Just took a deep dive into OTUS and I'm trying to figure out how to run decentralized AI workflows without burning a hole in my wallet. Has anyone cracked the code on minimizing latency while keeping compute costs down? Let's swap strategies and configs.