Verily Applied Scientist Interview Questions

Train and evaluate a classifier while measuring demographic disparities and applying mitigation strategies that preserve useful predictive performance.

data preprocessingfairnessModel EvaluationVerily

Design a reliable classification pipeline for a rare clinical condition using imbalance-aware training and precision-recall evaluation.

data preprocessingmodel selectionModel EvaluationVerily

Design a reliable pipeline to clean, standardize, validate, and harmonize heterogeneous EHR data for downstream ML.

data securitydata pipelineFeature StoreVerily

Design a real-time patient monitoring system that detects adverse events and delivers safe, actionable clinician alerts.

Feature Driftanomaly detectiondata streamsVerily

Explain how to trade off model complexity, prediction quality, inference cost, and production latency.

latencyproduction environmentinference latencyVerily

Explain how to evaluate a model when labels are noisy, inconsistent, or dependent on human judgment.

performance evaluationEvaluation TechniquesCalibrationVerily