I’m Michał Maik. I build small, fast, reliable agentic AI systems by balancing LLMs with traditional code.
Not just a record of the work so far, but a place for whatever I build next, however far afield.
A CAPTCHA built on motion perception. Trivial for a human eye, adversarial for a model. Includes attacks against my own design, because breaking it made the newer version more robust.
code →An LLM-powered tarot reader made for fun. Currently password-gated while I trim the rough edges.
open →My path ran physics, then data science, then NLP, computer vision, and now LLMs and agents. Each time the field changed shape I followed it somewhere new. What never changed is what intrigues me. I’m pulled toward the problems that aren’t supposed to be solvable, the cases that don’t fit, the work that starts with someone saying it can’t be done. Those are the ones I stay with.
The same curiosity spills past work. It’s why the things I make don’t sit in one place. One week it’s a serious tool, the next a silly side project. I’d rather chase an odd question into unfamiliar territory than settle into being an expert at one thing.
Off the clock, the same wiring keeps running. I read about how minds work and where they fail, lean on a bit of Stoicism to keep things in proportion, and generally can’t stop turning over how things fit together. The off switch, when I find it, is a walk with Ed, an old dog with crooked front paws who sets his own pace and doubles as my rubber duck, or an evening with Talking Heads, our favorite band.