Whenever you're ready, walk me through how you'd approach this.
How would you design a scalable architecture for an enterprise-level LLM deployment?
Explain how you would clarify workload requirements and design the end-to-end platform, including model routing, retrieval-augmented generation, online and batch inference, security, observability, evaluation, and cost controls. Address the tradeoffs between hosted APIs, self-hosted models, and hybrid cloud deployment, along with latency, availability, data governance, model drift, and training-serving consistency.