Software Engineer Machine Learning Interview Questions

This is the ML knowledge check inside a software engineering loop, 70 questions deep, and it asks you to explain rather than to build. Amazon leads with 5, OpenAI and NVIDIA follow with 4 each, and Microsoft, Anthropic, TikTok and Google are all present, which is why the material splits between classical fundamentals and the concerns that only appear once a model is serving traffic. On the fundamentals side: what overfitting looks like and how you catch it, the assumptions behind linear regression and what happens when you fit three points, how contrastive learning chooses its positives and negatives, what attention is actually doing inside a transformer. On the systems side: tensor against pipeline parallelism and when each one pays off, optimisation choices during training, human-in-the-loop review and permission controls for a retrieval-augmented product, and how you would keep an LLM feature from hallucinating. A few ask for small code, such as a bigram next-word predictor with weighted sampling. 38 of the 70 arrived in a technical screen against 9 onsite, so this is early-round filtering and usually a conversation. Every question carries a written solution, 34 sit at medium, and 72% is free.