Summary
Seok Song is a PhD candidate and graduate research assistant at Iowa State University with nine years of experience building and evaluating trustworthy GenAI systems, focusing on LLM and VLM reliability. He designs controlled benchmarks and mitigation methods that reveal when and why reasoning breaks down—quantifying real-world impacts like 15% accuracy drops from irrelevant noise and 17–65% losses from format changes—and has recovered substantial performance with targeted defenses. His multimodal work created a 3K+ chart dataset, a novel visual-bias metric, and fine-tuning strategies to improve chart understanding and faithfulness under conflict. Seok pairs this research with product-facing experience: he built an LLM-powered, grounded analytics assistant and an NL-to-SQL pipeline for billion-row BigQuery datasets at The Home Depot. He also has a background founding and shipping consumer apps and teaching large CS courses, showing an unusual blend of hands-on engineering, entrepreneurship, and pedagogy. Based in Ames, Iowa, he combines rigorous evaluation metrics with pragmatic grounding techniques to make GenAI more robust in messy, real-world settings.
9 years of coding experience
3 years of employment as a software developer
Doctor of Philosophy - PhD, Computer Science, Doctor of Philosophy - PhD, Computer Science at Iowa State University
Korean, English