Jeong Woo Lee

GitHub · Email · LinkedIn · Instagram

I will join the University of California, Irvine as a Master of Computer Science student in Fall 2026. I completed my B.S. in Electronics and Electrical Engineering at Dongguk University, where I researched table and chart reasoning for LLMs and VLMs at ML-Lab, advised by Prof. Jihie Kim. This work led to five papers, including an ACL Workshop Best Paper (TabBridge). Alongside research, I've shipped a full-stack AI web service end to end (E-MOJI) and open-sourced a table reasoning tool built on Ollama. I'm interested in understanding both the strengths and limitations of AI systems, and in how human insight can be designed into them to address those limitations. Feel free to reach out via GitHub or email.

Education

University of California, Irvine Sep. 2026 – Dec. 2027 (expected)

Master of Computer Science (MCS) · Irvine, CA

Dongguk University Mar. 2024 – Feb. 2026

B.S. in Electronics and Electrical Engineering (GPA 4.15/4.5, Rank 16/156) · Seoul, KR

Tech University of Korea Mar. 2019 – Feb. 2024

B.S. coursework in Embedded Systems (GPA 4.08/4.5) · Gyeonggi, KR

Experience

ML-Lab, Dongguk University — Undergraduate Research Assistant Jul. 2024 – Feb. 2026

Advisor: Prof. Jihie Kim · Seoul, KR

Led research on structured data reasoning, designing TabBridge (bridging table structure and context for LLM reasoning) and co-developing VisDoT (decomposed visual reasoning for VLMs); first-authored four of five published papers and received an ACL Workshop Best Paper Award. Along the way, built and open-sourced a table reasoning tool on Ollama to run table reasoning experiments on local LLMs without API costs.

Research

TabBridge — LLM Table Reasoning Best Paper, SURGeLLM Workshop at ACL 2026

Text-to-SQL table reasoning fails when question terms don't match table headers, generating queries over columns that don't exist. TabBridge bridges the table's structure and the question's context through TabSpec, a verified intermediate representation, reaching 73.94% on WikiTableQuestions (+5.3pp over prior SOTA).

VisDoT — VLM Chart Reasoning EACL 2026 Findings

LVLMs fail on chart questions with visual referents like "leftmost bar" or "dark red" because they cannot ground visual elements to their meaning — and chain-of-thought doesn't help when the model looks at the wrong place. VisDoT forces perception before logic, decomposing questions into perception-first sub-questions grounded in graphical perception theory, outperforming GPT-4o with 20× less training data than prior methods.

Jeongwoo Lee, Eunsoo Lee, Jihie Kim. “TabBridge: Bridging Structure and Context for Accurate Table Reasoning.” SURGeLLM Workshop at ACL 2026 — Best Paper Award.
Eunsoo Lee, Jeongwoo Lee, Minki Hong, Jangho Choi, Jihie Kim. “VisDoT: Enhancing Visual Reasoning through Human-Like Interpretation Grounding and Decomposition of Thought.” EACL Findings, 2026.
Jeongwoo Lee, Yongbin Kim, Jihie Kim. “Comparative Insights into Open-source and Closed-source LLMs for Complex Table Question Answering.” ISIS, 2025.
Jeongwoo Lee, Eunsoo Lee, Jihie Kim. “Table Interpretation using Least-to-Most Prompting for LLMs.” KICS, 2025.
Jeongwoo Lee, Minki Hong, Eunsoo Lee, Jihie Kim. “LLM-based Table Data Inference using Data Augmentation and Few-Shot Prompting.” HCLT, 2024.

Projects

E-MOJI — Personalized AI Emoji Web Service Dec. 2022 – Feb. 2023

Leader & Full-Stack Developer · Seoul, KR · GitHub

Led full-stack development (onboarding non-developer teammates onto core Django APIs) and deployed on AWS EC2 with Prometheus-Grafana monitoring; offloaded slow AI image generation to Celery workers via RabbitMQ, using Redis for status polling.

ReBike — Smart Waste Detection Website Jul. 2022 – Aug. 2022

Front-end Developer · Seoul, KR · GitHub

Built the React client (migrated JavaScript to TypeScript mid-project) and annotated a custom waste-image dataset to train a YOLOv5 model, building the detection UI that renders returned bounding boxes and class labels.

Leadership & Teaching

Dongguk University — Undergraduate Teaching Assistant Mar. 2025 – Dec. 2025

Machine Learning & Deep Learning Courses · Seoul, KR

Mentored 30+ students through office hours, clarifying core ML/DL concepts and providing feedback on project deliverables; awarded TA Scholarship.

English Study Club — Founder & Leader Apr. 2025 – Present

Seoul, KR

Founded and led a 6-member peer study group, designing the curriculum and leading discussions across 60+ weekly sessions over 15 months.

Awards & Honors

Best Paper Award — SURGeLLM Workshop at ACL 2026 (TabBridge)
TA Scholarship — Dongguk University, 2025
Academic Excellence Award — Dongguk University (2 semesters) & Tech University of Korea (3 semesters)
Top 10 Finalist, Techeer — 2022
Gold Award — TU Startup Competition, Silicon Valley Summer Bootcamp, Embedded Systems Competition for Freshman