Eugene Park

I am a software engineer at the Personal Robots Group of MIT Media Lab, where I focus on language model applications for educational tools (e.g. AI tutors). I design and develop system features that support personalized learning pathways for students and evaluate pedagogical effectiveness of language models.

My broader research objective is to advance our understanding of AI models and mitigate unintended or harmful behaviors, ultimately to provide reliable and personalized support for human decision-making under uncertainty. I am especially interested in high-stakes domains, such as education and financial decision-making, where model failures have real-world consequences.

In 2023, I graduated from the Courant Institute of Mathematical Sciences at New York University with a Master of Science in Mathematics. Under the mentorship of Professor Kenneth Winston and Professor Jonathan Goodman, I completed a master’s thesis on applying machine learning to uncover patterns and forecast stock returns (preprint). In 2021, I graduated from Boston College with a B.A. in Mathematics and B.A. in Economics (honors). I was advised by Professor Robert Murphy on my senior honors thesis examining how instability in the financial market propagates into the real economy.

News

Aug 2025	Our paper BehaviorSFT: Behavioral Token Conditioning for Clinical Agents Across the Proactivity Spectrum has been accepted at Findings of EMNLP 2025.
May 2025	Our paper VocalAgent: Large Language Models for Vocal Health Diagnostics with Safety-Aware Evaluation has been accepted as an Oral Presentation at Interspeech 2025.
Nov 2024	I started my appointment as a software engineer at the Personal Robots Group of MIT Media Lab!

Research

My research aims to build human-centered AI systems that provide reliable and personalized support for decision-making under uncertainty. To this end, I focus on three complementary research directions:

• Interpretability: Evaluating model behaviors and dissecting internal model architectures to advance our understanding of how models represent information and carry out reasoning.

• Alignment: Identifying embedded biases and failure modes and developing methods to mitigate harmful or unintended behaviors.

• Personalization: Developing models that surface meaningful, domain-specific insights to users while aligning model behavior with individual human preferences.

Selected Publications

ICLR 2026 (Under Review)	InvThink: Towards AI Safety via Inverse Reasoning Yubin Kim, Taehan Kim, Eugene Park, Chunjong Park, Cynthia Breazeal, Daniel McDuff, Hae Won Park Open Review, ArXiv
EMNLP 2025 Findings	BehaviorSFT: Behavioral Token Conditioning for Clinical Agents Across the Proactivity Spectrum Yubin Kim, Zhiyuan Hu, Hyewon Jeong, Eugene Park, Shuyue Stella Li, Chanwoo Park, Shiyun Xiong, MingYu Lu, Hyeonhoon Lee, Xin Liu, Daniel McDuff, Cynthia Breazeal, Samir Tulebaev, Hae Won Park ArXiv, Code & Model
Interspeech 2025 Oral Presentation	VocalAgent: Large Language Models for Vocal Health Diagnostics with Safety-Aware Evaluation Yubin Kim, Taehan Kim, Wonjune Kang, Eugene Park, Joonsik Yoon, Dongjae Lee, Xin Liu, Daniel McDuff, Hyeonhoon Lee, Cynthia Breazeal, Hae Won Park ArXiv
Master's Thesis 2023	Principal Component Analysis and Hidden Markov Model for Forecasting Stock Returns Eugene Park ArXiv

Selected Projects

Contact

Email: ewp@mit.edu