Neuralectic Systems Inc.

The model proposes. Physics decides.

Engineers should be working at the limits of physics. Instead, they are constrained by expensive simulations, fragmented tools, and slow design cycles. There are always more designs worth exploring than a team can afford to evaluate, so what gets built is often the best design they had time to check rather than the best design physics allows.

Frontier AI can close this gap, but engineering demands more than predictions. Engineers need to trust the reasoning that takes a system from requirements to something that can actually be built. Software and mathematics have shown what becomes possible when powerful models work with tools that independently verify their outputs. Engineering needs the same principle.

Neuralectic is building engineering superintelligence that is verifiable by design. Our platform gives AI a structured understanding of engineering systems, from requirements and components to dynamics and underlying physics. It brings together the models, tools, and checks needed to reason across the design process, evaluating results against numerical evidence and physical constraints. The model proposes, physics decides.

By making verification part of the intelligence itself, engineers can explore more ideas and build on what they have learned. Our goal is to connect the entire workflow, from initial requirements to final physical design, making systems as complex as a fighter jet 100 times faster to design at a tenth of the cost.

We met at Oxford during our PhDs in AI for physics and robotics. Between us, we have studied at Oxford, Berkeley, and MIT, and worked at NVIDIA Research, Daimler, and Siemens. Our research spans the world’s fastest GPU-accelerated quantum-dot simulator, learned dynamics for real-world robotic systems, open problems in relativistic quantum mechanics, and more than a dozen papers at ICLR, NeurIPS, and the Physical Review journals.

We’re building Neuralectic in San Francisco.

The Neuralectic team