My ML Learning Stack as a College Student (And the Honest Mess That Got Me Here)

I'm not going to pretend I had a clean, structured path into machine learning. I didn't.

Like a lot of people, I started with YouTube. Watched a few videos, got excited, watched more. But after a while something felt off — I was consuming a lot and building nothing. The videos explained concepts but never quite answered the question I actually had, which was: where do I go after this? I'd finish a 20-minute video feeling like I understood something, then open a blank file and realize I had no idea what to actually type.

So I'd go back and watch another video.

Then came the paralysis.

Everyone online seemed to agree that you need to "understand the basics first." Learn linear algebra. Learn calculus. Learn statistics. Build small projects. Then maybe, eventually, you get to do something interesting. It sounds reasonable. But in practice it meant I kept starting over — new resource, new "foundations" course, same feeling of not going anywhere. I was collecting starting points without ever actually starting.

I tried Andrew Ng's Machine Learning Specialization. It's well-made. He's a good teacher. But honestly? It felt boring to me at first. I couldn't connect what I was learning to anything I actually wanted to build. The math made sense on its own, but I kept asking myself so what does this let me do? and not finding a satisfying answer. I pushed through some of it, drifted off, came back, drifted off again.

I wasn't finding what I needed. And for a while I genuinely wondered if ML just wasn't for me — if maybe I'd come to it too late, or wasn't the right kind of person for it.

Then I found fast.ai.

The whole philosophy of fast.ai is almost the opposite of how ML is usually taught. You don't start with theory and work toward practice. You start with a working model — something real, something that produces actual output — and then you go back and learn why it works. Top-down instead of bottom-up.

Something about that just clicked for me. Suddenly I was training a model before I fully understood what was happening inside it, and somehow that made me more curious about the internals, not less. The confusion felt productive. I wanted to understand it because I'd already seen it work.

So I ran with it.

I'm still early. I won't pretend I've figured everything out or built anything impressive yet. But for the first time I feel like I'm actually moving — not just reading about moving. There's a difference between studying a map and taking a step, and fast.ai was the first resource that made me take the step.

My actual stack right now

• fast.ai — my primary learning resource. The course is free, the book is free, and the community on the forums is genuinely helpful without being condescending. If you're stuck like I was, start here.

• Google Colab — for everything. Running experiments, taking notes, breaking things, and not worrying about my laptop's specs. The fact that it runs in the browser and needs zero setup was a bigger deal than I expected. Less friction meant I actually opened it.

• Python — I already knew Python from other projects, which helped a lot. Most of the ML ecosystem revolves around it anyway, so learning ML naturally meant using Python more seriously.

If you're coming in with zero Python background, I honestly wouldn’t spend months “mastering” it first. Just get comfortable enough to read code, experiment, and build small things. You can learn a surprising amount along the way.

If you want a beginner-friendly starting point, Mosh Hamedani’s Python tutorial is one of the better resources I found early on.

• Kaggle — starting to use it for practice datasets and seeing how other people approach the same problems. Reading other people's notebooks taught me more in a week than some courses did in a month.

The thing nobody really tells you

The

learn the basics first

advice isn't wrong — it's just incomplete. Basics learned in a vacuum don't stick. Basics learned because you needed them to understand something you were already building — those stick.

I'm not saying skip the fundamentals. I'm saying maybe don't make them the prerequisite. Make them the explanation.

If you're stuck in the loop I was in — consuming, restarting, feeling like you're not making progress — try flipping the order. Build something that barely works. Then ask why it works. The confusion that comes after building is so much more useful than the confusion that comes before it, because it has a direction.

You don't have to have a clean stack. You don't have to follow the path everyone recommends. Find the resource that makes you want to open your laptop at midnight for no reason.

For me, that was fast.ai. Yours might be different.

Still figuring it out — just publicly now.