i am a final year phd student at mit eecs, advised by professor tamara broderick; i also collaborate with amazing folks including mitchell gordon, jacob andreas, and omar khattab.
my work has focused on surfacing false certainty in ai and making failure modes visible to the humans overseeing them: whether in ai evaluation, in the invisible sediment of long conversation histories, or in one-sided, confident answers to open-ended questions. my background in statistics and probability has helped shape how i approach these problems.
most recently, my work at google has been on making ai agent reports more honest (project landed in gemini post-training!)
i am grateful to be supported by the amazon ai research innovation fellowship and ibm research, and (prev) the mit presidential, fodsi, and quad fellowships.
selected publications
recent news
- oct 2026 — i will be a cbai research fellow in ai safety, collabing with nikola jurkovic (metr).
- oct 2026 — i will be at colm presenting our work, “do llms benefit from their own words?”.
- sept 2026 — i will be participating in the exploration phase of neel nanda’s mats stream.
- aug 2026 — i gave a talk at the stanford ai measurement seminar.
- may 2026 — i will be a student researcher at google (mtv office).
- apr 2026 — i have been selected as an amazon ai research innovation fellow for the 2026-27 year!
- apr 2026 — i will be attending iclr 2026 🇧🇷 — at the main conference and then at the memagents workshop.
- apr 2026 — i will be co-organizing the wids cambridge datathon on april 18th!
- feb 2026 - our work on the robustness of llm leaderboards was featured on the mit news!
thoughts lately
- slow ai: ai that meets a human’s pace - on designing ai systems that encourage slow thinking.
- simple ai - a proposed inference-time framework for reducing ai slop.
- ask your ai agent to be honest - experiments revealing how ai-generated accounts of work may be deceiving.