About

I am a researcher and entrepreneur working at the intersection of machine learning, interpretability, and scientific applications. Currently, I'm preparing to begin my DPhil in Computer Science at Oxford as a General Sir John Monash Scholar, where I'll continue my work on understanding and improving AI architectures.

My research focuses on mechanistic interpretability and meta-learning, particularly in transformers and language models. I've published on topics ranging from sparse autoencoders for neural embeddings to category theoretical perspectives on self-attention. I've had the privilege of working with Neel Nanda at Google DeepMind through the MATS program and David Klindt at Cold Spring Harbor Laboratory, exploring the intersection of neuroscience and machine learning interpretability.

Beyond academic research, I was the co-founder of Rake News, where we were building a peer-to-peer personalised news platform using LLMs for debiasing and synthesising news stories. This evolved into a provider of enterprise-grade RAG software solutions. I've also written software for statistically arbitraging Australian bookies against Betfair (a betting exchange). I'm now working on Iris, a stealth project applying mechanistic interpretability to medical applications.

As a Tuckwell Scholar at ANU, my background spanned both theoretical and applied machine learning, including roles at IMC Trading, Macuject (AI for ophthalmology), CSIRO and ICEDS (climate forecasting), and collaborations with universeTBD on applying language models to scientific discovery. I did further research through funded positions at the University of Sydney and Johns Hopkins University.

I'm particularly interested in two core questions: understanding the architectures of learning and meta-learning, and leveraging these insights to build better AI systems.