Skip to content
ML Academy

Production and system design

Interview question

  • seniorlevel
  • 35 minsuggested time
  • 1ML engineer

Whiteboard case. A payments-risk team and a recommendations team both want "fresh" per-user features, such as transactions in the last hour and average basket in the last 30 days, at serving time within a 50 ms total latency budget and 3,000 requests per second, while data scientists train on the same definitions. Design the feature pipeline and serving layer.

Tests: Designing batch + streaming features with training-serving consistency, point-in-time training data, an online store, freshness and latency budgets.

day 19day 9day 2day 58

Sign in to practise: answers are saved before model answers are shown.