Ask HN: How are you running AI experiments?

Curious to learn setups, run books, playbooks etc on how other HN'ers are running AI experiments (esp for personal projects and small startups).

What vendors are you using, what roadblocks have you faced, how have you overcome them, what are some up and coming tools other HNers should be trying. And, of course, what you are working on!

There was a post earlier today from a new GPU cloud service but it made me think it would be great to hear from the community what's good right now.