Skip to content

Real-Time Feature Store for ML

A unified online/offline feature store with streaming ingestion, point-in-time correctness, and feature governance.

PythonAWSKafkaPostgreSQL
Abstract near-black and electric cyan cover illustration for the Real-Time Feature Store project, showing online and offline feature planes joined by cyan bus lines.

Highlights

  • Point-in-time correct joins for training/inference parity.
  • Streaming ingestion with schema registry and drift detection.
  • Feature-level ownership, SLAs, and access controls.

Outcomes

  • Cut new-feature time-to-production from weeks to days.
  • Eliminated a class of training-serving skew bugs across 20+ models.

Stack

  • Python
  • AWS
  • Kafka
  • PostgreSQL
  • Redis