Research
We investigate questions that can change how teams build, evaluate, and operate AI—not research theatre detached from use.
Model architecture
An applied research line exploring how diffusion-style generation and mixture-of-experts architectures can change efficiency, specialization, and inference behavior.
Research notes, experiments, and reproducible artifacts will be linked here as they are released.
Evaluation initiative
A public evaluation initiative designed to compare useful model behavior with transparent tasks, judgments, and methodology—not a single opaque score.
The benchmark protocol and first results are in preparation.
Bring us the messy version
A process, an AI pilot, a growing bill, or a product idea. Start with the friction—not a solution brief.
