About
We measure before we claim.
Hatteria Labs is a research and engineering company in Plzeň. We measure what can be removed from a language model without breaking it, and build software on whatever survives the test.
Who we are
Hatteria Labs is based in Plzeň, Czech Republic. Radek Mužík and Miloslav Sklenička founded it in August 2026 as a place where researching the internal structure of language models sits next to building software they run themselves.
Research and products are not two departments here. The people who measure what a transformer can lose without breaking also keep Torumata, AIDJ and Slidify running.
Running the systems as well as writing them is a deliberate choice, not a constraint. Operating a service is how you find out how a research result actually behaves once real users depend on it.
From research to product
Research and products stand on the same discipline, not the same codebase. Our compression study ran nine escalating tests against a single question — how much of a transformer is redundant — and then widened into fourteen approaches to saving memory or compute. Three survived. Of the eleven that remove dimensionality or force structure onto the weights, not one can be applied twice without the model collapsing. A second line of work asks the same question about memory rather than weights: what a longer context window actually returns, measured instead of estimated.
What is left after a measurement like that is unremarkable, but it is dependable — and dependable is what software needs once other people run their business on it. Torumata, AIDJ, and Slidify do not run on the research results directly, but on the same procedure behind them: measure first, claim second.
People
Hatteria Labs was founded and is led by Radek Mužík and Miloslav Sklenička, the company's managing directors. Research, engineering and operations are shared work.
Radek Mužík
Managing Director
Miloslav Sklenička
Managing Director
Why Hatteria
Hatteria is the older genus name for the tuatara, a New Zealand reptile that carries a third, parietal eye on the top of its head. It forms no images. It registers light, the rhythm of the day and changes in the environment that ordinary sight passes over.
That is how we understand artificial intelligence: not as a replacement for human judgement, but as a third eye — a way to pick out patterns and signals in data that a person cannot see. Our work is to open that eye, and to build systems that see what would otherwise stay hidden.