I am a Research Scientist/Engineer at VUNO. I obtained my Master’s from the Graduate School of AI at KAIST, under the supervision of Prof. Sung Ju Hwang. If you are interested in collaborating with me on research projects, please contact me! Here is my CV (Curriculum Vitae).

My primary research interests include developing innovative AI solutions that address real-world challenges across diverse domains, with a commitment to making a meaningful impact on society by bridging the gap between AI and people through both research and engineering. My current interests involve:

• Training task-specific AI models: Leveraging the growing capabilities of foundation models in data-efficient settings, utilizing techniques such as knowledge distillation (KD), few-shot learning, self-supervised and semi-supervised learning, and active learning (AL).
• Building automated systems: Developing systems that genuinely automate and assist human workflows by integrating AI models such as large language models (LLMs), multimodal large language models (MLLMs) through instruction tuning, reinforcement learning (RL), and agentic AI frameworks.

Outside of research, I enjoy learning new languages 📚 (currently studying Spanish and Chinese), swimming 🏊 (been doing for a long time), and continuous self-improvement and acceleration of personal growth in expertise and research.

🔥 News

• 2026.01:  🎉 One paper accepted at ICLR 2026!
• 2025.07:  🎉 Renewed personal website with latest information

📝 Selected Publications

• Automated Structured Radiology Report Generation with Rich Clinical Context

  [paper] [project page] [github]
  Seongjae Kang *, Dong Bok Lee *, Juho Jung, Dongseop Kim, Won Hwa Kim, Sunghoon Joo (* equal contribution)
  Under Review

• PCoreSet: Effective Active Learning through Knowledge Distillation from Vision-Language Models

  [paper] [project page] [github]
  Seongjae Kang *, Dong Bok Lee *, Hyungjoon Jang, Dongseop Kim, Sung Ju Hwang (* equal contribution)
  Under Review

• Simple yet Effective Semi-supervised Knowledge Distillation from Vision-Language Models via Dual-Head Optimization

  [paper] [project page] [github]
  Seongjae Kang *, Dong Bok Lee *, Hyungjoon Jang, Sung Ju Hwang (* equal contribution)
  Under Review

• CostNav: A Navigation Benchmark for Real-World Economic-Cost Evaluation of Physical AI Agents

  [paper] [project page] [blog] [github]
  Haebin Seong, Sungmin Kim, Yongjun Cho, Myunchul Joe, …, Seongjae Kang, …, Youngjae Yu, Yunsung Lee
  Under Review

• Frequency-Balanced Retinal Representation Learning with Mutual Information Regularization

  [paper] [github]
  Seunghoon Lee, Seongjae Kang, Inhyuk Park, Gitaek Kwon, Jihyeon Baek, Doohyun Park
  ICLR 2026

📖 Education

• 2022.02 - 2024.02, M.S. in Graduate School of AI, KAIST
• 2016.02 - 2022.02, B.S. in Computer Science (major) and Electrical Engineering (minor), KAIST

💻 Work Experience

• 2024.05 - Present, Research Scientist/Engineer (Military Substitution) at VUNO
• 2024.02 - 2024.03, Research Scientist/Engineer at Deepauto AI
• 2020.09 - 2021.09, Research Scientist/Engineer at VoyagerX
• 2019.03 - 2019.08, Undergraduate Research Assistant at SGVR Lab, KAIST, Daejeon, South Korea. Prof: Sung-eui Yoon
• 2017.06 - 2018.06, Undergraduate Research Assistant at Computer Vision Lab, KAIST, Daejeon, South Korea. Prof: Hyunseung Yang

🚀 Selected Projects

• YouTube Multi-Subtitle for Chinese Learning
  [Demo]
  A Chrome extension that displays three subtitle tracks simultaneously (English, Chinese characters, and Pinyin) to enhance Chinese language learning through YouTube videos. The extension automatically converts Chinese characters to Pinyin in real-time and supports loading subtitles from multiple sources including machine translation, replacing YouTube’s default subtitle with a custom overlay that provides all necessary linguistic information in one place.

• Easy - Expense Report Automation Chrome Extension
  [Chrome Web Store] [PDF] [PPT]
  A Chrome extension that automates the monthly expense report (지출결의) process for company employees. This tool reduces what typically takes 1-2 hours of manual work each month into just a few minutes with minimal effort, significantly improving productivity and reducing repetitive administrative tasks.