"Tell me about a time you owned a Python service that was under pressure because synchronous ML inference was driving high CPU utilization and hurting reliability. How did you decide what to do first, how did you communicate trade-offs to stakeholders, and what was the outcome?"
