Shun Kiyono
Machine Learning Engineer at Cohere
Sendai, Japan
Summary
👤
Senior🎓
Top SchoolShun Kiyono is a Machine Learning Engineer with 11 years of experience specializing in NLP and a Ph.D. in Information Sciences from Tohoku University. His research on semi-supervised methods for grammatical error correction and machine translation evolved into hands-on development of Large Language Models across pre-training and post-training stages (SFT, DPO, etc.). He has combined academic rigor with production-focused engineering at RIKEN, LINE, SB Intuitions, and now Cohere, shipping data pipelines and model training workflows. An active contributor to speech and translation tooling, he has practical experience preparing subword tokenization and BPE pipelines for the widely used ESPnet toolkit. Based in Sendai, Japan, he is passionate about building impactful AI systems that bridge research advances and deployable solutions. His profile and publications reveal a pattern of turning novel learning techniques into reproducible, scalable training pipelines.
11 years of coding experience
3 years of employment as a software developer
博士（情報科学）, 情報科学研究科, 博士（情報科学）, 情報科学研究科 at 東北大学
Japanese, Chinese
Github Skills (9)
data-preprocessing10
machine-translation10
dataprep10
python10
byte-pair-encoding10
bash9
nlp9
pytorch3
text-to-speech3
Programming languages (4)
Github contributions (5)
espnet/espnet
Jan 2020 - Feb 2020
End-to-End Speech Processing Toolkit
Role in this project:
ML Engineer
ML EngineerContributions:16 commits, 7 PRs, 10 comments in 1 month
Contributions summary:Shun primarily focused on modifying and creating scripts related to data preparation, tokenization, and BPE (Byte Pair Encoding) training for machine translation tasks. Their commits show the integration of subword-nmt, indicating a focus on improving the model's ability to handle out-of-vocabulary words. These changes were implemented to prepare data for the ESPnet toolkit, specifically for IWSLT16 dataset, suggesting a hands-on role in preparing training data for machine translation models. The commits also showcase the modification of the run script, indicating the user's involvement in the training and evaluation pipeline.
speech-recognitionspeech-separationchainerspoken-language-understandingspeech-processing
butsugiri/gec-pseudodata
Sep 2019 - Dec 2019
Contributions:20 commits, 8 pushes in 3 months
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.