I'm a Founding Member of Technical Staff on the AI Research team at Edison Scientific in San Francisco. I'm primarily interested in agent epistemology: what sort of evidence and arguments convince a language model that a hypothesis is true? I’m particularly interested in using frameworks inspired by Bayesian Experiment Design to explore this question. Most recently, I've been training models to deliberate over scientific evidence to produce calibrated forecasts. I've also co-developed life-science benchmarks such as LABBench2, which OpenAI, Anthropic, and Google DeepMind use to evaluate their frontier models.

Before joining Edison, I completed my Ph.D. in Machine Learning at Carnegie Mellon University advised by Jeff Schneider, where I worked primarily on Bayesian Experimental Design, Reinforcement Learning, and other topics in Uncertainty Quantification. My thesis focused on equivariant neural architectures that significantly improve sample efficiency and generalization in certain Bayesian Experiment Design tasks. I also worked with Alex Beatson at Redesign Science on experimental design for drug discovery.

I first came to CMU through the RISS program, working with Stephen Smith and Isaac Isukapati in the ICL Lab on probabilistic modelling of transportation systems. Before that, I studied Electronic Engineering at UCD as an undergraduate, and spent some time at UCLA as an exchange student.

I like to write and talk about AI research in Ireland, most recently on Ireland's AI research gap for The Fitzwilliam, and as a guest on the Memo to File podcast discussing the compute Irish researchers will need over the next decade. I share posts about meta-science, epistemology, Bayesianism, and scientific progress on lynkmi, and I occasionally host a radio show on WRCT featuring my favourite Pittsburgh and Irish music producers.