From physical questions
to new science.

I am developing Generative Mechanics to turn AI’s growing capacities for prediction, design, and reasoning into physical understanding and materials we can make.

designing Matter, guiding Making, and inspiring Mind.

Fracture

How can disorder make an atomically thin sheet tougher?

Connect crack paths to the architecture of amorphous and crystalline regions.

Protein mechanics

How does a sequence become a response to force?

Move from molecular structure to motion, unfolding, and mechanical function.

Manufacturing

How does the way we make a material determine what it can do?

Connect composition, structure, and processing to performance.

New sciences, in the making

Powerful predictions.
Open physical questions.

AlphaFold makes the present opportunity concrete. AlphaFold 3 predicts structures of complexes involving proteins, nucleic acids, and other molecular components. This is a major expansion of what researchers can investigate. Yet a predicted structure alone does not explain a folding pathway, a response to force, or the motions that enable function. We need physical descriptions that connect these observations and support intervention.

This is what I mean by having the steam engine before thermodynamics: a powerful capability can arrive before we have general principles that explain its possibilities and limits. Existing physics remains essential. The opportunity is to develop the additional understanding that new capabilities and new phenomena demand.

My mission begins with those physical questions. Generative AI can capture complex patterns and open a wider space of hypotheses. AI reasoning can help formulate and execute tests. Physics-based models can identify mechanisms and compress observations into relationships. Experiments and independent calculations decide whether those relationships hold.

01 / Designing

Matter

Capture complex behavior. Find the principles within it.

Proteins, disordered solids, and alloys challenge descriptions that depend on a few idealized structures or well-separated scales. Their behavior can emerge from heterogeneous environments, competing mechanisms, and histories that are difficult to represent together.

I bring generative models and physics into a repeated exchange. A learned model helps explore patterns and candidate systems; a physical model asks which variables and mechanisms explain the response. The next test can refine both. My work on amorphous carbon and 2D titania supplies a grounding in material mechanisms, alongside the possibilities opened by protein design.

The ambition is new understanding of complex materials at scales where our current approaches struggle: relationships that connect structure to behavior, reveal their limits, and guide a change we can actually make.

02 / Guiding

Making

Make manufacturing part of materials design.

Performance is a consequence of composition, structure, and processing together. Learning from nature means asking how a material is formed as well as what it is made of. Spider silk, for example, motivates questions that connect molecular building blocks, assembly, spinning, and the resulting mechanical response.

For engineering, manufacturing belongs inside the design problem. Printing can introduce structures and failure modes that an idealized material model misses. Those difficulties are also opportunities for discovery: the reality of making reveals physical questions we might otherwise never ask.

AlloyGPT and the AlloyGen preprint explore this connection through generative design and physics-based workflows for additive manufacturing. I want AI and physical principles to work together throughout the process, connecting a proposed design to the conditions under which it can deliver its intended performance.

03 / Inspiring

Mind

Use AI as a mirror for intelligence—and for science.

AI gives us an unfamiliar participant in activities we associate with understanding: perceiving patterns, interpreting evidence, and constructing arguments. Comparing its capabilities with our own can help us examine how learning, collaboration, and reasoning depend on the systems that perform them.

I want to study intelligence through task performance and its costs, with definitions that permit meaningful comparisons across physical systems. I want to test when diversity and differentiation improve collective reasoning, and build environments in which learning and the evolution of capabilities can be investigated. These are open research questions, rather than claims that one universal metric or collaborative architecture is already established.

There is also a human question. I see science as understanding we can operate: compact, transferable explanations that fit within—and extend—our ability to reason. Powerful prediction makes this ambition more valuable. It asks us to turn capabilities into principles that humans can examine, use, and share.

One mission, three connected directions. The behavior of matter gives us questions. Making exposes new constraints and phenomena. Studying intelligence changes how we investigate both. I am building GenMech Lab as the scientific identity for this developing research vision.

See the work behind the vision