I’m Michał Maik — I build agentic AI systems for a living and adversarial things that confuse machines for fun.
Recovering physicist. I spent years studying how the universe behaves, then switched to making language models behave. Mostly the latter sticks.
Currently leading a research team building agentic AI systems — orchestration, tool use, and the unglamorous engineering that makes autonomous agents actually reliable.
An LLM tarot reader built on a single premise: hallucination isn’t the bug, it’s the feature. Draws from multiple providers, stays faintly mystical, never claims to be right.
open →A CAPTCHA built on motion perception — trivial for a human eye, adversarial for a model. I also wrote the static attack against my own design, because someone had to break it first.
code →I came to machine learning from physics — a PhD, then a detour through finance, then years leading R&D on LLM-powered systems. I like problems that sit at the seam between perception and computation: where a human and a model look at the same thing and disagree. I work in the open where I can, prefer local and on-device tools, and have a soft spot for synths and physics-flavoured wordplay.