Machine Learning Engineer interview
at Spotify.

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 collaborated across squads to take an ML model from prototype to production.
2. 02 Describe a situation where you had to communicate a model tradeoff to non-ML stakeholders so they could make a decision.
3. 03 Walk me through a time you exercised autonomy to deprioritize model accuracy in favor of latency or interpretability.
4. 04 Give an example of when you used a bias for action to ship a simpler model quickly rather than waiting for the ideal one.

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

1. 01 Explain the differences between collaborative filtering and content-based filtering and how you would combine them for music recommendations.
2. 02 How would you address the cold-start problem when recommending tracks to a brand-new user with no listening history?
3. 03 Beyond accuracy, how would you evaluate a recommendation model using metrics like NDCG, diversity, novelty, and skip or save rate?
4. 04 How would you detect and mitigate popularity bias so that emerging artists are not crowded out of recommendations?

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

1. 01 Design the end-to-end Discover Weekly recommendation pipeline, from listening data ingestion to weekly personalized playlist generation.
2. 02 Design a real-time recommendation system that serves personalized playlists to millions of users with low latency.
3. 03 Design the feature store and serving infrastructure that powers autoplay and Daily Mix personalization.
4. 04 Design a model training and deployment pipeline that retrains recommendation models on fresh listening data and monitors them in production.