Your question is High-Frequency Inference Under Constraints. 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).
Given a fixed budget and hardware constraints, how would you optimize a machine learning pipeline for high-frequency inference?