Surrogate models
Fast approximations of expensive physics — structural, thermal, fluid — trained so a search can run millions of evaluations instead of five.
AI Advanced IncModel architectureSpace Unknown
Solvers that know the physics
Design at the scale of what is possible.
Your engineers reach the part of the design space a human mind can hold. We build the instrument that searches the rest — model architecture for aerospace, defence, space, medical, automotive and robotics.
01
The clearing
The mark is the argument: a frame that does not close, and one small settled block low and left.
Every design problem contains a territory of configurations that would have worked. What a team actually reaches is a clearing inside it — small, bounded by working memory, fatigue, training and time. Everything else is still there, unvisited.
Our work is to build the instrument that walks out of the clearing. Where a team prototypes five ideas, the pipeline prototypes five million in simulation — and returns the handful worth building, with the reasoning attached.
We are not a neural-net shop. The models we build know meshes, vector fields, contours and wavefronts — physics, held in a form a machine can search.
The definition, drawn
Space Unknown n.
The set of solutions to a design problem that are entirely valid — physically lawful, manufacturable, often elegant — but that no human mind has ever reached. Not mystical: unknown only in the sense that nobody has stood in it.
Space Unknown = Solution Space − Reachable Space
The frame breaks at the top right because the territory has no far edge.
02
Model architecture
Four capabilities that compose into one pipeline. A billion candidates across the top, one decision at the bottom.
Generative geometry
Topology and form search that proposes candidates well outside any training tradition, with manufacturability treated as a first-class constraint.
Multi-objective search
Twenty or fifty competing requirements held at once — mass, stiffness, thermal margin, build time, cost, qualification burden — where a person can balance four or five. The answer comes back as a Pareto set, so the trade is visible instead of assumed.
Verification & qualification
When a candidate looks unfamiliar, intuition stops working as a sanity check. We replace it with a simulation and test strategy that stands up to review.
03
Where it is put to work
We work where the constraint count is high, the physics is unforgiving, and intuition fails earliest.
Aerospace
Structural and thermal components where every gram carries a cost per kilometre of altitude.
Defence
Mission-critical parts under hard qualification requirements and constrained supply.
Space
Launch and orbital structures where a prototype cannot be recalled and mass is the budget.
Medical
Devices and implants that have to satisfy physics, regulation and a frightened patient's hand.
Automotive
Structure, thermal and packaging problems where twenty objectives compete and none can be dropped.
Robotics
Actuated systems where geometry, control and materials have to be designed as one object.
04
Volume moves to the model. Judgment does not.
When a model proposes ten thousand viable answers, taste matters more than productivity. The question becomes which of these is right for these people, in this place, at this moment — and answering it means knowing what we are optimising for. That is a question about us, not about the design.
The model
What it does
- Enumerates candidates at volume
- Evaluates against physics, process and cost
- Balances many objectives at once
- Searches without fatigue or preference
- Reaches forms outside every training tradition
The engineer
What stays human
- Framing the problem worth solving
- Stating the constraints the model cannot see
- Judging candidates against values and use
- Noticing what the model has missed
- Owning the decision, and its consequences
“Design has always stopped where the mind stops. That was never where the possibility stopped.”