Chanwoong Yoon

        



Ph.D. student
Manning College of Information & Computer Sciences
University of Massachusetts Amherst
Center for Intelligent Information Retrieval (CIIR) Lab

I'm a first-year Ph.D. student at UMass Amherst, advised by Negin Rahimi and Andrew McCallum. My research focuses on self-evolving agentic systems and long-horizon search.

profile photo

News


Selected Publications (full list on Google Scholar)

agentmercury

AgentMercury: Your Agent Can Synthesize Verifiable Environments for Business Scenarios at Scale
Minbyul Jeong, Chanwoong Yoon
arXiv 2026.

Paper / Project Page / Hugging Face

We introduce AgentMercury, a framework that turns high-level business scenarios into executable, verifiable environments with persistent state, services, tools, and cross-service invariants. We build 4,783 environments across 14 industries and use them as training worlds for agentic RL.

solaropen2

Solar Open 2 Technical Report
Upstage Solar Team
Technical Report 2026.

Paper

Solar Open 2 is a 250B-A15B MoE language model built for long-horizon agentic tasks. I built the tool-use task synthesis pipeline behind its agentic post-training.

safetycost

Who Pays More for Safety? Measuring the Disparate Cost of Safety Alignment across Languages
Chanwoong Yoon, Jungsoo Park, Alan Ritter
EMNLP 2026 Main.

Paper / Code

We introduce Safety Cost, the loss in response utility that safety alignment imposes on benign requests, and show that non-English users consistently pay more of it than English users, often through a quiet loss of informational depth that refusal rates miss.

retpo

Ask Optimal Questions: Aligning Large Language Models with Retriever's Preference in Conversational Search
Chanwoong Yoon*, Gangwoo Kim*, Byeongguk Jeon, Sungdong Kim, Yohan Jo, Jaewoo Kang
NAACL 2025 Findings.

Paper / Code

We present Retriever’s Preference Optimization (RetPO), which optimizes a language model (LM) for reformulating search queries in line with the preferences of the target retrieval systems.

compact

CompAct: Compressing Retrieved Documents Actively for Question Answering
Chanwoong Yoon, Taewhoo Lee, Hyeon Hwang, Minbyul Jeong, Jaewoo Kang
EMNLP 2024 Main.

Paper / Code

We propose a novel framework that employs an active strategy for compressing extensive documents. CompAct dynamically preserves query-related contexts, focusing on the integration of information across documents.


Awards and Honors

Academic Service

Template based on Jon Barron's website.