I “Learned” AI for 18 Months and Had Nothing to Show For It. Here’s What Changed

The uncomfortable truth about how developers actually learn and the simple fix that finally worked for me

Let me paint you a picture.

It’s Sunday afternoon. You’ve got four browser tabs open. One is a Coursera module you’re 40% through and haven’t touched in three weeks. Another is a YouTube playlist titled “Machine Learning Full Course 2024” with 47 videos, of which you’ve watched eleven. A third tab is someone’s GitHub repo with a neural network you cloned, ran once, and vaguely understood. The fourth is Reddit, because you needed a break from all the learning.

Sound familiar?

That was me. For almost two years.

I wasn’t lazy. I genuinely wanted to learn AI, I work in software, I saw what was happening in the industry, and I knew I needed to level up. I put in real hours. I finished courses. I read papers. I built small projects that mostly worked.

But if you’d asked me to map out what I actually knew versus what I’d half-absorbed? I couldn’t do it. Everything felt vaguely connected and specifically incomplete.

The problem wasn’t the resources. The internet has more free, high-quality ML content than any human could consume in a lifetime. The problem was that I had no structure, no measurement, and no way to see the full picture of what I was building toward.

The Bookmark Graveyard

Every developer I know has one. A folder in your browser called something like “Learning” or “Resources” or, if you’re optimistic, “Goals 2027.”

Mine had 340 links in it.

I’d find something useful a great tutorial on attention mechanisms, a clear explanation of gradient descent, a deep dive on transformer architecture and I’d bookmark it. With every intention of coming back.

I came back to maybe 12% of them.

The rest became a museum of good intentions. A digital record of all the things I meant to learn but never quite got around to.

The Roadmap Trap

So I tried roadmaps.

You’ve seen them those massive flowcharts that show you every skill a DevOps engineer or ML practitioner needs to know. They’re actually great. The problem is they’re static. You look at them, feel a combination of inspiration and mild panic, and then… close the tab.

There’s no tracking. No way to mark what you’ve done. No memory of where you left off. You come back a month later and start from scratch trying to figure out where you were.

I needed the structure of a roadmap with the track ability of a todo list. Those are different things, and I couldn’t find anything that combined them without being either too simple or so feature-heavy it became another thing to manage.

What Actually Worked

A few months ago I started using NeuralPath (https://neuralpath.app), a free tool that’s basically a structured roadmap you can actually track your progress through.

The concept is simple. You pick a career track AI Engineer, ML Engineer, Data Scientist, Backend Developer, Cloud AWS, whatever fits where you’re going. The roadmap is already structured into phases. Within each phase, there are topics. Within each topic, there are subtopics, resources, and projects you can tick off as you go.

https://neuralpath.app/#/explore

What got me wasn’t the feature list. It was the progress dashboard.

For the first time, I could see actually see what percentage of my ML foundations I’d covered. I could see that I understood supervised learning pretty well, had a shaky grasp of neural network architectures, and had completely skipped anything to do with model deployment.

That last one stung. I’d been “learning ML” for 18 months and hadn’t touched deployment once. In a job, that’s the part that matters most.

The Dashboard That Embarrassed Me Into Action

That screenshot above is from my actual account after about 6 weeks of using it seriously.

The streak counter is both motivating and guilt-inducing in exactly the right proportion. Miss a day and it resets. Not in a punishing way more in the way that running apps work, where the streak itself becomes a small game you want to keep going.

The activity feed on the right shows what I’ve completed recently. Seeing “Completed: Attention Mechanisms” and “Completed: Transformer Architecture” in a list felt different than just closing a tab after watching a video. It felt like I’d actually done something.

How I Use It

My actual workflow, for what it’s worth:

1. Pick one phase at a time. Don’t look at the whole roadmap, it’s overwhelming. Just focus on the current phase.

2. Before starting a new topic, expand it and read what subtopics it covers. This 30-second step prevents wasted time going deep on something you already know.

3. Mark subtopics as you go, not all at once at the end. It sounds trivial but the act of ticking something off mid-learning keeps you engaged.

4. Use the project section honestly. If you didn’t build something, don’t mark the project done. This keeps the completion percentage meaningful.

The Specific Part That Changed How I Think About Learning

When you expand any topic, you see the full breakdown, subtopics, linked resources (papers, videos, articles), and any projects attached to it.

This sounds basic. But it changed something for me.

Before, “learn transformers” was one task. Vague, unbounded, impossible to feel done with. Now it’s: 8 subtopics, 3 papers, 2 videos, 1 project. I can see the edges of the thing I’m trying to learn. That makes it finite. Finite things get done.

Is It For Everyone?

Honestly? No.

If you learn well from courses that provide their own structure and progression, you probably don’t need this. Platforms like fast.ai or deeplearning.ai have built-in progress tracking and curated paths that work well for many people.

NeuralPath is for people who learn from multiple sources. YouTube, papers, blogs, side projects and need a home base to track it all. It’s for people who’ve tried roadmaps but found them too passive. And it’s for people who need to see the full picture of a career track to stay motivated.

It’s free. No paywall on the core features. You can try it in 60 seconds.

If you’re somewhere in the bookmark graveyard phase of your ML journey, it might be the thing that helps you climb out.

https://neuralpath.app