My Story
I'm currently on garden leave and was previously a quantitative researcher at Two Sigma. I received my PhD from CMU LTI in 2023, and before that spent four years at Bosch Center for Artificial Intelligence.
Over the past decade, I've worked on deep learning systems across audio, speech, and multimodal settings, with a consistent focus on robustness and failure modes—understanding not just when models work, but how and why they break.
Real-world systems are messy—inputs are incomplete, distributions shift, and signal-to-noise ratios are low. I build models that remain useful under these conditions, rather than optimizing for idealized benchmarks.
With LLMs, the challenge has shifted: they are more capable but harder to reason about. I focus on probing their behavior, adapting them through post-training, and turning them into reliable components for decision-making systems.
Over time, I've come to view research as a process of allocating attention under uncertainty. Progress rarely comes from chasing every new idea, but from recognizing which directions are worth committing to early—and going deep enough to extract real value.
Ultimately, I care about developing taste and intuition—knowing which problems matter, which signals are real, and when a new capability is worth taking seriously. As the field accelerates, these become more important than any single technique.
LLM Post-Training
Pioneering RLHF workflows and alignment strategies to enhance reasoning and verifiable logic in frontier models.
Robust AI
Building systems that maintain high performance and safety under adversarial conditions and distribution shifts.
Audio & Multimodal
Exploring the integration of complex sensory data to create more grounded and versatile intelligent agents.