A small research lab building foundation models from the ground up

Auxerta started in 2025 as a domain annotation service. For a year we cleaned and labeled training data for clients in finance, law, and other regulated fields.

By early 2026 we had shipped about a million labeled samples, and the limit was no longer the data. Better data only takes you so far when the constraint is the model itself, not the labels.

So in 2026 we pivoted. Auxerta is now a foundation-model lab working on post-transformer architectures: sub-quadratic sequence models that aim past the Transformer's cost curve, and representations that are not bound to next-token prediction. The work is in progress; the technical case is in the research note.

Founders

Where we work