I stopped asking AI for answers. I asked it to quiz me instead. (Fast forward)

Parts 1 and 2 were about a Docker reactivation — a couple of AI-coached challenges that stuck better than re-reading docs.

That was the starting point.

A lot has happened since.

I kept the method, but I started feeling its limits. The AI had no memory of what I’d practiced the session before. Every time: cold start. No profile, no tracking, no “you struggled with networking last week — let’s revisit.”

So I started building what I was missing.

A small platform. Python. Designed around one idea: the system should know what you’ve practiced, where you’re weak, and what you’re ready for next — not just today, but across sessions.

I’m dogfooding it with the Docker course that started all this. It’s still early. But it’s already a different experience.

The repo is public: github.com/elmika/learning-engine

What started as a 20-minute experiment is turning into something I actually want to use every day. And building it surfaced something I didn’t expect: the method was always the moat — not the AI. The AI is just the delivery layer. The value is in the memory, the profile, the engine that knows where you left off.