Your question is Low-Latency Inference Challenges. 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).
What are the common challenges when deploying models that require low-latency inference?
Discuss the end-to-end path from input preprocessing through model execution and response delivery. Cover model selection, feature availability, infrastructure, concurrency, tail latency, monitoring, and accuracy tradeoffs. Explain how you would measure p50, p95, and p99 latency, identify bottlenecks, and validate that optimizations do not introduce training-serving skew or unacceptable model degradation.