Design Streaming Home Personalization
Combine cross-device Continue Watching with location- and history-informed recommendations under clear freshness and fallback rules.
Problem statement
Design the personalized home experience for a large streaming product. It must show Current/Continue Watching items that a viewer has started and can resume, retaining those entries for a configurable number of days, and Recommended Shows informed by coarse location, prior watch history, and other eligible signals.
Cover client-side caching, database selection and data modeling, how progress is stored and retrieved, how recommendations are generated, and service interactions.
For this exercise, distinguish correctness-sensitive playback progress from best-effort ranking, support multiple devices per profile, and degrade usefully when personalization is unavailable. Video delivery, encoding, billing, ad serving, and the internal mathematics of a recommendation model are outside scope.
What to build
- Record replay-safe playback progress and completion from multiple devices.
- Return resumable titles in recent-activity order with configurable retention.
- Return eligible ranked shows using watch history, coarse location, and other consented signals.
- Serve safe cached or non-personalized rails when feature or ranking systems are unavailable.
At scale
- Assume 100 million monthly profiles, 10 million concurrent viewers at peak, and up to 500,000 playback progress events per second.
- Assume the home response should complete within 250 ms at p95 while accepted progress should appear on another device within five seconds at p95.
- Assume Continue Watching retains a title for a configurable X days after the last qualifying activity and must resist delayed offline events.
- Assume recommendations may be eventually consistent, but catalog, market, and entitlement eligibility must be enforced at read time.