Uber Machine Learning Engineer Mock Interview

Tell us about a time a machine learning model in production caused measurable business impact at a large scale—how did you diagnose the root cause (data drift, label issues, feature leakage, or system changes), and what monitoring and retraining strategy did you put in place to prevent recurrence?

Uber uses ML heavily for matching, pricing, and safety—describe how you would design an experiment to evaluate a new recommendation or ranking model for rider-driver matching, including metric selection, offline/online alignment, handling delayed feedback, and dealing with selection bias.

Given a stream of trip events where some rides are missing key signals (e.g., pickup coordinates quality, cancellation reasons), how would you build a robust model and training pipeline that can handle missingness and noisy labels—what imputation, loss functions, and validation approach would you choose?

Situation: you’re asked to improve ETA or demand forecasting under tight latency constraints and with rapidly changing seasonality around events/cities. How would you prioritize modeling approaches (baseline vs. advanced), decide what to retrain, and ensure the solution is reliable across markets with different data volumes?

Walk us through a recent technical decision you made between optimizing for accuracy versus optimizing for calibration, fairness, or user experience (for example, ride acceptance rates or cancellation risk). What trade-offs did you consider, how did you measure success, and what did you do when the results didn’t meet expectations?