What you need to know
Prepare for AI engineer interviews with question patterns and answer frameworks covering ML foundations, LLM systems, production design and responsible AI.
How to use these questions
Answer aloud, state assumptions and connect theory to a project. Interviewers often care less about a textbook sentence than about how you diagnose ambiguity, choose a metric and respond when evidence changes. Use a simple structure: clarify the objective, describe the approach, identify risks, define evaluation and explain trade-offs.
Machine-learning foundations
How would you detect and prevent data leakage?
Define leakage as information unavailable at prediction time or contamination between training and evaluation. Discuss splitting before learned preprocessing, time-aware splits when appropriate, fitting transformers only on training data, checking duplicate entities and reviewing feature provenance. Give a concrete example and explain how leakage would distort the metric.
Accuracy is high, but users report poor results. What do you do?
Check whether accuracy matches the business cost, then inspect class balance, confusion matrix, segments and recent data. Validate the labels and serving pipeline. Compare offline and online inputs, define a better metric and create error slices. The strongest answer treats this as both a model and system investigation.
Explain overfitting and your response
Describe the gap between training fit and generalization. Responses can include better validation, simpler models, regularization, more representative data, augmentation or early stopping. Explain which evidence would lead you to each option.
LLM and RAG systems
How would you evaluate a RAG application?
Separate retrieval and generation. Create representative questions with expected evidence, measure retrieval relevance or recall, verify citation support, score answer usefulness and test refusal when evidence is absent. Track latency and cost. Include adversarial or prompt-injection cases and review important failures with humans.
When should you use prompting, RAG or fine-tuning?
Start from the required behaviour. Prompting is suitable when the base model already has the capability and instructions/context are enough. RAG is useful when answers need current or private knowledge with traceable evidence. Fine-tuning can help consistent behaviour or task adaptation, but it does not automatically provide current facts. Mention maintenance, evaluation data, privacy and cost.
What can go wrong with an AI agent?
Discuss incorrect tool choice, unsafe side effects, excessive loops, prompt injection, leaked secrets, stale state and unbounded cost. Mitigations include least-privilege tools, confirmation for consequential actions, typed inputs, timeouts, budgets, sandboxing, audit logs and evaluation of complete trajectories.
Production system design
Design an online inference service
Clarify traffic, latency, freshness and availability. Cover input validation, feature or retrieval dependencies, model versioning, autoscaling, caching, fallbacks, observability, staged rollout and rollback. Define service metrics and model-quality signals. Google's production guidance is a useful reminder that data and operational components usually outweigh the model code.
How would you monitor model quality without immediate labels?
Use input schema checks, distribution shifts, embedding or prediction changes, proxy outcomes, sampling for human review and delayed-label backfills. Establish thresholds cautiously and connect alerts to an investigation runbook. State that drift does not prove performance loss; it signals a need to investigate.
Responsible-AI reasoning
A model performs worse for one group. What next?
Verify the measurement and sample size, then inspect data coverage, labels and operational context. Involve domain and affected stakeholders, assess harm, compare mitigation options and document the decision. Do not promise that a single fairness metric solves the problem.
Behavioural and project questions
Prepare examples for a failed experiment, a production incident, disagreement about a metric, a trade-off under time pressure and a new skill learned quickly. Use situation, task, action and result, then add what you would do differently.
Finally, choose questions from the actual vacancy. A machine-learning role, LLM role and MLOps role should not receive identical preparation.
Methodology
Editorial synthesis of cited sourcesQuestions are organized around competencies common to applied AI roles and checked against official ML, production and risk-management guidance. The answer frameworks identify reasoning to demonstrate; they are not scripts to memorize or claims about a particular employer's process.
Read our editorial and research policy →Sources
- Machine Learning Crash Course
Google for Developers · Accessed September 2026
- Production ML systems
Google for Developers · Accessed September 2026
- scikit-learn User Guide
scikit-learn · Accessed September 2026
- Artificial Intelligence Risk Management Framework
US National Institute of Standards and Technology · Accessed September 2026