🎧 Eric Riddoch's Linkedin Analytics

I'm an ML Platform engineer! What is that? Yes, Platform Engineering is rebranded DevOps. But the rebrand is trying to get back what DevOps was originally meant to be. And I'm trying to do that for MLOps and Data Science! It's a super challenging problem because: • SWE's speak a different language: often, the "best practices" for handling data defy the conventional wisdom in "traditional software engineering". • Simplicity: MLOps--like DevOps--is a practice, not a person. MLOps is best practiced by the people creating inference pipelines. For DevOps, that's SWE's, but for MLOps, it's... statisticians, academics, domain experts. These are generally people with weak SWE skills. The demand for simplicity is high. • Risk: ML/AI is about making inferences (guesses). By definition, when you're guessing, it's impossible to write tests that guarantee if your thing works before deploying. You *have* to test in prod. (they call it A/B testing). ML platform engineering is a new field. And it's awesome.