Sean Yi
Machine Learning Engineer at Toss
Seoul, South Korea
Summary
👤
Senior🎓
Top SchoolSean Yi is a machine learning engineer with 9 years of experience, currently focused on productionizing RAG systems for Channel Corporation’s AI agent ALF and recently joining Toss. Trained in NLP during graduate school at Korea University, he has broadened into multimodal and retrieval-focused systems, having built image-text models, fashion NER, and embedding pipelines in industry. He’s driven MLOps improvements—revamping labeling-to-training pipelines, dataset management, and deploying models with Triton and TEI—to shorten QA cycles and scale serving. An active contributor to the Hugging Face Transformers ecosystem, he added robust MLflow integration and a “best” save strategy to Trainer, helping teams manage experiment tracking and model selection. Based in Seoul, he combines research instincts with hands-on engineering across vector search, OpenSearch, and A/B experimentation to turn prototypes into reliable products. Collected bilingual datasets and custom multimodal models reveal a practical focus on cross-lingual product discovery beyond typical NLP work.
8 years of coding experience
3 years of employment as a software developer
Master's degree Computer Science, Master's degree Computer Science at Korea University
English, Korean, Chinese, Spanish
Github Skills (10)
pytorch10
machine-learning10
deeplearning-ai10
mlflow10
deep-learning10
python10
pre-trained-model9
hub9
nlp9
language-model8
Programming languages (10)
Github contributions (5)
huggingface/transformers
Dec 2019 - Mar 2025
🤗 Transformers: State-of-the-art Machine Learning for Pytorch, TensorFlow, and JAX.
Role in this project:
ML Engineer
ML EngineerContributions:12 reviews, 7 PRs, 36 comments in 5 years 3 months
Contributions summary:Sean contributed to the MLflow integration within the Hugging Face Transformers library. They added functionality to set the tracking URI for the MLflow callback, allowing users to specify where to store their MLflow runs. They fixed existing issues with the tracking URI settings, preventing clashes with existing MLflow implementations and ensuring correct behavior when the URI is set. The user also implemented a new `"best"` option for the `save_strategy` in the `Trainer` class, enabling saving of the best model based on evaluation metrics.
pythonbertspeech-recognitionstate-of-the-artflax
seanswyi/paper_breakdowns
Mar 2020 - Jul 2020
Contributions:179 pushes, 1 branch in 4 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.