02.1Ai
Atomic AINow building: Labs for Future Materials, San Francisco Bay AreaRead more →
Home / Our approach / Atomic AI
Metal-expert AI that designs a new alloy in minutes.
Trained on 20 years of real production data and every result from our labs, not just on public literature.
What it does
From target spec to candidate alloy.
Atomic AI agents propose alloy compositions and processes for a target spec, then predict their properties for the way the part will actually be made, such as high-pressure die casting.
Watch the agent · inverse design
Physics-informed models
Predictions that follow metallurgy.
ICME physics plus real production data feed our physics-informed ML (PIML) platform. This directional learning builds accurate models from little data, so predictions respect how metals behave and not just patterns in text.
Predicted vs measured on held-out alloysEach dot is one alloy. Dots on the dashed line are perfect predictions.
Metal strength MPa · measured →, predicted ↑
Metal ductility % · measured →, predicted ↑
The data flywheel
Every result makes the next prediction better.
01
Open data
Agents harvest open-source data for new metals.
02
PIML check
Physics-informed models filter and correct it.
03
Synthetic data
The models generate data where real data is thin.
04
Reinforcement
RL tunes the models to match production behavior.
05
Lab truth
Dynamic Labs results close the loop.
Ohio labAgents that run the lab
It doesn’t just propose. It runs the test.
Our AI agent is taking over experiments in the Ohio lab, so every prediction is tested in metal, on the same kind of machines that make the part.
Start a programBring us a spec.
Bring us a spec.
We’ll bring you a part.
Tell us the part, the load case and the process. We’ll tell you how fast we can get there.