If features created yesterday exist only in a notebook dataframe, the next project must create the same features again. Worse, transformations used during training might subtly differ from those used during inference, silently destroying model performance. SageMaker Feature Store is a dedicated repository solving both problems: feature reuse and training/inference consistency.
In MLA-C01 exams, Feature Store appears with keywords like "online/offline store differences," "preventing training/serving skew," "feature sharing." Today we cover three axes: structure, two stores, and consistency.
Feature Store is a central repository storing, managing, and serving ML features