Skip to content
ML Academy

Production and system design

Interview question

  • seniorlevel
  • 15 minsuggested time
  • 1ML engineer

A ranking service must answer within 300 ms at the 99th percentile. For each request it calls five independent backend services in parallel (inventory, price, user profile, ML features, ad filters); each has a 99th-percentile latency of 100 ms (hypothetical). Leadership says "every dependency meets 100 ms at p99, so we are fine". Are they? How would you design the request path?

Tests: Tail-latency amplification across fan-out, deadline budgets, timeouts and fallbacks, hedging, and degrading gracefully.

day 19day 54day 58

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