Lyft’s Feature Store: Architecture, Optimization, and Evolution

This post provides an updated architectural overview of Lyft's Feature Store, detailing its batch, online, and streaming components. It highlights the use of Spark SQL and JSON for batch feature ingestion, DynamoDB and ValKey for the online serving layer, and Apache Flink for streaming pipelines. The post also emphasizes improvements in user experience, feature governance through metadata and versioning, and integration with Kyte for accelerated local development workflows.