Your question is Diagnosing Production Latency Spikes. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
How would you approach a situation where a system's latency has spiked unexpectedly in production at Seven Ai?
Explain how you would triage the incident, localize the bottleneck across the request path, distinguish infrastructure issues from model-serving or feature-store problems, and restore the latency objective safely. Include the observability, rollback, capacity, and follow-up steps you would use. Discuss how you would protect model quality while applying mitigations.
