Machine Learning Engineer interview
at Meta.

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 fast and had to decide what modeling complexity to cut to hit the deadline.
2. 02 Describe a time a model performed well offline but failed in production, and how you diagnosed and owned the fix.
3. 03 Tell me about a disagreement with a partner team over a ranking metric and how you resolved it.
4. 04 Describe the most technically ambitious ML project you drove and the measurable impact it had.

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 and delayed/biased labels when training a model to predict ad click-through rate?
2. 02 Explain the trade-offs between a lightweight model for low-latency inference and a heavier ensemble for accuracy in a real-time ranking system.
3. 03 How would you design features and prevent training-serving skew for a model ranking content in Feed?
4. 04 How would you detect and mitigate feedback loops where a recommendation model's own outputs bias the training data it later learns from?

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

1. 01 Design the News Feed ranking system, covering candidate generation, ranking, and online serving under tight latency constraints.
2. 02 Design Instagram's Explore recommendation system, including retrieval and ranking stages.
3. 03 Design an ads ranking and evaluation framework that optimizes for both relevance and long-term value.
4. 04 Design a 'People You May Know' friend recommendation system at Facebook scale.