feast

Why we included this project

Feast is an open source feature store for teams wiring raw analytic data into model training and live inference without letting the two drift apart. Define a feature view once, and Feast serves it from an offline store for batch training and a low-latency online store for real-time prediction, so production sees the same values the model learned on. Point-in-time correct retrieval stops future data from leaking into historical training sets, which silently skews a model. ML platform teams and data engineers who own many models get a single data access layer instead of one hand-rolled feature pipeline per project. It works with the warehouse or batch infrastructure you already run, such as Snowflake or Redshift, so adopting it does not mean standing up new storage.