An autonomous AI system that takes raw data and problem descriptions, then automatically develops, trains, and deploys machine learning models using Google's Gemini and Vertex AI.

This is a simplified version focused on core ML pipeline functionality:

• ✅ Automated problem analysis
• ✅ Data processing and labeling
• ✅ Model selection and training
• ✅ Real-time progress tracking
• ✅ Model deployment and testing

.
├── backend/                 # FastAPI backend
│   ├── app/
│   │   ├── api/            # API endpoints
│   │   ├── core/           # Config and core utilities
│   │   ├── schemas/        # Pydantic data models
│   │   ├── services/       # Business logic
│   │   │   ├── agent/      # AI agent components
│   │   │   └── cloud/      # GCP integrations
│   │   └── utils/          # Helper utilities
│   ├── tests/              # Backend tests
│   └── pyproject.toml      # Python dependencies
│
├── frontend/               # React frontend
│   ├── src/
│   │   ├── components/     # React components
│   │   ├── pages/          # Page components
│   │   ├── hooks/          # Custom hooks
│   │   ├── services/       # API client
│   │   └── contexts/       # React contexts
│   └── package.json        # Node dependencies
│
├── docs/                   # Documentation
│   └── guides/             # Setup guides
│
└── docker-compose.yml      # Local development setup

• Python 3.11+
• Node.js 18+
• Docker & Docker Compose
• Google Cloud account with billing enabled

Follow the detailed guide: docs/guides/google-cloud-setup.md

Quick version:

# Enable required APIs
gcloud services enable aiplatform.googleapis.com storage.googleapis.com

# Create service account and download key
gcloud iam service-accounts create agentic-platform-sa
gcloud iam service-accounts keys create gcp-key.json \
  --iam-account=agentic-platform-sa@YOUR_PROJECT_ID.iam.gserviceaccount.com

# Move key to credentials directory
mkdir -p gcp-credentials
mv gcp-key.json gcp-credentials/

# Copy example env file
cp .env.example .env

# Edit .env and update:
# - GOOGLE_CLOUD_PROJECT
# - GCS_BUCKET_NAME
# - GEMINI_API_KEY

# Start all services
docker-compose up

# Backend will be available at http://localhost:8000
# Frontend will be available at http://localhost:3000

cd backend
poetry install

cd backend

# Install dependencies
poetry install

# Run development server
poetry run uvicorn app.main:app --reload

# Run tests
poetry run pytest

# Format code
poetry run black .
poetry run ruff check .

cd frontend

# Install dependencies
npm install

# Run development server
npm run dev

# Run tests
npm test

# Build for production
npm run build

See .kiro/specs/agentic-model-training-platform/tasks.md for detailed task list.

Completed:

• ✅ Project infrastructure setup
• ✅ Docker configurations
• ✅ Environment configuration
• ✅ GCS storage structure

In Progress:

• 🔄 GCS storage manager and Pydantic schemas
• 🔄 Core agent components
• 🔄 API endpoints
• 🔄 Frontend UI

1. Automated Problem Analysis: Upload data and describe your problem - the system analyzes and determines the best ML approach
2. Smart Data Processing: Automatic data validation, cleaning, and feature engineering
3. Intelligent Labeling: Zero-shot labeling for unlabeled datasets using Gemini
4. Model Selection: AI-powered model architecture and hyperparameter recommendations
5. Automated Training: Hands-off training on Vertex AI with progress monitoring
6. Real-time Updates: WebSocket-based live progress tracking
7. Easy Deployment: One-click model deployment with API endpoints
8. Interactive Testing: Test your models directly in the UI

Backend:

• FastAPI (Python web framework)
• Google Cloud Storage (Metadata and file storage)
• Celery + Redis (Task queue)
• Google Vertex AI (Model training)
• Google Gemini (AI agent)
• Pydantic (Data validation)

Frontend:

• React 18 + TypeScript
• Material-UI (Components)
• React Query (Data fetching)
• React Router (Routing)
• Vite (Build tool)

Infrastructure:

• Docker + Docker Compose
• Google Cloud Storage
• Google Cloud Logging

Once the backend is running, visit:

• Swagger UI: http://localhost:8000/docs
• ReDoc: http://localhost:8000/redoc

# Check if services are running
docker-compose ps

# View logs
docker-compose logs backend
docker-compose logs redis

# Verify credentials file exists
ls -la gcp-credentials/gcp-key.json

# Test authentication
gcloud auth activate-service-account --key-file=gcp-credentials/gcp-key.json

# Stop all containers
docker-compose down

# Check what's using the port
lsof -i :8000  # or :3000 for frontend

• Google Cloud Setup Guide
• Design Document
• Requirements Document

This is a hackathon project. Feel free to fork and experiment!

MIT License - feel free to use this for your own projects.