Korea Advanced Institute of Science and Technology (KAIST), South Korea
Ph.D. degree, Computer Science · Sep 2026 – Present
Research topic: Efficient LLM Agent Serving Systems
Advised by Youngjin Kwon

Korea Advanced Institute of Science and Technology (KAIST), South Korea
Master's degree, Computer Science · Sep 2024 – Aug 2026
GPA: 4.1 / 4.3
Master's Thesis: Multi-Agent Disaggregation for Serving Mixture-of-Experts LLMs
Advised by Youngjin Kwon

Korea Advanced Institute of Science and Technology (KAIST), South Korea
B.S., School of Computing
Minor in Electrical Engineering
Graduated as a Top 10 Leadership & Volunteer Honoree

Chungbuk Science High School, South Korea
Early Graduation

AGSPEC: Pushing the Limits of Retrieval-Based Speculative Decoding in Coding Agent Pipelines
Sumin Lee, Sukmin Cho, Seungjae Lim, Youngjin Kwon
arXiv Preprint

TTServing: Efficiently Serving Large Language Models with Test-time Search for Inference-time Scaling
Sukmin Cho, Seungjae Lim, Sangjin Choi, Hyunsu Ye, Wooseok Gwak, Youngjin Kwon
Under Review

COSPEC: Efficient LLM Serving with Colocated Speculative Decoding
Sangjin Choi, Hyunjae Lee, Seungjae Lim, Sukmin Cho, Youngjin Kwon
Under Review

BEAM: Joint Resource–Power Optimization for Energy-Efficient LLM Inference under SLO Constraints
Hyunjae Lee, Sangjin Choi, Seungjae Lim, Youngjin Kwon
MLSys 2026 · Ninth Annual Conference on Machine Learning and Systems
pdf · abstract

RAG-based Unanswerable Question Detection in Clinical Text-to-SQL
Donghee Han, Seungjae Lim, Mun Yong Yi
CIKM 2025 · The 34th ACM International Conference on Information and Knowledge Management
pdf · abstract

Leveraging LLM-Generated Schema Descriptions for Unanswerable Question Detection in Clinical Data
Donghee Han*, Seungjae Lim*, Daeyoung Roh, Sangryul Kim, Sehyun Kim, Mun Yong Yi
*Equal contribution
COLING 2025 · The 31st International Conference on Computational Linguistics
pdf · abstract

CASYS KAIST, South Korea
ML Systems Research Intern · Jun 2023 – Aug 2024
Worked on efficient LLM inference and serving systems.

NAVER CLOVA, South Korea
ML Engineering Intern · Aug 2022 – Feb 2023
Built delivery-time prediction models and an AutoML forecasting pipeline for time-series data.

Top Excellence Award (1st Place) 🥇 · Ewha × KAIST C-TECH Sprint, 2026
Developed an XAI-based solar power forecasting system for the renewable-energy generation forecasting incentive program.

Leadership & Volunteer Excellence Award 🥇 · 2025
KAIST President's Award · Top 10 among all 2024–2025 graduates for exceptional leadership and volunteer work.

Excellence Award (2nd Place) 🥈 · Elice AI Hellothon, 2024
Caregiver cognitive-activity lesson guide creator & elderly interactive AI drawing diary service · SAEM·SAM

Grand Prize (1st Place) 🥇 · SPARCS Science Hackathon, 2024
Daejeon Mayor's Award · Web-based XR labs for all types of experiments · ALLeX

Grand Prize (1st Place) 🥇 · Social Problem-Solving Volunteer Showcase, 2024
Gyeonggi-do Governor's Award · Generative volunteer reward system · GreenThread

Gold Prize (1st Place) 🥇 · ICISTS Hackafair, 2024
GenEraser · Customizable profanity-filtering B2B service with RAG

Best Prize (1st Place) 🥇 · Social Problem-Solving Volunteer Idea Hackathon, 2024
Generative volunteer reward system for creating customizable characters · GreenThread

Best Prize (3rd Place) 🥉 · K-Data Science Hackathon, 2024
Fine-tuned & served distortion-free LLM with RAG on Korean history · K-Hyunmoo GPT

Top 10 (Finalist) 🏅 · Elice Generative AI Edu Hackathon, 2023
AI Oneul · Literacy-promotion app using generative AI with few-shot CoT prompting

Silver Prize (2nd Place) 🥈 · KAIST-POSTECH-UNIST Data Science Contest, 2023
ECMiner Award · Business-owner prediction & pop-up / survey ad planning with Hana 1Q usage data

Best Prize (1st Place) 🥇 · Social Problem-Solving Volunteer Idea Hackathon, 2023
Real-estate listing & brokerage recommendation app for single female-headed households · Homey

Grand Prize (1st Place) 🥇 · Korean Donation for Education Idea Contest, 2023
Minister of Science and ICT Award · Dynamic coding-based fairy-tale platform for English education