Reviewing a Feature Pipeline PR and Serving Features by Freshness

Reported in interviews at Netflix

Problem statement

How would you review a churn feature pipeline for production risks? Its features feed both training and online scoring, and the pull request passes tests on small fixtures. Then explain how to serve features whose freshness needs range from slow aggregates to live activity and request-dependent values.

Review the pipeline, choose serving tiers and handle stale features.

Show hint

Unit tests run on small, clean fixtures frozen in time; list the properties of real production data that such fixtures usually lack.

Relevant roles: ML Engineer, MLOps Engineer and Applied Scientist.

Related free problems