A one-minute live loop: LLM calls through a gateway, into tracing and monitoring, then evals on the output.