ML Engineer Interview Scenarios

Scenario — Model quality drops after deployment

A model's production quality has degraded even though the training pipeline has not changed.

Discuss:

• Data and feature drift.
• Label quality and delayed feedback.
• Training-serving skew.
• Evaluation datasets and metrics.
• Model/version comparison.
• Rollback and monitoring.

Scenario — Deploy a model safely

Explain how you would use staged rollout, observability, rollback, and validation to reduce production risk.

Always state the failure signal and rollback condition, not only the deployment mechanism.