Jaehyun Jun

I build AI agents at AWS Professional Services. What holds my attention is self-improvement — loops where a system evaluates its own output and refines it — and the loop engineering that makes those cycles converge rather than drift.

Where I want to take this is physical AI. A self-improving loop is comparatively easy when feedback is cheap and text-shaped. It gets hard when the signal comes from the physical world, where observations are noisy, partial, and costly to collect. That is where loop engineering stops being plumbing and becomes the problem worth solving.

My multimodal modeling background is what makes that direction tractable. An agent acting in the physical world has to ground its decisions in perception, not only in language:

• Vision-language models — developed VLMs end to end: training pipelines, architecture selection, and evaluation benchmarks.
• Visual grounding — research in visual question answering (VQA), alongside object detection and segmentation.
• Language modeling — personalized dialogue, instruction tuning, and shipping LLM features to production.

Work Experience

Amazon Web Services

2026 – Present AI/ML Engineer, AWS Professional Services Building AI agents for customers, focused on self-improvement and loop engineering.

SK Telecom

2019 – 2026 (joined from SK T-Brain) AI Engineer, Model Alignment team Developed multimodal large language models and personalized dialogue systems. See Projects for details.

Samsung Electronics Co., Ltd.

2013 – 2015 Software Engineer, Ultrasound Development Group Developed a measurement software platform for ultrasound diagnostic instruments.

Samsung Software Membership

2012 Software Engineer Developed creative software applications.

Education

Seoul National University

2017 – 2019 — M.S., Interdisciplinary Program in Neuroscience Bio-Intelligence Laboratory, advised by Prof. Byoung-Tak Zhang

Hanyang University

2006 – 2013 — B.S., Biomedical Engineering