Mock Interviewer
Ask Coach
Tap a question to start, or type your own below.
Recommendation System Design
Visual Problem Diagram
Scenario
Homepage loads in 180 ms but recommendations are 6 hours stale—product accepts that trade until a viral item never surfaces and revenue dips 3%. The interview separates offline training, online serving, and exploration without designing a full ML course.
Design a recommendation system that personalizes content for users on a homepage or feed. Production systems are two-stage (retrieve then rank) with strict serving latency and honest staleness SLAs.
You should support offline model training, candidate generation, real-time ranking, feedback collection, and A/B hooks. Be ready to explain cold start, ANN retrieval, and exploration.
Constraints
Collect events, train models offline, generate candidates, rank, serve recommendations, experiment hooks
< 100 ms serve p95, refresh staleness minutes–hours acceptable if stated, high availability
100M users, billions items, millions QPS aggregate retrieve, terabyte feature store
Stages ahead
Log in to start practicing
Preview the problem below. Sign in to start your timed session, save progress, and use the AI coach.