Data & AI

Dual-Store Feature Store Design

Data & AI Pipelines diagram · made in Skemio

Architecture diagram of Feast feature store: raw data flows to an offline batch store for training and syncs via materialization to an online Redis store for low-latency inference.
Architecture diagram of Feast feature store: raw data flows to an offline batch store for training and syncs via materialization to an online Redis store for low-latency inference.
Open it in SkemioMake your own

What it shows

  • Offline stores (BigQuery/S3) compute batch features for model training.
  • Online stores (Redis/DynamoDB) serve low-latency features in <10ms.
  • A registry ensures both stores share identical schemas and definitions.

The takeaway

A dual-store feature store guarantees consistent feature logic from training to inference.

More data & ai pipelines diagrams