buildbox
building AI capability for the physical sciences
Scientific breakthroughs in the physical world takes more than knowing the literature. It requires choosing the right calculation, interpreting measurements, and reasoning through the next hypothesis when a result is inconclusive. Scientists develop these skills through hands-on research, but much of that experience never lands into a published paper.
We build datasets and environments around practical scientific work. Working with scientists, we identify where models struggle in a research lab, and develop tasks and feedback that make them more capable of real-world scientific work.
the vision
Behind many advancements in everyday life is decades of scientific research. We want AI to power the next generation of scientific work, including the experiments that fail and the ideas that need rethinking. We hope to contribute to scientific progress whose benefits are felt in people’s health, living conditions, and access to basic needs.
our approach
We focus on problems that bring together scientific reasoning, specialized software, and experimental evidence. We capture how an investigation unfolds: what information was available, which approaches failed, and what led a scientist to change direction.
We work with scientists to define what a successful result would look like and how to assess it. That might mean checking a calculation, comparing a prediction with an independent measurement, or running another experiment. Sometimes, it means recognizing that the evidence does not yet support a conclusion.
work with us
We’re looking for scientists and AI researchers to help shape the problems we work on and how we evaluate progress. If you have a research problem you’d like to work on together, .