Seungyeon Jwa
Seungyeon Jwa

I am a second-year M.S. student at Seoul National University, advised by Prof. Jonghyun Choi. Previously, I spent six months as a visiting graduate student with the Minnesota NLP Group, hosted by Prof. Dongyeop Kang.

My research focuses on the evaluation and alignment of AI systems. I aim to use evaluation to guide the development of AI systems that are not only high-performing but also reliable, responsive to users’ diverse needs, and capable of creating social and economic value in real-world settings.

Before graduate school, I spent three years as an NLP research engineer at NCSOFT (now NC AI), post-training dialogue and judge models. This gave me hands-on experience across the post-training pipeline, from data synthesis to training and evaluation.

Selected Publications
evaluation
Becoming Experienced Judges: Selective Test-Time Learning for Evaluators
EACL 2026 (Short) Β· EvalEval @ ACL 2026 Oral Γ—2
evaluation alignment
OffsetBias: Leveraging Debiased Data for Tuning Evaluators
Junsoo Park*, Seungyeon Jwa*, Meiying Ren, Daeyoung Kim, Sanghyuk Choi
alignment
Mitigating Hate Speech in Korean Open-domain Chatbot using CTRL
Seungyeon Jwa*, Young-rok Cha*, Moonsu Han, Donghoon Shin
News
Nov 2026 I wrapped up my visit at Minnesota NLP Group and returned to SNUMPR. πŸ‡°πŸ‡·
Sep 2026 We released EvoDuet, which co-evolves web search queries and solutions for LLM-driven scientific discovery. Another collaboration at Minnesota NLP Group.
Sep 2026 Our paper on recursive self-improvement, Metan, was accepted to NeurIPS 2026. This is my first work from my visit at Minnesota NLP Group.
more β†’