chatbot

Fine-tune a T5 model. Ask it anything. Get domain-aware answers in real-time.

---

📸 Screenshots

🏠 Landing Page

---

💬 Chat Interface

---

🔷 How It Works

 ┌──────────────┐     ┌─────────────────────┐     ┌──────────────────────┐
 │  User Query  │ ──▶ │   Custom Rule Check  │ ──▶ │  Domain Keyword Check │
 │  (typed text)│     │  (identity/greetings)│     │  (healthcare/finance) │
 └──────────────┘     └─────────────────────┘     └──────────────────────┘
                                                              │
 ┌──────────────┐     ┌─────────────────────┐                ▼
 │  React UI    │ ◀── │   Flask REST API     │ ◀── ┌──────────────────────┐
 │  HamChat     │     │   JSON Response      │     │  Fine-tuned T5 Model  │
 └──────────────┘     └─────────────────────┘     │  Beam Search (n=5)    │
                                                   └──────────────────────┘

---

✨ Features

Feature	Description
🤖 Fine-tuned T5	T5-small fine-tuned on 3,000 Healthcare & Finance Q&A pairs
💊 Healthcare Domain	Appointments, medications, flu symptoms, vaccine side effects
💳 Finance Domain	Account balance, loans, credit cards, interest rates
🧠 Custom Rules Layer	Identity, greetings and creator questions handled before the model runs
🚫 Smart Fallback	Out-of-scope questions get a helpful redirect instead of wrong answers
💬 Persistent Chat History	Conversation stays intact when navigating back to the home page
🎨 Two-Page UI	Professional landing page + dedicated chat interface
⚡ Suggested Questions	Clickable chips so users always know what to ask
🔄 Reset Chat	One-click clear to start a fresh conversation
📱 Fully Responsive	Works cleanly on desktop, tablet, and mobile screens

---

🏗️ Tech Stack

Layer	Technology	Purpose
Frontend	React 18 + CSS	UI Dashboard
Backend	Flask + Python 3.10	REST API Server
ML Model	T5-Small (Hugging Face)	Text Generation
Training	Hugging Face Trainer API	Fine-Tuning Pipeline
Framework	PyTorch	Deep Learning Backend
Tokenizer	T5Tokenizer (SentencePiece)	Text Preprocessing
Frontend Host	Vercel	Auto-deploy from GitHub
Backend Host	Hugging Face Spaces (Docker)	Flask API Hosting
Model Weights	Hugging Face Hub	Model Storage & Download

---

🚀 Getting Started

Prerequisites

• Python 3.9 or higher
• Node.js 18 or higher
• pip and npm

---

⚙️ Backend Setup

# 1. Clone the repository
git clone https://github.com/hamza0426/Domain-Specific-Chatbot.git
cd Domain-Specific-Chatbot/backend

# 2. Install Python dependencies
pip install flask flask-cors transformers torch sentencepiece

# 3. Start the backend server
python app.py

✅ Backend live at http://localhost:5000 📡 Health check at http://localhost:5000/health

---

🖥️ Frontend Setup

# 1. Navigate to frontend directory
cd frontend

# 2. Install dependencies
npm install

# 3. Start the development server
npm start

✅ Frontend live at http://localhost:3000

Note: Make sure the backend is running on port 5000 before starting the frontend.
For the live version, the frontend connects to the deployed backend on Hugging Face Spaces automatically.

---

📡 API Endpoints

Method	Endpoint	Description
GET	/	API info and available endpoints
POST	/chat	Send a query — returns model response + source tag
GET	/health	Check if the model is loaded and API is alive
GET	/suggestions	Get sample questions grouped by domain

Example Request

curl -X POST https://hamza0426-chatbot-backend.hf.space/chat \
  -H "Content-Type: application/json" \
  -d '{"query": "What are the symptoms of flu?"}'

Example Response

{
  "response": "Flu symptoms include fever, cough, sore throat, runny nose, body aches, and fatigue.",
  "source": "model"
}

---

🧠 Model Training Details

Parameter	Value
Base Model	t5-small
Dataset Size	3,000 samples
Domains	Healthcare, Finance
Unique Intents	10
Train / Val Split	80% / 20%
Epochs	10
Batch Size	8
Max Length	250 tokens
Warmup Steps	500
Weight Decay	0.01
Final Train Loss	~0.001

---

📁 Project Structure

Domain-Specific-Chatbot/
│
├── backend/
│   ├── app.py                  # Flask API — inference, custom rules, fallback logic
│   ├── requirements.txt        # Python dependencies
│   ├── Dockerfile              # Docker config for Hugging Face Spaces
│   └── chatbot_model/          # Fine-tuned T5 weights (not in GitHub)
│                               # → Loaded from Hugging Face Hub at runtime
│
├── frontend/
│   ├── src/
│   │   ├── App.jsx             # Router — switches between Home and Chat
│   │   ├── Home.jsx            # Landing page component
│   │   ├── Home.css            # Landing page styles
│   │   ├── Chat.jsx            # Chat interface component
│   │   ├── Chat.css            # Chat interface styles
│   │   ├── index.css           # Global CSS variables and reset
│   │   └── index.js            # React entry point
│   ├── public/
│   │   └── index.html
│   └── package.json
│
├── .gitignore
└── README.md

---

🌐 Deployment Architecture

 GitHub Repo
      │
      ├──── Vercel ─────────────────▶  Frontend (React 18)
      │     Auto-deploy on push         https://domain-specific-chatbot-hamza.vercel.app
      │
      └──── Hugging Face Spaces ──────▶  Backend (Flask + Docker)
            Manual push                   https://hamza0426-chatbot-backend.hf.space
                  │
                  └── Hugging Face Hub ─▶  T5 Model Weights
                                            Auto-downloaded at container startup

⚠️ Note: Hugging Face free tier may take 30–60 seconds to wake up on first request if idle.
Subsequent requests are instant once the container is warm.

---

👨‍💻 Author

Muhammad Hamza Owais

Computer Science Graduate — passionate developer continuously learning AI, automation, and new technologies. Completed Bachelor's in Computer Science with a strong focus on building practical, full-stack AI-powered applications. Always eager to explore what's next in the world of machine learning and software engineering.

---

Made with ❤️ by Hamza — Domain-Specific AI Chatbot