Technology
onepot Data
Models are only as good as the reactions behind them. We generate reaction data on POT-2: uniform, traceable, and dense in chemistry, where public datasets are sparse.
Negative results included
Public datasets overrepresent successful reactions. Ours captures the full distribution: failed reactions, partial conversions, and side products, because those outcomes teach models what not to predict.
Reaction-class expansion
Coverage includes all reactions in CORE v1.1: amide coupling, Suzuki-Miyaura coupling, Buchwald-Hartwig amination, CDI-mediated urea synthesis, TCDI-mediated thiourea synthesis, N-alkylation, and O-alkylation. New reaction classes are added every quarter.
Linked to building blocks
Every reaction is tied to physical building blocks with known purity and provenance. Conditions are stored as structured presets, not free-text procedures.
Powering CORE and C1
The same data backs ML feasibility scoring in onepot CORE and conditional generation in onepot C1.
Each color is a different reaction type · drag to pan · scroll to zoom