Research-led physics AI: engineering simulation software.

Our model learns from simulations your team has already run and operates entirely within your secure infrastructure. Your data stays yours.
Fast enough to explore. Accurate enough to build from.

 See it in practice
How it works

How NPco works

A 3D mesh goes in. The optimal solution comes out. What happens in between is the architecture we built, purpose-designed for the physical reality you are testing.

1

Upload mesh

STL, VTK, or any standard format. Auto pre-processing and tokenization.
2

Fine-tune

50 simulation samples. Adapts to your physics domain in 24h on one GPU.
3

Predict

Sub-minute inference. Full-resolution pressure, temperature, stress fields.
4

Validate & export

Compare against reference runs. Export to your post-processing stack.
In practice

Simulate in seconds. Trust the result.

Thermal, structural, fluid. Across any geometry, at any scale.

THERMAL
Wireframe mesh of a car for structural stress simulation

Car engine

Structural stress simulation across complex, irregular assemblies. Adapts to any geometry without retraining from scratch.

1,000x

Faster

Same

Accuracy

24h

To train
CFD
Wireframe mesh of a turbine for 3D CFD of blade aerodynamics

Turbine

Full 3D CFD of blade aerodynamics and thermals. Benchmarked and validated against Safran's internal standards.

55ms

Inference

0.4%

Error

50 samples

Needed
FATIGUE
Wireframe mesh of a drone for fatigue and deformation analysis

Drone

Fatigue and deformation analysis on irregular geometries. No retraining needed across design variants.

55ms

Inference

50

Samples

1 GPU

24h
Performance

Numbers that change what's possible.

Samples

50

Pretraining eliminates the cold start. Adapt simulators to your geometry and physics in hours.
Less than

1 min.

Sub-minute inference for pressure, velocity, and temperature across meshes 100M+ nodes. No coarsening. No approximation.
Nodes

100M+

No mesh is too large. Handle industrial geometries that choke conventional solvers, turbine blades, crash structures, reactor components.
WHY NPco

The physics hasn’t changed. The tools finally have.

Traditional solvers gave you the right answer, eventually.
Earlier AI gave you a fast guess. NPco gives you both.

Accuracy
Speed
Full Assembly
Training Data
Traditional solvers
Accuracy
HIGH FIDELITY
Speed
WEEKS TO MONTHS
Full Assembly
COST PROHIBITIVE
Training Data
HIGH COMPUTE COST
NPco
Accuracy
HIGH FIDELITY
Speed
UP TO 50,000X FASTER
Full Assembly
ROUTINE
Training Data
AS FEW AS 50 SAMPLES

FAQ

What engineers ask before a first benchmark.

How accurate is NPco compared with a classical solver?

The best way to judge accuracy is on your own case. We benchmark on a problem your team has already solved, and you compare our predictions side by side with your reference results.

What data does NPco need to get started?

The simulations your team has already run. NPco's models learn from your own past simulations and adapt to your physics domain.

How does NPco fit with our existing CAE tools?

NPco takes the same 3D meshes your team already uses and returns full-resolution fields you can export to your post-processing stack. It connects to existing CAE workflows through an API.

How do we evaluate NPco?

Bring a case your team has already run. We train on it and show you what the model does, side by side with your reference results. Talk to an engineer to set it up.

The rest is a conversation.

If you’re an engineer evaluating us against your pipeline, or a researcher who wants to build the models, write to us.

Talk to an engineer