Biography

I’m an Applied Research Engineer at The Wellness Center @ LA General Hospital, where I build AI and statistical systems for overdose surveillance in Los Angeles County: nowcasting neighborhood-level overdose deaths months before official records finalize, detecting emerging drugs in medical examiner data, and deploying fine-tuned LLMs to extract substance use history from 1M+ electronic health records.

I earned my PhD in Computer Science at UCLA in 2025. My dissertation, From Metrics to Meaning, tackled a big challenge in AI: how do we really know if an AI system is reliable, useful, and trustworthy? I explored new ways to evaluate them, looking closely at everything from data diversity, quality, and provenance to how people truly experience AI-generated work.

I was fortunate to be advised by Prof. Miryung Kim and to collaborate with incredible mentors like Prof. Nanyun (Violet) Peng (PLUS Lab), Prof. Quanquan Gu (Statistical ML Lab), Prof. Amit Sahai, and Prof. Muhammad Ali Gulzar.

During my PhD, I got to dive into some fascinating areas:

• Making AI More Robust & Secure: This meant stress-testing models through fuzzing and adversarial attacks, refining AI testing metrics (like looking beyond neuron coverage), and for AI content provenance, critically re-evaluating theoretical limits on watermarking—showing that robust techniques for verifying authorship can be more resilient and quality-preserving than previously assumed.
• Innovating with Data: I worked on new ways to augment data (like sibylvariant transformations and feature-aware methods) to make AI models smarter and more generalizable and robust to adversarial attacks.
• Understanding Human-Perceived AI Quality: I led research measuring subjective qualities like engagement, authenticity, and empathy through the stories that LLMs tell, which received an Outstanding Paper Award at EMNLP 2024 (top 0.25% of submissions). We also used these ideas to build specialized models for applied tasks, like high-accuracy drug detection in medical texts, and to think about cultural inclusion in vision language models.

I also continue independent research on LLM watermarking and on measuring linguistic diversity in generated text.