Applied Research · Est. MMXXII
Your model answers everything. Ours answers once.
An LLM's first draft is a paragraph; ours is a decision. We distil, quantise and prune until one answer is enough — smaller models, fewer tokens, less of everything except the part that decides.
Ethos
Less is the whole method.
A model that does more with less is not a smaller model — it is a better-aimed one. We take away until only the decision is left.
No noise, no theatre — only the work, and what it leaves behind.
Intelligence
AI that does the work, not the demo.
Models are material — studied, distilled and sharpened until they disappear into the work.
What remains is the system: running, deciding, finishing its job without us.
Experiments
Experiments, art, and everything in between.
We treat research as a studio practice: build the strange version first, measure what it does, and keep only what surprises us.
An interface is a composition too. Restraint, rhythm, timing — the craft of a system is not decoration; it is how the work is understood.