[breckenedge]

I dipped my toes in about 7 months ago with a 3-month-long project to make content recommendations using ML. I started with off-the-shelf collaborative filtering libraries and ended with PyTorch. ChatGPT was a huge help. I would have been OK continuing down that route, but execs wanted faster and better results, and 3 months is just enough time to really get a flow going when starting from no experience.

A lot of ML was cleaning up and preparing datasets, which wasn’t that much fun. I had an exec pushing for using Amazon Personalize, which I gave a good try but ultimately it didn’t deliver. Was that because of data problems or underlying models? That’s the crux of the problem when using black-box ML services: you can’t analyze what’s going on. And Amazon Personalize makes changing the data layer a pain, so you never know if you’re getting closer to a better solution. Who knows, Personalize is probably great in the hands of experienced ML folks.

So, if you can swing it, definitely recommend doing an ML trial project to see if you like it before committing your career to it.