Lilly Tech Systems
Build an ML Feature Platform
Build a production-ready ML feature platform from scratch. Learn feature definitions, offline and online stores, real-time serving, data quality monitoring, and drift detection — all with full working code.
Project Build Path
Follow these lessons in order to build the complete project step by step, or jump to any section you need.
1. Project Setup
Architecture overview, Feast and Redis setup, tech stack, and scaffolding.
2. Feature Definitions
Feast entities, feature views, data sources, and feature engineering.
3. Offline Feature Store
Batch computation, point-in-time joins, and historical features.
4. Online Feature Store
Redis materialization and low-latency feature serving.
5. Feature Serving API
FastAPI endpoint for real-time and batch feature retrieval.
6. Feature Monitoring
Data quality checks, drift detection, and freshness alerts.
7. Enhancements
Streaming features, governance, team onboarding, and FAQ.
What You Will Build
By the end of this project, you will have a fully functional application that can:
Define Features Declaratively
Use Feast to define entities, feature views, and data sources in Python code.
Serve Features Online & Offline
Materialize features to Redis for low-latency serving and PostgreSQL for batch training.
Build a Feature API
Create a FastAPI service that serves features in real-time for model inference.
Monitor Feature Quality
Detect data drift, track freshness, and alert on quality issues automatically.