Decades of R&D, in days.
AI models that discover, design, and predict novel materials — trained on quantum-mechanical physics, validated in the lab.
The Materials Project
A Department of Energy–funded open database of computed properties for 150,000+ inorganic compounds, run on DOE supercomputers — the public data foundation most later generative and graph-network models are trained or benchmarked against.
ALIGNN
The National Institute of Standards and Technology built a graph neural network trained on its own JARVIS-DFT database plus the Materials Project, predicting over 100 material properties from formation energy to phonon spectra — the same core approach later scaled up by GNoME.
GNoME
A graph-network model for materials exploration expanded the catalog of known stable inorganic crystals by roughly an order of magnitude — from tens of thousands to over 384,000 predicted stable structures.
MatterGen & MatterSim
Generative diffusion models — the same underlying approach behind modern image generators — run in reverse on matter: state a target property, and the model proposes an atomic arrangement designed to hit it.
GNN surrogates + DFT-in-the-loop
Graph neural networks now stand in for hours-long DFT simulations at millisecond speed, with active-learning loops routing low-confidence predictions back to full simulation before they reach a shortlist.
Materials Discovery Platform
Generative search over candidate structures, screened against your target properties.
generative + screeningProperty Prediction Engine
Graph neural networks trained on DFT simulation data predict properties in milliseconds.
band gap · eV · GPaSynthesis Route Generation
Proposed synthesis pathways with precursors and conditions for each candidate.
precursors · °C · atmScientific Data & APIs
Curated datasets and inference endpoints that plug into your existing R&D stack.
REST · datasets