Everything we ship runs on the same chain: first-principles simulation of the atmosphere, measured weather-to-energy data, machine-learning surrogates with honest uncertainty, and deployment on whatever hardware the decision lives on, from an HPC cluster to a device the size of a paperback.
Weather and climate modelling from global reanalysis down to turbine scale. Mesoscale simulation nested into turbulence-resolving large-eddy simulation, which resolves the eddies that load a rotor rather than parameterising them away, plus ocean and wave modelling and coupled atmosphere-ocean runs. Demonstrated down to 33 m and 11 m grid spacing inside a hurricane eyewall.
ERA5 reanalysis, Copernicus satellite products and site measurements, engineered into features that connect what the sky does to what the asset produces, loses and endures.
Coupled structural simulation of energy infrastructure, onshore and offshore: wind turbines and floating platforms, high-voltage towers and overhead lines, foundations, moorings and subsea cables. Fatigue, extreme loads, availability and lifetime, resolved with the same first-principles physics and grounded in the standards we help write at IEC TC88.
Machine-learning surrogates trained on high-fidelity simulation give physics production speed. Probabilistic outputs are calibrated with conformal methods, so an 80% band covers 80% of reality and a reviewer can check it.
Containerised services with standard interfaces (REST, MQTT, CIM-aligned payloads). Models compress to a few megabytes and forecast in under a millisecond, so they run next to the meter as comfortably as in the cloud.
Resolution is reported together with its limits. If you are going to rely on a simulation, you should know which physics it actually resolves, which it approximates, and which it does not represent at all.
On a public four-year household dataset with a fully held-out test year, our probabilistic forecasters cut pinball loss by 30 to 60% against seasonal-naive baselines for both active and reactive power, with empirical band coverage of 80.0% against 80% nominal, from a model bundle under 1 MB running in 0.52 ms per forecast on ARM hardware. The full protocol and figures are published.
Tell us the decision you need to make, and when you need it.
Talk to us