Everyone is building LLM Inference

Since everything is offline, what ended up being the way to go. Insider take: routing will not be written by AI. Maybe "co-authored by Claude and friends", but not authored. So I read a whole bunch of text trying to figure out a viable method of model routing.