COMPEL Glossary / feature-store
Feature Store
A feature store is a centralized, managed repository for storing, versioning, and serving the processed data features (engineered variables) used to train and run AI models, enabling feature reuse across teams, ensuring consistency between training and serving environments, and reducing the redundant data processing that occurs when each team independently creates the same features.
What this means in practice
For organizations scaling AI beyond a few models, a feature store prevents the common problem where different teams create slightly different versions of the same feature, leading to inconsistent model behavior and wasted engineering effort. In COMPEL, the feature store is assessed as part of the Technology pillar's AI platform maturity during Calibrate and represents a key infrastructure component of the enterprise AI platform designed during Module 3.3, Article 2.
Why it matters
Without a feature store, different AI teams independently create slightly different versions of the same data features, leading to inconsistent model behavior, wasted engineering effort, and training-serving skew that degrades production performance. Feature stores enable reuse, consistency, and governance of the processed data variables that AI models depend on, reducing redundant work while improving reliability.
How COMPEL uses it
The feature store is assessed as part of the Technology pillar's AI platform maturity during Calibrate. During Model, it is designed as a key infrastructure component of the enterprise AI platform in Module 3.3, Article 2. The Produce stage implements the feature store with versioning and governance controls. The Evaluate stage measures feature reuse rates and consistency between training and serving environments.
Related Terms
Other glossary terms mentioned in this entry's definition and context.