Geewook Kim
Tech Lead & Applied Research Scientist at 네이버클라우드(NAVER Cloud)
Gyeonggi, South Korea
Summary
🤩
Rockstar🎓
Top SchoolGeewook Kim is a tech lead and applied research scientist at NAVER Cloud with five years of experience building LLM-based multimodal products and research platforms. He drives R&D from prototype to production, leading projects such as HyperCLOVA-X-Vision, Cream, and contributing to high-profile open-source work like the Donut OCR-free document understanding model. Combining hands-on ML engineering with academic rigor, he is pursuing a Ph.D. in AI at KAIST while teaching AI Engineering in Production at the University of Seoul. His background spans Kyoto University (B.Eng., M.Informatics) and research stints at RIKEN, and he has a track record of integrating research advances—like categorical special tokens and evaluation tooling—into real-world codebases. Comfortable bridging research and engineering, he focuses on vision-and-language systems that power productized multimodal solutions. Colleagues describe him as a pragmatic researcher who moves ambitious models into usable, production-grade services.
5 years of coding experience
3 years of employment as a software developer
Master of Informatics, Systems Science, Master of Informatics, Systems Science at Kyoto UniversityDoctor of Philosophy - PhD, Artificial Intelligence, Doctor of Philosophy - PhD, Artificial Intelligence at 한국과학기술원(KAIST)
English, Korean, Japanese
Github Skills (12)
computer-vision10
pytorch10
trainings10
python10
modeling10
model-optimization9
nlp9
jsonp8
json8
json-parser8
test-automation8
parsing8
Programming languages (2)
Github contributions (5)
clovaai/donut
Jul 2022 - Nov 2022
Official Implementation of OCR-free Document Understanding Transformer (Donut) and Synthetic Document Generator (SynthDoG), ECCV 2022
Role in this project:
ML Engineer
ML EngineerContributions:8 releases, 44 commits, 9 PRs in 4 months
Contributions summary:Geewook primarily contributed to the Donut model, focusing on improvements and refinements related to the model's functionality. They added features such as the integration of categorical special tokens and functions to calculate F1 accuracy scores. Additionally, they made version updates and fixed bugs, including typos and adjustments to the `max_iter` parameter.
nlpdonutunderstandingtransformer-architecturemultimodal-deep-learning
naver-ai/cream
May 2023 - Jun 2024
Contributions:1 release, 11 pushes, 1 tag in 1 year
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.