Machine Learning Engineer interview
at Amazon.

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 took ownership of an ML model in production end-to-end, including monitoring it after launch and fixing it when it degraded (Ownership).
2. 02 Describe a time you disagreed with a data scientist or stakeholder about a modeling choice or metric, held your position, and then committed once a decision was reached (Have Backbone; Disagree and Commit).
3. 03 Give me an example of when you dove deep into messy training data and found a labeling or data-quality issue that was hurting model performance (Dive Deep).
4. 04 Tell me about a time you shipped a simpler model quickly to deliver results rather than waiting to perfect a more complex one (Bias for Action / Deliver Results).

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

1. 01 How would you build the retrieval and ranking stages of Amazon's 'customers who bought this also bought' recommender, and which offline metrics like NDCG or MRR would you optimize?
2. 02 Explain how you would handle the cold-start problem for newly listed products in Amazon's recommendation system.
3. 03 How would you design online and offline evaluation for a model that ranks search results on Amazon, and how would you guard against feedback loops?
4. 04 Describe how you would detect and mitigate training-serving skew for a fraud-detection model serving Amazon transactions.

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

1. 01 Design the real-time personalized product recommendation system serving Amazon's homepage at millions of requests per second, including offline training and online serving.
2. 02 Design the ML model-serving and feature-store infrastructure behind Amazon search ranking.
3. 03 Design a fraud-detection system for Amazon payments that scores transactions in real time and adapts to new fraud patterns.
4. 04 Design an ad-matching and ranking pipeline for Sponsored Products on Amazon's search results page.