ML systems

Feature Platform

Real-time feature computation and multi-model serving.

How it works
  1. Kafka
  2. Feast
  3. Redis
  4. Ray Serve

The problem

Every model team ends up rebuilding the same features, with different freshness and different bugs, and serving each model on its own stack.

How it works

Kafka streams transaction events to feature processors that compute real-time features and store them in Redis through Feast, with online and offline serving. Multiple models share those features. Ray Serve handles deployment with A/B testing and traffic splitting, Prometheus monitors it, and MLflow tracks experiments. A FastAPI service sits in front.

03 /

Key decisions

  • One feature store, many models

    Fraud detection, recommendation and forecasting models share the same feature infrastructure instead of each computing their own.

  • A/B testing in the serving layer

    Ray Serve splits traffic between model versions, with Prometheus metrics and MLflow experiment tracking.

04 /

Results

Events / sec
10K
Feature lookup
5–10 ms
Inference
20–50 ms