Learn the signal, keep the physics.
Why AI for industrial energy has to learn from plant data and answer to physics
Variational Thermodynamic Optimization Through Multi-Agent Orchestration.
How physics-informed multi-agent systems can optimize industrial operations — Ampere's first published research on the approach behind GENIE.
GENIE's measured results — equipment identification, anomaly detection on unseen machines, energy disaggregation, and transferable pretraining — are detailed on the model page. The V2.1 white paper reports the pretraining result in full: about twice the macro-F1 of an untrained encoder, in 5 of 5 seeds.
Physics-grounded anomaly detection (generalizing to unseen machines), equipment identification from power signatures, energy disaggregation, and transferable pretraining beyond models trained from scratch.
Extending the model across more sensor modalities, learning how systems respond to interventions, and widening safe closed-loop autonomy.
Against specialized models we're at parity; the edge is the physics grounding and the gap vs. general AI.