Modern AI and GenAI platforms running on Kubernetes depend on derived data that must be processed in batch and served with low latency. Teams often stitch together data processing engines, object stores, databases, and caching layers, creating operational complexity and inconsistent data access patterns.

This session presents Feast as a cloud native data system that bridges batch processing and online serving on Kubernetes. We explore how Feast integrates with Kubernetes-native data engines such as Spark and Ray, reuses existing data lakes and warehouses as offline stores, and materializes data into online stores optimized for real-time access.

Using Open Data Hub and the Feast Operator, the talk demonstrates a Kubernetes-native architecture for managing data pipelines. Attendees will learn practical design patterns and operational trade-offs for building scalable, cloud native data platforms that support ML and RAG workloads without introducing new data silos.