Difficulty: 8/10collaborative filtering, real-time serving, cold startCollaborative filtering and matrix factorization-based architecture with a distributed database and a caching layer, handling 100,000 user requests per hourScenario Questions0-2 years experience1How would you build a simple 'customers also bought' feature on a product page using only purchase history data?2Given a CSV of user IDs and product IDs they bought, what steps would you take to generate the top‑3 recommendations for a specific user?3What would happen if you tried to compute recommendations on the fly for every request without any caching?2-5 years experienceFollow-up QuestionsWhat metrics would you track to evaluate recommendation quality?How would you ensure the system remains resilient to traffic spikes?Can you describe your approach to rolling back a model that underperforms?Suggest Improvement
