I am a Research Scientist at the Wharton AI & Analytics Initiative, working with Hamsa Bastani, Osbert Bastani and Dean Knox.​

​

I led the research, development, and deployment of AI systems in high-stakes domains such as healthcare and education. My expertise spans several areas of AI:

1. LLM Pretraining & Post-training: Pretrained large encoder models with masked language modeling; applied RL (e.g., GRPO) and SFT to improve reasoning, alignment, and personalization.

2. LLM Uncertainty & Safety: Developed algorithms to quantify hallucination risk and detect incomplete information in high-stakes domains.

3. LLM Benchmark & Evaluation: Designed and collected new benchmark datasets for LLM tutors.

4. LLM Deployment: Deployed an LLM tutoring system used by 600+ students over a school year.

​

My research has been published in leading scientific venues such as ICML, PNAS, and ITCS. It has had significant impact, including being cited in the 2025 Economic Report of the President, submitted by the U.S. President to the Congress.

​

I earned my Ph.D. in Politics at Princeton University with a specialization in quantitative methods, where my research focused on statistical modeling and machine learning in social science contexts.  As part of my doctoral work, I pretrained a transformer encoder model, TESS, on a large social science corpus and released it on Hugging Face, where it has been downloaded over 1,100 times.

​​