Getting past the AI hackathon cold start problem – Remote Cluster Auto-setup

I'm curious as to what people do currently, without runhouse. And how does runhouse work, under the hood? Could I trivially configure model serving without resorting to FastAPI or some such?