hi i'm kevin!
i'm currently on leave from studying math at Princeton to cook up something new!
i like to spend time thinking about how AI models think, how humans think, the future of education, and the financial markets
previously, i worked on:
- quantitatively researching markets @ Mirae Asset Securities and Oraclum Capital
- ML forecasting and RAG systems @ Hyundai Engineering & Construction
- VLM biases and reasoning model post-training research @ Princeton
- RL for scalable ESL education @ Heu Learning
- fixed income models + data pipelines @ Migrations.ml
- LLM model compression research @ Seoul National University
reach out at kevinjeon@princeton.edu or @kejerial!
technical projects & papers
- "Auditing Aha! Moments in Finetuned Language Models: A Cognitive-Theory Driven Analysis of CoT": Developed a psychology-driven audit of LLM reasoning traces showing how different post-training methods affect reasoning strategies beyond accuracy.
- "Towards Foundation Models for 3D Vision: How Close Are We?" (3DV 2025): Introduced UniQA-3D benchmark to evaluate whether foundation models understand 3D visual–spatial structure (tl;dr they don't).