A fully modular, production-like Feature Store framework built from scratch — supporting batch and real-time feature engineering, metadata tracking, and feature serving.
- Feature Registry: PostgreSQL with FastAPI API for registering and managing features
- Offline Store: Parquet files (via PySpark) for batch feature generation and backfills
- Online Store: Redis for real-time feature serving
- Materialization Tracker: Auto-updates feature freshness + writes logs
- Feature Syncer: Moves batch features to Redis after materialization
- Monitoring: Prometheus + Grafana dashboard integration
- Time-Travel Retrieval: Point-in-time feature lookups for safe ML training
make up # Start Postgres + Redis + Spark
make serve # Run FastAPI on http://localhost:8000
make sync # Sync features to Redis
make test # Run all unit tests