Your question is LLM Serving with Orchestration. 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).
Design a system for LLM serving that incorporates real-time data processing and intelligent orchestration.
Explain how you would route requests across models, incorporate continuously arriving data, manage multi-step or tool-using workflows, and meet production latency and reliability requirements. Cover system sizing, online versus batch components, evaluation, observability, model and feature drift, training-serving skew, cost controls, and failure recovery. State reasonable assumptions where requirements are unspecified.