In ML, features are used as inputs to train predictive models – a feature can be represented by a data table column or a transformation [of raw data]. A feature store is a platform that facilitates the transformation, storage, and serving of features for both training and inference. In this blog post, Tecton co-founder Mike Del Balso and Feast creator Willem Pienaar explain how feature stores enable the development of operational ML applications. They identify the key components of a feature store, including capabilities to monitor online-offline inconsistencies and to enable feature reuse.