We assemble frontier AI into R&D pipelines, file the IP before wet lab, and spin out the companies. Six projects across de novo enzyme design, sovereign regulatory AI, and connectome-scale neural simulation.
The first computationally designed enzyme family targeting per- and polyfluoroalkyl substance degradation. Luxembourg patent LU605811 filed April 2026; wet lab validation in planning with academic partners.
A language model trained on the complete CSSF regulatory corpus and Luxembourg fund law, paired with a RAG-based compliance assistant. Built for the €6.3T Luxembourg fund industry on sovereign infrastructure.
Marcus theory and Dutton-Moser rules encoded into a generative pipeline for designing multi-cofactor redox enzymes, the tool that makes the next generation of Nekazyme-class work possible.
Building on the FlyWire connectome (139,000 neurons, full Drosophila brain) to run biologically-grounded behavioral simulation in MuJoCo. A testbed for understanding emergent intelligence without training.
A foundation model over borehole records, mineral surveys, and lithology maps to replace expensive geological consulting in critical-minerals exploration.
A next-generation application of de novo protein design targeting osteochondral regeneration. The enzyme-design pipeline transfers directly to scaffolds, vascularization signals, and growth factor analogs: the protein-engineering layer of a decade-long regenerative medicine program.
We pick problems where the computational stack is ninety percent complete and assemble the last ten.
We file patents before we validate, because IP protects the design space while experiments run.
We work with academic partners for wet lab validation rather than owning experimental infrastructure.
We spin out the projects that work; we open-source the ones that don't. We publish preprints regardless.