● Research · Machine Learning

Surrogate Model Sandbox

The materials-research workflow in miniature: sweep a physics engine to build a dataset, then train a small neural network in-browser to predict it. Watch the surrogate’s surface converge to the true physics, live, with R² → 1.

No install · No signup · Runs in your browser. Educational and research tool — not a medical device.

Surrogate Model Sandbox
Surrogate
Materials
Testbench
Movement
Train a surrogate
Design sweep
Sample the physics
DOE
Training loss
Falling live
Surrogate fit
vs true physics
R²→1
Prediction
Instant inference
ms
Illustrative — live values update as you change the inputs

What you can explore

See how a fast ML emulator is built on top of a slow physics engine — the modern R&D loop.

Sweep the physics

Run a design-of-experiments sweep over the engine to generate a training set.

🧠

Train in-browser

A small neural network trains live — watch the loss fall epoch by epoch.

📉

Convergence (R²)

See the surrogate surface approach the true physics surface as R² climbs toward 1.

Instant prediction

Once trained, the surrogate predicts in milliseconds where the physics took seconds.

How it works

The build-a-surrogate loop, made visible.

1

Sweep

Choose the parameter ranges and sample the physics engine to build a dataset.

2

Train

Fit a small neural network to the data, watching the training loss fall.

3

Compare

Overlay the surrogate against the true physics and read R² as it converges.

Ground-truth in the loop

The surrogate is trained on, and scored against, the same validated physics engine that powers the other tools — so the convergence you watch (R² → 1) is measured against real ground truth, not a synthetic toy.

Who it's for

For teaching the ML-emulator workflow with a transparent ground truth.

🔬

Educators & researchers

A transparent, citable model for teaching and study. Every value traces to a published equation.

🎓

Dental & ortho students

Build real intuition before the clinic — change a parameter and watch the mechanics respond, instead of memorizing rules.

🏭

Manufacturers & R&D

Translate material and design parameters into the force system delivered — a virtual bench before you cut metal.

Frequently asked questions

What is a surrogate model?

A surrogate (or emulator) is a fast machine-learning model trained to mimic a slow, expensive simulation. Once trained, it predicts results in milliseconds — useful for optimization and design sweeps. The sandbox builds one live.

Does training run on my machine?

Yes — the neural network trains in your browser, so you can watch the loss fall and the fit converge without any server.

Is it free? Do I need to install anything?

It's completely free, with no signup, and runs entirely in your browser — nothing is downloaded or installed.

Is this a medical device?

No. It is an educational and research tool for training machine-learning surrogate models on synthetic, user-generated data. It does not use patient data and is not intended for the diagnosis or treatment planning of individual patients.

Open the simulator

Free, in your browser, right now.

Launch the Surrogate Sandbox →