Basetwo is bringing Physics-informed AI into the manufacturing environment. We are a group of engineers, researchers, and company builders developing learning algorithms that can autonomously predict and control any biological or chemical production process.
A hybrid mechanistic–ML model of a fed-batch mammalian cell culture drives a closed-loop controller that sets feed, pH, dissolved oxygen and temperature in real time, producing a HER2-targeting monoclonal antibody without manual recipe adjustments.
Kinetic parameters for mAb process models, found by refining global search candidates with bounded quasi-Newton optimization.
Data-driven discovery of mechanistic ecosystem models with LLMs: language models propose candidate model structures, TMB fits and scores them against process data.
A network architecture that enforces thermodynamic consistency by construction, so predicted activity coefficients can be used directly in process simulation.
Exact gradients through property models and phase-equilibrium calculations, so thermodynamic simulations can be optimized and trained end to end alongside neural networks.
Work with engineers and researchers building autonomous models for biological and chemical production.
