feathr

Why we included this project

Hand-building training datasets from raw logs, transactions, or event streams is a recurring chore for many ML teams, and Feathr is built to take that pain away. It's a feature store where you define transformations once in Python, register them by name, and reuse the same features for both offline training and online serving, with point-in-time-correct joins that stop data leakage. Because it runs on Spark and plugs into Databricks and Azure Synapse, it handles the billion-row workloads that homegrown pipelines often choke on. Data scientists get a Pythonic API with UDF support, while platform teams get a registry, governance, and materialization for production. If your org is trying to standardize how features are built, shared, and versioned across models, this is a solid reference to study.