Summary
Kevin Ko is an AI research engineer with nine years of experience building and scaling large language models, currently focusing on LLM post-training and agentic capabilities at Upstage. He has led pre-training and post-training efforts across Kakao, Kakao Brain, and EleutherAI, driving projects like Polyglot-Ko and Kanana that achieved performance comparable to much larger models. A prolific open-source maintainer, Kevin created widely used tools including Transformer (4k+ stars), Pororo, KoChat, and Parallelformers—work that spans model implementation, deployment, and Korean-language NLP utilities. He combines deep systems expertise (Megatron, DeepSpeed, TP/PP optimizations, OSLO) with hands-on engineering from serving optimized models in production (ONNX/TensorRT, Triton). Beyond code, he publishes papers and technical write-ups and organizes community learning through talks and manager roles in Korean NLP groups. His background in software engineering and consistent open-source impact make him a bridge between research, tooling, and practical LLM deployment.
9 years of coding experience
5 years of employment as a software developer
학사, Software Engineering, 학사, Software Engineering at Jeonbuk National University
Korean, English