Summary
Wonho Song is an AI research engineer with eight years of experience specializing in large language model pretraining and fine-tuning, currently on Upstage’s LLM Pretraining Team. He led the pretraining of a 22B-parameter foundation model that achieved an 80 MMLU score and previously pushed model scaling, multilingual continued pretraining, and context-length extension work. A competitive Kaggle Master with multiple top finishes and academic competition wins, he brings a strong empirical mindset from reproducible contest environments into production ML. Notably, he engineered cost-efficient depth up-scaling from 7B to 10.7B and drove a first-place solution on the HuggingFace Open LLM Leaderboard. He combines rigorous academic training from KAIST and Chung-Ang University with hands-on product deployments across healthcare, finance, and e-commerce domains. Beyond model building, he mentors as an object detection instructor at Boostcamp, blending teaching with cutting-edge research practice.
8 years of coding experience
2 years of employment as a software developer
Bachelor's degree, Industrial Security, Summa Cum Laude, Bachelor's degree, Industrial Security, Summa Cum Laude at Chung-Ang University
Master's degree, Graduate School of Information Security, Master's degree, Graduate School of Information Security at Korea Advanced Institute of Science and Technology(KAIST)