Problem Understanding
Restate the problem in your own words.
Design Netflix-style Recommender System
Design a Netflix-class recommender system: produce a personalised, ranked top-K of items for any user in any context (home, 'because you watched X', search re-rank) at sub-200 ms p95. The architecture is two-tower retrieval (user + item embeddings, nearest-neighbour search in a vector DB) feeding a multi-stage re-ranker that scores ~1 000 candidates with rich real-time + batch features. An online learning loop closes the feedback: clicks, plays, skips, dwell time flow back into the user model within minutes. Decisive trade-offs are sync vs async re-rank, online vs offline training, and how to balance exploration against exploitation without hurting near-term engagement.
- NetflixFoundational recommender at petabyte scale; ~270 M subscribers; multiple recommender models per surface.
- YouTubeRecommender drives the watch page; two-tower retrieval + ranker is the public reference design.
- TikTok / PinterestRecommender IS the product; ranking model dominates everything else.
- Amazon'Customers who bought this' is the canonical e-commerce recommender at hyperscale.
Your task: read the problem above, then write what the system is, who uses it, the rough scale, and the headline UX expectation — in your own words. Submit for AI review when you're ready.
Click any step in the sidebar to jump around — sections don't have to be done in order. Press ? any time to see all shortcuts.