Video Recommendation

ML system-design round: architect a video recommendation system covering candidate generation, ranking models, and serving. The interviewer pushes hard on the infrastructure side — how services are split, how data flows end to end, and especially event logging, pipelines, and observability that power the feedback loop, often spending more time here than on the model itself. Video Recommendation ML system-design round: architect a video recommendation system covering candidate generation, ranking models, and serving. The interviewer pushes hard on the infrastructure side — how services are split, how data flows end to end, and especially event logging, pipelines, and observability that power the feedback loop, often spending more time here than on the model itself. MLE SWE system-design mlsd recommendation ranking retrieval ml-infra metrics data-engineering personalization Frequency Single report Last asked 2026-03-22 Stage onsite-system-design Video Recommendation Problem Statement Design a video recommendation system. Walk through how candidate generation, ranking, serving, and feedback loops work end to end, and how user interactions are logged so the system can continuously improve. Areas to Cover Candidate generation — retrieve a candidate set from the video corpus (embedding-based / two-tower retrieval, recency and engagement-based sources, filtering). Ranking — the scoring model(s) that order candidates for a given user, the features consumed, and how multiple objectives are balanced. Serving — how a request flows through the stack at request time and how latency is kept bounded. Feedback loops — how observed engagement is fed back to continuously improve model quality. Interviewer Emphasis: Infra, Data Flow, and Logging Expect the round to spend more time on the infrastructure side than on the model itself. Be ready to go deep on: Service split — how the recommendation stack is decomposed into services and where the boundaries sit. Data flow — how data moves through the system from user interaction to model training and back to serving. Event logging & observability — how user interactions are collected as events, the pipelines that carry them, and how the resulting logs feed metrics, monitoring, and the training/feedback loop. This is the central deep-dive of the round.