Q1
Walk me through how you would approach feature engineering for a high-dimensional dataset with 500+ raw features. What techniques would you use to reduce dimensionality and avoid overfitting?
Why they ask this:* They're assessing your practical knowledge of feature selection, PCA, regularization, and domain expertise—critical skills for building production models that generalize well.
Q2
Describe a time you built a machine learning model in production. What metrics did you use to evaluate it, and how did you handle model drift or performance degradation over time?
Why they ask this:* They want to understand your experience moving models beyond notebooks into real systems, including monitoring, retraining strategies, and MLOps fundamentals.
Q3
Explain the difference between batch and real-time prediction pipelines. What are the trade-offs, and which would you choose for a recommendation system processing millions of user events daily?
Why they ask this:* This tests your understanding of system design, scalability, latency requirements, and practical trade-offs between architecture choices relevant to data science in industry.
Q4
How would you design an A/B test to evaluate whether a new recommendation algorithm improves user engagement? What statistical concepts would you apply, and what pitfalls would you watch for?