GhostpilotHome · Interview questions · Machine Learning Engineer
Machine Learning Engineer interview questions
ML engineer interviews sit between software and data science. Besides knowing models, you need to show you can put one in production and notice when it starts to fail.
What interviewers look for
- Engineering. Production-quality code, not only notebooks.
- ML knowledge. Training, evaluation and trade-offs.
- ML systems. Serving, monitoring and retraining.
How the interview usually runs
- Recruiter call.
- Coding screen.
- ML fundamentals interview.
- ML system design, then behavioural.
Stages vary by company, so ask the recruiter what to expect before you start preparing.
10 Machine Learning Engineer interview questions, and how to answer them
1. Design a recommendation system.
What they want to see: ML system design.
How to answer: Cover candidate generation, ranking, features, offline metrics and an online test.
2. Bias-variance trade-off in plain words.
What they want to see: Fundamentals.
How to answer: Too simple misses patterns, too complex memorises noise. Say how you'd tell which you have.
3. How do you serve a model with low latency?
What they want to see: Production thinking.
How to answer: Use batching, caching, smaller models or quantisation, and measure p95.
4. How do you detect model drift?
What they want to see: Monitoring.
How to answer: Track input distributions and live metrics, with retraining triggers.
5. How do you keep training and serving features consistent?
What they want to see: Avoiding skew.
How to answer: Share feature code or use a feature store, and log features at serving time.
6. What's the attention mechanism?
What they want to see: Deep learning basics.
How to answer: It lets a model weigh parts of its input. Connect it to transformers.
7. When would you not use machine learning?
What they want to see: Judgement.
How to answer: When rules work, data is poor, or errors are too costly.
8. How do you run experiments reproducibly?
What they want to see: Engineering discipline.
How to answer: Version data, code and parameters, and track with a tool such as MLflow.
9. How do you reduce inference cost?
What they want to see: Business sense.
How to answer: Use smaller models, batching, caching and the right hardware.
10. Tell me about a model you put in production.
What they want to see: Impact.
How to answer: Cover the problem, metric, serving and the result.
A sample answer
How do you detect model drift?
On a fraud model, I logged input features and scores daily, and alerted when distributions moved past a threshold. One alert traced to a new payment type our training data never saw. We labelled a fresh sample and retrained. Without the alert, precision would have kept falling unnoticed, since real labels arrive weeks later.
A made-up example to show the structure. Use your own experience and real numbers.Common mistakes
- Stopping the story at model accuracy.
- No plan for monitoring.
- Weak general coding skills.
- Ignoring latency and cost.
Practise this out loud
Run a mock interview for a Machine Learning Engineer role. Questions change every time, you answer by voice or typing, and each answer gets a score with feedback.
Frequently asked questions
How is an ML engineer different from a data scientist?
ML engineers focus more on building and running systems. Data scientists focus more on analysis and experiments, though titles vary.
Do I need to know LLMs?
It's increasingly useful. Know embeddings, retrieval and evaluation even if the role isn't focused on LLMs.