Heungsub Lee

[email protected]
Web Sites
subl.ee, github.com/sublee, linkedin.com/in/sublee


Software architectures, distributed systems, cost optimization, developer experience, and the open source culture.


Programming Languages
Go, Python, TypeScript, Bash
Service Development
Linux, AWS, K8s, Pulumi, React, ZeroMQ, NoSQL, Concurrent programming, Testing
ML Engineering
PyTorch, NVIDIA Nsight Systems, NCCL

Work Experience

Software Engineer
Global AI Platform, Sep 2023 – Present
Focusing on making everyone utilize AI for their daily productivity. Developed and maintained the client and server of Gistty, a Chrome extension summarizing Amazon reviews.
Software Engineering Manager
NAVER, Aug 2020 – Jul 2023

Supervised MLOps platforms by leading 25 software engineers to optimize the inference performance and productivity of HyperCLOVA, an LLM specializing in Korean culture.

Developed NSMLv2, a large-scale ML research platform in CLOVA. Designed its multitenant architecture based on economics to share GPU clusters for HPC among diverse organizations with complex desires while utilizing GPU resources efficiently.

Software Engineer
Kakao Brain, Dec 2018 – Aug 2020

Developed and published a pipeline parallelism library named torchgpipe in open source.

Developed a serverless training framework and a distributed hyperparameter search platform for an AutoML service.

Game Server Engineer
NEXON, Mar 2011 – Dec 2018

Developed cloud-based distributed MMORPG servers for Durango using pub/sub communication over the spatial grid system. Achieved up to 70k concurrent users per game world.

Developed online racing game servers and matchmaking for KartRider Dash and KartRider Coin Rush.

Back-end Web Developer
nPine, Dec 2008 – Feb 2011
Developed web services selling stock images.

Open Source Experience

torchgpipe, Feb 2019 – Apr 2020
Implemented GPipe, a multi-GPU pipeline parallelism technique for training giant models, as a PyTorch library with optimization for CUDA, the autograd engine, and long skip connections. This project has become a part of PyTorch as Pipe APIs
Hangulize, Oct 2010 – Present
Invented a Hangul transcription algorithm and served as a web tool at zero cost. Many professional Korean translators use this tool to translate undocumented proper nouns.
TrueSkill, Jan 2012 – Dec 2015
Implemented TrueSkill™, the rating algorithm for Xbox Live, as a Python library. This project was introduced in PyData Berlin 2019.
  • For PyTorch, fixed potential GPU memory violation (#27371); deprecated inconsistent API (#21006, #25985); discussed a counterintuitive behavior of the autograd engine (#18568).
  • For ZeroMQ, discussed a PUB socket crash (#2942).
  • For Flask, fixed a bug to generate a URL with a subdomain (#108).
  • For jQuery 1.4.3, fixed a bug on content negotiation in Ajax requests.


*Contributed equally

Public Speeches



Computer Software, Kwangwoon University, 2008, Completed the first year only.