You already know: LLM serving basics and the LLM platforms case (gateways in outline). This case goes deep on the gateway itself: an infrastructure component where the hard parts are quotas measured in tokens you cannot predict, failover that does not multiply load, streaming connections as the unit of capacity, and an overhead budget of a few milliseconds.
How to read this lesson. No code, on purpose. Numbers marked assumed are assumptions you state out loud; prices are illustrative.
1. The prompt, and the questions that change the design
A mid-level answer: "an API proxy with rate limiting". A staff candidate sees what makes LLM traffic different: cost is per token, not per request, output length is unknown at admission, responses stream for seconds, providers fail partially (slowdowns, not just errors), and retries are expensive because every retry is billed.
Assumptions you state: