Get in touch
One-person lab. Stuttgart. Open to the right conversation.
The work here is about structural correctness — making the most damaging structural errors in supervised ML workflows unrepresentable, not just detectable. The ML grammar is the first result. The approach is meant to generalize.
If you have a problem where the tools give you plausible numbers but the results don’t hold, or where correctness depends on ordering and data flow that nothing currently enforces — that’s the kind of thing worth talking about.
Also open to: research collaboration, academic visits, and questions about the papers. Not open to: generic consulting or unsolicited pitches.