Design YouTube's video recommendation system that serves 2 billion users and selects the best 10-20 videos for each user's homepage feed from a catalogue of 500 million videos.
Requirements:
- •Personalised recommendations for each user
- •Serve recommendations in ≤ 100ms
- •Maximise watch time while maintaining creator equity
- •Handle cold start for new users and new videos
- •Support A/B testing of ranking models
What you'll be assessed on
Two-stage retrieval + ranking architecture, cold start solutions, feature engineering, exploration vs exploitation, and how to evaluate recommendation quality.