Physics and AI, applied to industry
Neural Continuity Labs builds physics-grounded AI that reconstructs what the sensors miss: the state of an industrial system everywhere, not only where it is measured, with uncertainty an engineer can sign off on.
The real world, unlike the simulated one, is rich and noisy. It is not a spherical cow. We build for that world: every model we train is a hypothesis about a physical system, and every deployment is a test of it. We keep what survives.
A generative model trained on simulated physics across many regimes, so it arrives at a plant already knowing how fields behave.
It is then tuned on this plant: its geometry, its history, its sensor feeds. Measurements carry what no simulation does: wear, drift, the habits of this particular machine.
Every estimate comes with its uncertainty. An engineer should know when to trust the model and when to walk over and look.
Plants grow, age and add components. The sensors never keep up, and faults happen in the gaps in between.
We want to build superintelligence for science.
A system that proposes hypotheses about the physical world, tests them against measurements, and keeps what survives. We start in industry, where every deployment is a falsifiable claim.
Neural Continuity Labs grew out of doctoral research in AI and 3D reconstruction at Université Paris-Saclay.
It is founded by Akash Malhotra and supported by Inria Startup Studio.
The name is the research program behind the company: Neural Continuity, the study of differentiable computation as a regime with a boundary.
If you run a plant with a blind spot you would like to see into, write to us. Pilots, partnerships and questions are welcome.