Feature Store

A feature store is a centralized repository for storing, managing, and serving the engineered features used in machine learning models. It bridges the gap between raw data (from pipelines and APIs) and the model training and inference systems that need computed, model-ready features.

Features are derived data columns computed from raw data - for example, 'average price per sqft in the property's neighborhood over the last 90 days' is a feature derived from raw listing data. Computing this feature identically for training and production is critical; a feature store ensures consistency.

Without a feature store, feature computation code is often duplicated between the training pipeline and the real-time inference service, leading to training-serving skew - where the model sees different values in production than it was trained on. Feature stores solve this by centralizing feature definitions and serving them from a shared platform.