I build tools. I'm a third-year Mathematics and Engineering student at Queen's University, in the Computing and Communication option, from Toronto.
AI is going into everything, and most of my projects come back to one problem: understanding what these systems are actually doing, and building the checks that keep them compliant and accountable. I'm lead author of a paper on testing interpretability methods in vision-language models, accepted at a NeurIPS 2026 workshop. I also work on q-deformed numbers as an NSERC undergraduate researcher. At QUBE, where I'm Director of Capital, I co-built the club's portfolio competition app, which rolls out this term.

Researchers use tools that look inside AI models to check why they answer the way they do. We tested one of the most common, along with its standard placebo check. The tool claimed to find how a model sees color. It had really found a shortcut to answer "A", and the check passed it anyway.

Language models propose and referee proofs on q-deformed real numbers, and a claim only counts once exact computation, Lean 4 and a human read agree. Across 668,745 numbers, a 1.000 classifier score turned out to be a 512-term window artifact; rechecking at 4,096 terms left just √2 and √3, matching the theorem we proved.
Queen's undergraduate business and engineering club. I co-direct the Capital team and co-built the club's portfolio competition app, which rolls out this term.
Used machine learning and large-scale computation to look for patterns in pure mathematics, with Profs. Dimitrov, Paquette and Wehlau. More
Tutored mathematics and computer science.






Understudy and Roleweave are built by Carrel Labs.


