Online feature stores provide precomputed feature stores to real-time AI systems by clients requesting features using entity IDs (primary keys).

You need, however, to decide what columns go into which tables in your feature store. The design of the table schemas is informed by the features the real-time AI systems need to retrieve at runtime and what entity IDs (user ID, a session ID, credit card number, store ID, etc) are available in those real-time AI systems to retrieve those features. There is also a desire to design the online tables such that you can reuse its features across many models, so that each model does not require its own (denormalized) online table(s).

Existing feature stores support the Star Schema data model for designing online tables. This limits the features you can retrieve to the entity IDs available in the online client. In this talk, we describe how we added support for Snowflake Schema Data Model to Hopsworks 4.0, which enables the use of more features than just those associated with the entity IDs, through the use of foreign keys in online tables.