A full-stack application for generating synthetic data using weak supervision, generative models, and AI agents.

# 1. Run installation script
./install.sh

# 2. Create .env file with your NVIDIA API key
nano .env
# Add: NVIDIA_API_KEY=nvapi-your-key-here

# Start all services (Frontend + Backend + Agent)
./start.sh

# Open in browser: http://localhost:8080

# Stop all services
./stop.sh

If you prefer to set up manually:

# Create virtual environment
python3 -m venv venv
source venv/bin/activate

# Install dependencies
pip install -r backend/requirements.txt

# Create .env file
echo "NVIDIA_API_KEY=your_key_here" > .env

# Terminal 1: Start Flask API (Port 5000)
python backend/flask_api.py --port 5000

# Terminal 2: Start Orchestrator (Port 8000)
python backend/agents/serve_orchestrator.py

# Terminal 3: Start Frontend (Port 8080)
cd frontend
npm install
npm run dev

• Frontend (Port 8080): React + TypeScript + Vite
• Flask API (Port 5000): Direct tool access for labeling, training, generation
• Orchestrator (Port 8000): AI agent for natural language workflows

Backend:

• Flask - RESTful API
• FastAPI - Agent orchestrator
• PyTorch - Deep learning models
• Snorkel - Weak supervision labeling
• LangChain + LangGraph - Agent orchestration
• NVIDIA NIMs - LLM services

Frontend:

• React + TypeScript
• Vite - Build tool
• TanStack Query - Data fetching
• Shadcn UI - Component library
• Tailwind CSS - Styling

• Backend API: See backend/FLASK_API_README.md
• Agent Setup: See backend/AGENT_SETUP.md
• Quick Reference: See backend/API_QUICK_REFERENCE.md

Create a .env file in the project root:

# Required for Agent Mode
NVIDIA_API_KEY=nvapi-xxxxxxxxxxxxx

Get your API key from NVIDIA Build

• Upload Dataset: CSV or Parquet files
• Data Labeling: Weak supervision with Snorkel
• Model Training: CVAE, CTGAN, VAE+GMM models
• Data Generation: Generate synthetic samples
• Export: Download as Parquet or CSV

• Natural Language: Describe your goal in plain English
• Automated Workflow: Agent orchestrates all steps
• Self-Correcting: Adjusts hyperparameters automatically
• Multi-Step Tasks: Handles complex workflows

HackUTDNVIDIA/
├── backend/
│   ├── agents/                 # AI agent pipelines
│   ├── DataFoundry/           # Core data processing
│   ├── testing_data/          # Sample datasets
│   ├── output_data/           # Generated outputs
│   ├── flask_api.py           # Main API server
│   └── requirements.txt       # Python dependencies
├── frontend/
│   ├── src/
│   │   ├── components/        # React components
│   │   ├── pages/            # Page components
│   │   ├── lib/              # API client & utilities
│   │   └── hooks/            # Custom React hooks
│   └── package.json          # Node dependencies
├── logs/                      # Service logs
├── install.sh                # Installation script
├── start.sh                  # Start all services
├── stop.sh                   # Stop all services
└── .env                      # Environment variables (create this)

View logs for each service:

# Flask API
tail -f logs/flask.log

# Orchestrator
tail -f logs/orchestrator.log

# Frontend
tail -f logs/frontend.log

# Activate venv
source venv/bin/activate

# Test backend API
cd backend
python -m pytest

# Test agents
python -m agents.test_agents --test all

# Check what's using a port
lsof -ti:5000  # Flask
lsof -ti:8000  # Orchestrator
lsof -ti:8080  # Frontend

# Kill process on port
lsof -ti:5000 | xargs kill -9

1. Make sure virtual environment is activated
2. Check all dependencies are installed
3. Verify .env file exists with API key
4. Check logs for specific errors

1. Verify Flask is running on port 5000
2. Check CORS is enabled in Flask
3. Verify VITE_BACKEND_URL in frontend/.env is set to http://localhost:5000

[Add your license here]

[Add contributors here]

• NVIDIA for NIM services
• Snorkel AI for weak supervision
• LangChain for agent framework