Hacker News Reader

Comment by tnolet | Hacker News Reader

By all means, read the books mentioned in this thread.

And then drop it and just start working.

There is a large swath of self-branded Indie Hackers who cling to books, blogs, conferences etc. but are too paralyzed by all the advice and input to actually do the one thing they need to: get customers.

Disclaimer: I run my own SaaS and visit Indiehacker meetups and stuff. I’m not a nay sayer, just stay skeptical.

View on HN

I've found this to be true in a lot of subjects. When I started really getting into machine learning a few years ago, I did all the courses, refreshed on my college math, followed tutorials to learn things like tensor flow. But despite all that, I was totally flummoxed on my first day as an ML engineer at work. It wasn't until I started building things to solve problems and learned practical lessons by doing that (with messy data vs toy prepared datasets) that I really excelled. I wouldn't be surprised if this tactic of just starting working would help here too.

So, even though I haven't specifically built a saas business by myself, though I have a bunch of experience growing small startups to medium to large companies, I would echo this advice.

Did you work on any projects before getting the ML job? i try to learn by myself but never seems to me any employer would hire me on the basis of doing online courses. So how did you end up crossing that gap?

You need to find an employer who is hiring based on potential or potential + a track record in a different area. I wouldn't hire an ML practitioner with no practical experience. I would hire someone who's produced in other areas and is looking to make the move to a junior ML practitioner (or who has domain expertise in our corner of the world).. It's like the rest of engineering... just because you've taken a course in Python doesn't make you a software engineer or proficient in it. Now show me that you've done things in other languages and we can start talking. Like, courses in ML don't _really_ teach you about things like:

* reproducing model training * deployment of experiments in a CI/CD pipeline * observability of models * discoverability and governance of results * versioning of data / models * optimizing for latency vs throughput * when to use batch vs real time etc

Just like a course in a computer language wouldn't necessarily teach you about CI/CD, horizontal vs vertical scaling, or domain-specific bits.

Just seeing this. This is exactly right ime, I had years of senior full stack SWE experience /track record of SWE at my employ and was able to get myself on a RNN project initially, then expanded from there. I didn't get hired at a new company as an ML person out of the gate.

Reply on news.ycombinator.com