tideline

live · checked 03:02 UTC

AI and machine learning interview preparation

General preparation, not from any one company

AI and ML interviews range from research questions to production engineering. Most test the basics of models and evaluation, how you work with data, and how you’d ship and watch a model in use.

  1. Foundations

    What they’re checking: Whether you understand why models behave as they do, not only which library call to make.

    • Explain overfitting to a product manager, and how you’d spot it.
    • When would you choose a simple model over a large one?

    How to prepare: Revise the core ideas behind the models in the posting, and practise explaining each without jargon.

  2. Evaluation

    What they’re checking: How you decide whether a model is good enough, and for whom it might fail.

    • Your model scores well offline but users complain. What do you look at?
    • How would you choose the right measure for a model that flags rare events?

    How to prepare: Prepare one project where the first measure you picked was wrong, and how you found a better one.

  3. Data

    What they’re checking: Care with data: where it came from, what’s missing, and what could leak into the results.

    • How do you check a training set before you trust it?
    • Tell me about a time a data problem changed your results.

    How to prepare: Be ready to describe the data behind one of your projects: its source, its gaps and how you cleaned it.

  4. Shipping and monitoring

    What they’re checking: Whether you can take a model from a notebook to something people use, and notice when it drifts.

    • How would you serve a model that must answer in under 100 milliseconds?
    • What would you monitor once a model is live?

    How to prepare: Sketch the path one of your models took to production, or would take, including how it’s rolled back.

  5. Responsible use

    What they’re checking: How you think about harm, fairness, privacy and limits, and whether you raise concerns.

    • How would you find out whether a model works worse for some groups of users?
    • Tell me about a time you pushed back on how a model would be used.

    How to prepare: Read what the company publishes about how it builds and uses AI, and have one honest question ready.

Answering well

For “tell me about a time” questions, the STAR shape keeps an answer clear: the situation, your task, the action you took, and the result, with what you learned. Use your own stories; there is no right script to memorise.

Companies differ: check what each one publishes about its own interviews before you prepare for it.