Dongseong Hwang
Staff Research Scientist at Apple
Seattle, Washington, United States
Summary
🤩
Rockstar🎓
Top SchoolDongseong Hwang is a Staff Research Scientist in Seattle with 13 years of engineering experience bridging foundational physics, graphics, and cutting-edge ML research. He helped build production speech and long-context LLM solutions at Google—including TransformerFAM and the FAdam optimizer that reframed Adam through information geometry—and now conducts fundamental research at Apple Intelligence. His background ranges from low-level Chromium and WebKit graphics optimizations that shipped in Chromebooks and WebGL tests to full-stack contributions such as a Chrome drone controller sample, reflecting rare fluency across systems, graphics, and ML. A self-taught programmer who began in VR game development, he combines deep mathematical intuition (MS Physics) with pragmatic engineering to deliver state-of-the-art models and runtime performance wins. He’s driven by a curiosity about intelligence itself, focusing on the mechanisms of memory, reasoning, and planning that current ML lacks.
13 years of coding experience
17 years of employment as a software developer
Master of Science - MS Physics, Master of Science - MS Physics at Seoul National UniversityBachelor of Science - BS Physics, Bachelor of Science - BS Physics at Korea Advanced Institute of Science and Technology
Korean, English
Github Skills (31)
webgl10
machine-translation10
debug10
javascript10
crosswalk10
c-language10
tizen10
testing10
attention-mechanism10
chrome-extension10
pad10
gamepad10
tensorflow10
chrome10
speech-recognition10
Programming languages (9)
Github contributions (5)
tensorflow/lingvo
Jan 2021 - Jan 2023
Lingvo
Role in this project:
ML Engineer
ML EngineerContributions:44 commits in 2 years
Contributions summary:Dongseong primarily contributed to the Lingvo project by implementing and modifying components related to machine learning models, specifically in the area of speech recognition and related tasks. Their work includes refactoring of existing code, adding new features such as frequency warping for data augmentation, and optimizing attention mechanism computations, indicating a focus on improving the performance of the models. They also worked on correcting identified WER (word error rate) reporting and supporting of causal attention, and the introduction of block masking, showcasing a strong understanding of speech models.
asrtranslationctcspeech-recognitiontensorflow
crosswalk-project/crosswalk
Aug 2013 - Feb 2015
A web runtime built on Chrome. This project is currently unmaintained.
Role in this project:Full-stack Developer
Full-stack DeveloperContributions:26 commits, 1 PR, 7 comments in 1 year 6 months
Contributions summary:Dongseong primarily focused on adapting the Crosswalk project for Tizen mobile devices, addressing platform-specific requirements. Their contributions included defining Tizen-specific build configurations, modifying the application's user interface, and adjusting the application's behavior to align with Tizen's framework. The user also worked on integrating the application with Tizen's task switcher by implementing AppcoreContext and addressing issues that arose during the integration process.
chrome-extensionjavascriptruntimechrome
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.