_
Computer interpretation of handwritten math is a challenging problem due to the variablity in size and placement of many characters. I created and explored the viability of a novel computer-vision algorithm to solve this problem.
Read the paper
Generative AI is flexible but unreliable, while code-based theorem provers are always correct but hard to use. By combining the strengths of each, my research group and I are creating a system to automatically optimize and clarify formal proofs, along with other tools and infrastructure to assist mathematicians.
Check out the ImProver repository
What started as messing around out of curiosity one afternoon has turned into a year and counting working with the Carnegie Mellon Rocket Command, my school's competitive rocketry team. I've developed and implemented data collection, filtering, and control algorithms on embedded systems to keep our rocket on target as we demonstrate its capabilities to NASA.
A bit of what we do
I've found that the most interesting applications of science are those with real-world impact. My team and I worked with the maintainers of Pittsburgh's public riverfront trails to analyze and predict traffic patterns to help maintainence and future expansion.
About our organization
Myself and a co-author identified a new form of jailbreaking attack in large language models, and analyzed its effects on models' reasoning. Our paper has been accepted for publication at AAAI 2026.
The latest on our paper