To reduce the time spent managing feature preparation pipelines, LinkedIn developed Feathr; a feature store designed to aggregate time-sensitive data from disparate sources, compute features over historical time series data while preserving point-in-time-correctness, and persist features to online data stores for low-latency online inference. Recently, LinkedIn OSS’ed this system, which also eliminates online-offline inconsistencies and enables feature reuse across multiple models. With Feathr, ML practitioners can more easily define and register features and offer a consistent way to compute, serve, and access them by name from within ML workflows.