Feast as a Cloud Native Data System: Bridging Batch Processing and Online Serving...
summary
Feast is presented as a Kubernetes-native feature store that materializes computed features from data lakes/warehouses into online stores using Spark, Ray or Flink, and exposes them via SDK/APIs and Feast Operator as a single CR for deployment. Integrations include Snowflake, BigQuery, Redshift, Prometheus, OpenTelemetry and OpenLineage; it supports embeddings and MCP for RAG use cases and claims 2 ms latency at ~10 entity rows per request and 500+ RPS while providing RBAC, freshness metrics and autoscaling via HPA/KEDA.
AI-written summary · about
chapters (14)
- 0:00Feast and data challenge
- 1:05Data versus ML models
- 5:24Feature store importance
- 8:35Feast architecture
- 13:14ML lifecycle and governance
- 15:00Feast on Kubernetes
- 16:45Observability with Prometheus
- 20:15Scalability benchmarks
- 22:26RBAC and compliance
- 24:15Data processing with Spark Ray
- 26:03Feast for GenAI
- 27:52Feast in Open Data Hub
- 29:19Data quality and MLflow
- 30:24Closing and resources
related talks
32mBuilding a Cloud Native Feature Store with Feast on Kubeflow
27mFeast: feature store for Machine Learning
36mIntegrating Feast Online Feature Store with KFServing
51mSelf-serve Feature Engineering Platform Using Flyte and Feast
29mIntegrating High Performance Feature Stores with KServe Model Serving
46mScaling ML Pipelines with Feast, Ray and Kubeflow
33mFeathr, an Open Source, Battletested, Scalable and Enterprise-Grade Feature Store
23mFeathr: An Enterprise-Grade, High Performance Feature Store