Method: Data → Physics → AI → Decision
Research
Our research is organized by the four stages of our method. We list a research area only when there is working software behind it. Ideas without software yet are listed as open problems on the related research page.
Data
→ Harmonized datasets
We collect data from regulators, national surveys, satellites, clinical trials and the scientific literature, and harmonize it under one schema with the source of every value recorded.
Physics
→ Simulation
We solve the governing equations for flow, heat, mechanics and chemistry on parallel computers, and test the solvers against analytic solutions and published benchmarks.
AI
→ Models and agents
We train models where they can be checked against simulation or measured data, and build agents that plan tasks, call our solvers and record each step.
03 · AIAgentic science
Agents that plan scientific tasks, call simulation and analysis software, and weigh evidence, with human approval required for actions that change data or systems.
- Liveframework and Agent Studio
- In buildproduct agents
Scientific machine learning
Surrogate models, reduced-order models and domain language models, each used only within the range where it has been tested against simulation or data.
- Livesurrogate and reduction library
- Researchphysics-guaranteed surrogates
Decision
→ Decisions under uncertainty
We combine models with measurements using Bayesian inference, so that forecasts include their uncertainty and the next measurement can be chosen by how much it would reduce it.
Verification applies to every stage. Each research page describes how its methods are tested; shared libraries are described on the platform page.