I am currently working as a Machine Learning Engineer at
Mantel, and co-supervising a Master’s thesis at the University of Melbourne.
In 2025 I spent 8 months overloading coursework and research to learn more about LLMs and AI safety. I previously worked as an engineer on guidance, navigation and control (GNC) vehicle systems.
Writing
-
Aug 2026
Visualising Refusal as One Direction
A reconstruction of the famous "refusal direction" result, aimed at learning. Every figure here is interactive, to help you intuit this result for yourself.
-
Jan 2026
Expert Collapse and Compositional Failure in Simple Multimodal MoE
Supervised by Caren Han · AAAI 2026 Workshop on Bias in Multimodal AI
· Code
Projects
-
Sep 2026
The Effect of Peer-Framing on Model Behaviours and Beliefs
A time-boxed (~15 hour) research project for Neel Nanda's MATS 12.0 stream, testing
whether peer-framing shifts a model's harmful compliance and belief once refusal is ablated.
