CV

Research Intern

Hitachi Central Laboratory

• Assessed Chain of Thought (CoT) faithfulness via control experiments and unlearning techniques
• Accepted to ACL 2026 Student Research Workshop (oral)

Research Intern

Qatar Computing Research Institute

• Extracted output language features in Cross-Layer Transcoders (CLT)
• Developed a novel dataset for antonym prediction and closed-set enumeration
• Accepted to EMNLP 2026 (Main)

Engineering Intern

SB Intuitions

• Applied GRPO to Sudoku and competitive coding with NeMo-RL (~90% Sudoku accuracy on Qwen2.5 7B)

Software Engineering Intern

Fixstars Corporation

• Built a memory consumption estimator for Llama 3 architectures
• Automated evaluation with Bash and Python on a Megatron-LM environment

Research Assistant

Andlinger Center for Energy and the Environment, Princeton University

• Developed a web platform visualizing energy transition bottlenecks
• Extended an open-source traffic simulator with custom JS/CSS/HTML visualizations

B.A.

Princeton University

Computer Science

• Minors: Applied and Computational Mathematics, Humanistic Studies, Statistics and Machine Learning
• Cumulative GPA: 3.82/4.0; Major GPA: 3.97/4.0

B.S. (partial)

University of Tokyo

Computer Science

• Enrolled for a semester and completed 15 credits toward the Bachelor of Science

Deconstructing In-Context Learning (Prof. Tom Griffiths)

Princeton University

• Analyzing memorization and generalization failures in ICL and improving them via interventions

MLE Agent (Prof. Karthik Narasimhan)

Princeton University

• Testing how SFT improves metrics and agent behavior on MLE Bench

OccludeBench (Prof. Zhuang Liu)

Princeton University

• Benchmark testing VLMs’ ability to recover fully occluded objects
• Accepted to Computer Vision in the Wild, CVPR 2026

Compute–Supervision Tradeoff in RLVR

Princeton University

• Studied how rollouts compensate for noisy verifiers in GRPO
• Accepted to CTB (Combining Theory and Benchmarks), ICML 2026

Guess the Unified Model

Princeton University

• Classifier attributing image origins across 7 open- and closed-source models
• Accepted to ICLR 2026 DATA-FM workshop

Fast Convolution of Toeplitz Kernel (Prof. Marc Gilles)

Princeton University

• Optimizing Chebyshev-series convolution from O(n³) toward O(n²)

Toshin Study Abroad Scholarship

Toshin

95,000 USD/year for four years of undergraduate study abroad

OURSIP Summer Grant

Princeton University

7,200 USD for on-campus summer research

Conference Travel Grants

Princeton University

ICLR 2026 (1,250 USD from university); ACL 2026 (600 USD from university + 750 USD workshop grant)

Co-President

Japanese Student Association, Princeton University

Member

Men's Club Soccer, Princeton University

Working Unit Member

Civic Leadership Council, Princeton University

Core Member

Mielka (voter engagement nonprofit, Japan)

Captain

Varsity Soccer Team (High School)

Peer Reviewer

ICLR 2026 DATA-FM Workshop