We are a research‑driven company working on the foundations of numerics for training and inference. We are exploring the fundamental limits of the algorithmic layer underpinning AI.
We work on two main interconnected areas, building a positive feedback loop. On one front, we develop hardware-aware algorithms, kernels and new primitives for model architecture surgery. On the other, we design agentic infrastructure and verification techniques to automate runtime optimization.
Our long-term mission is to systematize and automate algorithmic advances in AI.
We are fully funded and backed by leading investors and research institutions in the Bay Area and Tokyo.
