MLOps & Infrastructure · intermediate · concept 214 of 311 · Updated Nov 4, 2026
Feature Stores
A feature store is a shared system that computes, stores and serves the inputs machine learning models use, so training and live prediction see the same values. It prevents a common bug where a feature is calculated one way in training and another in production. It earns its cost once several teams or models reuse the same features.
Key terms
Training-serving skewOnline and offline storeFeature reusePoint-in-time correctnessLineage
Learn these first
Where you meet it in the real world
Fraud detection, recommendations, real-time pricing
Guides and articles