Backend ML inference service with asynchronous task processing (FastAPI + Redis + Celery), designed to handle concurrent workloads and long-running training jobs. Scalable ML Inference & Training API Service
XGenerator is a scalable backend service for training and serving XGBoost machine learning models via REST APIs. (Documents)

Backend: FastAPI (Python)
Queue: Celery + Redis
Database: Not currently used (planned PostgreSQL integration for persistent storage and scaling)
Infrastructure: Docker, AWS EC2, Nginx
ML: XGBoost

Client (HTTP Requests)
    |
    v
FastAPI API Server
    |
    v
Redis (Message Broker / Task Queue)
    |
    v
Celery Worker (Background Processing)
    |
    v
XGBoost Model (Training / Inference)

• FastAPI handles HTTP requests and validation
• Celery processes long-running tasks asynchronously
• Redis acts as a message broker between API and workers
• Docker containers isolate services for deployment

XGenerator separates API request handling from compute-heavy tasks using Celery workers.

This allows the API server to remain responsive while training jobs are processed asynchronously.

• Task queue architecture for long-running jobs
• Stateless API design for horizontal scaling
• Redis-based message broker for worker coordination
• Containerized services for flexible deployment

• Dataset Upload - CSV upload with schema inference and validation
• Asynchronous Model Training (Celery + Redis) - XGBoost classification & regression with background task processing
• Batch Prediction API - Predict on multiple rows with missing value handling
• API Key Authentication - User management with tier-based access control
• Rate Limiting - Request throttling to prevent abuse
• Automatic Model Evaluation - 80/20 train/validation split with performance metrics

• Docker & Docker Compose
• OpenAI API key (optional, for premium tier schema inference)

git clone https://github.com/YOUR_USERNAME/XGenerator.git
cd XGenerator

# Create .env file
cp .env.example .env
# Edit .env and add your OPENAI_API_KEY (optional)

docker compose up -d --build

This starts 3 containers:

# Health check
curl http://localhost:8000/health

# API Documentation
open http://localhost:8000/docs

curl -X POST http://localhost:8000/users \
  -H "Content-Type: application/json" \
  -d '{"username": "myuser", "email": "[email protected]"}'

Response:

{
  "user_id": "abc123...",
  "api_key": "sk_live_xxx...",
  "tier": "free"
}

Save your API key! It's only shown once.

curl -X POST http://localhost:8000/datasets \
  -H "X-API-Key: sk_live_xxx..." \
  -F "[email protected]" \
  -F "user_id=abc123..." \
  -F "dataset_name=My Dataset"

curl -X POST http://localhost:8000/train \
  -H "Content-Type: application/json" \
  -H "X-API-Key: sk_live_xxx..." \
  -d '{
    "user_id": "abc123...",
    "model_name": "My Model",
    "dataset_id": "ds_xxx...",
    "task_type": "classification",
    "target": "label"
  }'

Response: Returns job_id (training runs in background)

curl http://localhost:8000/jobs/{job_id}

curl -X POST http://localhost:8000/predict \
  -H "Content-Type: application/json" \
  -H "X-API-Key: sk_live_xxx..." \
  -d '{
    "model_id": "model_xxx...",
    "rows": [{"feature1": 10, "feature2": "A"}]
  }'

Full API docs: http://localhost:8000/docs

Create .env file:

OPENAI_API_KEY=sk-...      # Optional: for premium tier schema inference
OPENAI_MODEL=gpt-3.5-turbo # Optional: default model
REDIS_URL=redis://redis:6379/0
TESTING=false              # Set true to disable rate limits

For EC2/Docker deployment guide, see: projectHint_deploy.txt

XGenerator/
├── app/
│   ├── main.py              # FastAPI entry
│   ├── config.py            # Configuration
│   ├── celery_app.py        # Celery config
│   ├── routers/             # API endpoints
│   ├── services/            # Business logic
│   ├── tasks/               # Celery tasks
│   ├── models/              # Pydantic schemas
│   └── utils/               # Utilities
├── data/                    # Data storage (gitignored)
├── docker-compose.yml
├── Dockerfile
├── requirements.txt
└── ProjectHint.txt          # Internal dev documentation (Chinese)

MIT

• API Docs: http://localhost:8000/docs (Swagger UI)
• Internal Dev Guide: ProjectHint.txt (Chinese)
• Deployment Guide: projectHint_deploy.txt (Chinese)