Your question is Backpressure and Rate Limiting. 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 handle backpressure and API rate-limiting in a multi-tenant control plane when multiple users execute large-scale inference workloads simultaneously?
Explain how you would align engineering and operations on capacity assumptions, prioritization, queuing, tenant isolation, observability, and rollout safeguards. Address trade-offs between fairness, throughput, latency, reliability, and customer experience, including how you would respond if demand exceeds forecast.