Research Projects
My work spans AI for scientific computing, generative modeling, efficient adaptation, multimodal learning, and personalized language models.
NeurIPS 2026 · AI for scientific computing
Breakeven Complexity
Supervised by Prof. Misha Khodak, we developed an evaluation framework centered on breakeven complexity: the number of forward solves required before a learned solver becomes cost-effective relative to an error-equivalent traditional solver.
AI for scientific computing · Plasma physics
Driftless-Star
An interdisciplinary UW–Madison project developing an end-to-end pipeline for transport-consistent stellarator optimization. The work brings together machine learning, computational optimization, plasma physics, and fusion science.
ECCV 2024 · Multimodal generation
Audio-Guided Visual Animation
Supervised by Prof. Pedro Morgado, we developed audio-to-video generation methods focused on producing highly synchronized visual animations from audio guidance.
NAACL 2025 · Model personalization
CHAMELEON
Supervised by Prof. Frederic Sala, we extended AlignEZ to personalize language models for many users while keeping resource requirements low.
Training-free alignment
AlignEZ
Supervised by Prof. Frederic Sala, we developed a nearly cost-free approach to model alignment using self-generated preference data and representation editing—without additional model training.
Honors Thesis · Data-efficient learning
Domain-Specific Fine-Tuning for Generative Models
My senior honors thesis, supervised by Prof. Frederic Sala, studied whether synthetic datasets from fine-tuned generative models can improve downstream classification and how to generate higher-quality data efficiently.