Research
I work on post-training and evaluation of LLMs, with a focus on agents operating in open-ended environments on tasks that have no single best answer. My work includes scaling data pipelines for reinforcement learning and creating evaluation benchmarks that measure the scientific reasoning capabilities of LLM agents.
Before moving into machine learning, I completed a PhD in physics at Berkeley, where (with my collaborators) I helped to develop the Quantum Focusing Conjecture, and to formulate and provide the first proof of the Quantum Null Energy Condition. I later worked at the Perimeter Institute as an It from Qubit fellow, studying information transfer in evaporating black holes.
Machine Learning
Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities. arXiv:2507.06261 (2025).
Gemini 1.5: Unlocking Multimodal Understanding Across Millions of Tokens of Context. arXiv:2403.05530 (2024).
STRUM-LLM: Attributed and Structured Contrastive Summarization. arXiv:2403.19710 (2024).
Measuring an LLM’s Proficiency at Using APIs: A Query Generation Strategy. KDD-ADS (2024).
ReST meets ReAct. arXiv:2312.10003 (2023).
Impakt: A Dataset for Open-Schema Knowledge Base Construction. arXiv:2212.10770 (2022).
Making Transformers Solve Compositional Tasks. ACL (2022).
LogicInference: A New Dataset for Teaching Logical Inference to Seq2Seq Models. arXiv:2203.15099 (2022).
ETC: Encoding Long and Structured Inputs in Transformers. EMNLP (2020).
Physics
Evaporating Black Holes Coupled to a Thermal Bath. JHEP 2021 (1), 65.
Information Flow in Black Hole Evaporation. JHEP 2020 (3), 152.
A Semiclassical, Entropic Proof of a Weak Gravity Conjecture. arXiv:1706.08257 (2017).
Quantum Focusing Conjecture. Phys. Rev. D 93, 064044 (2016).
Proof of the Quantum Null Energy Condition. Phys. Rev. D 93, 024017 (2016). Editors’ Suggestion.
Entropy on a Null Surface for Interacting Quantum Field Theories and the Bousso Bound. Phys. Rev. D 91, 084030 (2015).
Proof of a Quantum Bousso Bound. Phys. Rev. D 90, 044002 (2014).
Pulsed Multipole Injection for the ALS Upgrade. Proceedings of IPAC (2010), 2642.
Patents and Disclosures
Artificial Intelligence Generated Badges for Search. U.S. Patent Application 18/636,844 (2024).
Automated Optimization of Structured Large Language Model Prompts via Iterative Evaluation. Technical Disclosure Commons (2026).