In the 2022 SERI Summer Research Fellowship with Stephen Casper as my mentor, eventually publishing the paper Explore Establish Exploit: Red Teaming Language Models from Scratch. My role on the project primarily involved setting up evaluation pipelines, communicating our methodology, and designing the survey and collecting responses resulting in the CommonClaim Dataset.
In 2024, my colleague and I completed a small interpretability research project measuring the similarities between embedding model representations and the error arising from stitching embeddings from one model into the embedding space of another model.
In 2025, I had a limited collaboration with METR and the MIT Algorithmic Alignment Group on developing an E2E evaluation determining whether models are able to replicate influential AI R&D papers from project proposal to results & figure generation. I contributed to methodology brainstorming, reviewing the final report, and integrating SWE-Agent and UK AISI’s Inspect framework (influenced Inspect’s development and had some minor successes).
A majority of my research and non-research experience is heavy on engineering and light on math and theoretics. However, I’m excited to further explore the latter.