Your question is Distributed Inference Caching. 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 architect a distributed inference caching layer to minimize redundant compute across concurrent high-volume user sessions?
Discuss cache-key design, request coalescing, consistency, invalidation, capacity planning, and failure handling. Explain how you would evaluate cache effectiveness while avoiding stale or cross-user results.