One job for fourteen years
Same job for fourteen years, just different material: give a system a goal and constraints, let it search. The hard part was never the computer science, it was trust.
Three systems, one decision loop
Built at Heineken's Global Analytics & AI Hub to work as one: from annual budget planning down to week-by-week retail execution.
“Most importantly, he has great working habits: extremely well organized in all aspects, always looking into possible ways to improve his own work and his team's.”
“Tomasz made a great effort to investigate the problem, and through the research process he kept the highest standards and followed very clear logic.”
The model isn't the hard part
It's the human layer — the trust, adoption, and judgment calls that decide whether people actually act on it.
I spent a decade building generative and optimization systems: genetic algorithms, evolutionary solvers, and large-scale parametric work at MAD Architects and AREP. The same mental model of handling uncertainty, constraints, and trade-offs now applies directly to LLM agents and agentic workflows.
At Heineken, I applied this thinking at scale: shaping the product layer for AI systems built to optimize a ~€2.9B marketing budget. I approach AI products as a builder who owns outcomes, from discovery to shipped systems, not just as a designer of interfaces.
My specialty is uncertainty UX: turning model output and confidence into explainable, simulatable interfaces that non-technical teams trust and act on. On every product I've shipped, the bottleneck was never model accuracy. It was adoption.
The long way here
Before AI agents, I optimized buildings with genetic algorithms. Same operation, different material: set the objective, let the system search.
Buildings that were programs
A decade before "AI agent" was a phrase, I was writing evolutionary solvers that bred building geometry from a fitness function. Same operation, different material: set the objective, let the system search.
