Machine Learning Engineer interview
at Google.

Tell-me-about-a-time stories on ownership, conflict, and impact — scored against this company's real values.

1. 01 Tell me about a time you shipped an ML model whose offline metrics looked strong but online performance disappointed, and what you did.
2. 02 Describe a situation where you had to balance model accuracy against latency or serving cost for a production system.
3. 03 Tell me about a time you disagreed with a data scientist or researcher on modeling approach and how you resolved it.
4. 04 Describe how you handled a model that started degrading in production after a data distribution shift.

The hard screen for the craft itself — talked through out loud, not whiteboarded in your head.

1. 01 How would you handle severe class imbalance when training a click-through-rate model for Google Ads?
2. 02 Explain how you'd choose evaluation metrics for the YouTube recommendation ranker beyond simple click-through rate.
3. 03 Walk me through diagnosing and mitigating training-serving skew in a deployed ranking model.
4. 04 Implement and explain a function that computes embeddings similarity for approximate nearest-neighbor retrieval at scale.

Open-ended design of the systems this company actually runs, with the interviewer probing your tradeoffs.

1. 01 Design the YouTube video recommendation system, including candidate generation, ranking, and re-ranking stages.
2. 02 Design a real-time model-serving pipeline for ads click prediction at Google with strict latency budgets.
3. 03 Design the ML system behind Google Photos search that lets users find images by natural-language queries.
4. 04 Design a multi-stage recommendation and ranking pipeline for Google Play app suggestions, including the feature and embedding store.