About

I am a Senior Machine Learning Scientist at Matterworks, where I build foundation models for mass spectrometry. My work sits at the intersection of AI × Science: building, scaling, and deploying models that learn from messy, high-dimensional scientific data. At Matterworks, I developed the core architecture and self-supervised pretraining strategy for the Large Spectral Model (LSM), and led the development of models for small molecule identification.

Drawn to hard problems across STEM, I began my career with a B.S.E. in chemical engineering and a minor in computer science from the University of Michigan. During undergrad, I explored a wide range of problems, from leading teams developing aerodynamic wind turbine blades to working at a large industrial gas supplier, before joining ExxonMobil to model refinery kinetics. As I saw the potential for machine learning to transform scientific modeling, I returned to school for an M.S. in computer science at the University of Wisconsin. My graduate research in the Van Lehn Laboratory focused on geometric deep learning for molecular dynamics simulations.

Set on pursuing frontier research at the intersection of AI and science, I joined Ford as an ML researcher working on an automated materials discovery lab. That work led to publications at NeurIPS workshops, including a spotlight talk. I later joined Matterworks as its third ML researcher, drawn to the opportunity to build a research-driven product from the ground up. Since then, I have explored a field of mass spectrometry largely untapped by machine learning, taking research projects from initial ideas through customer deployment.

My key research interests include:

• Foundation models for scientific data: Multimodal pretraining over spectra, molecules, and structure, and understanding when those representations transfer across tasks.
• World models: Learned simulators of physical and chemical systems that are accurate enough to design against, rather than simply predict with.
• Grounding models in physics: Incorporating physical laws into model architectures and training objectives, and understanding when doing so actually helps.
• Automated discovery loops: Closing the loop between generative models, simulation, and experiment so that systems can propose and test their own candidates.
• Machine learning for engineering and hardware: Applying modern ML methods to process, mechanical, and hardware problems where classical modeling still dominates.

Above all, I am motivated by difficult problems with the potential for positive impact, and by the opportunity to learn something new along the way. My diverse background across various disciplines has made me more interested in the problem than the domain itself. Chemistry, biology, physics, or something I have not worked in before — if the problem is interesting, I am interested.

Always happy to connect and chat!