I'm currently researching mechanistic interpretability with support from Emergent Ventures. Before this, I was doing cybersecurity development at Allbirds. I really like understanding how things work at a fundamental level, whether that's neural networks, compiled binaries, or geometric structures.
Built security tools during my internship at a unicorn. Founded an AI education platform during my A levels that reached 500+ users before compute costs forced a shutdown. Explored sparse dictionary learning for interpretability before a better implementation of the method emerged. Previously hit Radiant 8x on Valorant EU servers. Currently focused on mechanistic interpretability research, trying to understand what and how neural networks actually learn.
Reverse engineering fascinates me. There's something beautiful about taking apart compiled code to understand its logic. I spend most of my time on mechanistic interpretability, reverse engineering neural networks. Mathematical art, computational geometry and frontend web development are my creative outlets. Always looking for elegant solutions to complex problems.
Building IDA Pro for deep neural networks. Just finished first year of my cybersecurity degree at the University of Gloucestershire with a 73% average. Two more years to go while I work on making neural networks as debuggable as compiled binaries.