CV

Software Engineering Intern, ML Brain SW Development Team

Samsung Electronics

• Research on Retrieval Augmented Generation (RAG) for internal chatbot system
• Fine-tuned retriever and re-ranker using contrastive learning
• Improved internal document retrieval performance from 20% to 71% Hit@1 for user queries

Undergraduate Research Intern

KAIST, Data Science and Artificial Intelligence Lab

• Advisor: Prof. Chanyoung Park
• Conducted literature review and implementation of representative papers in recommender systems
• Explored LLM-based recommendation approaches (LLM4Rec, TALLRec)

Research Intern

LG AI Research, EXAONE Lab

• Advisor: Dr. Hyeongu Yun
• Built a multi-modal (vision + text) data extraction pipeline to curate large-scale pretraining data for the EXAONE foundation model
• Combined vision-based layout detection models with rule-based parsing to extract structured text, tables, and images from PDF documents
• Produced 73GB+ of pretraining-grade data, validating scalability toward web-scale corpora

Software Engineering Intern

Kounosoft

• Advisor: Dr. Woongmyung Kim
• Constructed Arduino-related Q&A dataset for education platform
• Performed supervised fine-tuning of KoGPT2 model on the custom dataset
• Developed complete chat interface and system using Vue3.js, FastAPI

M.S.

Seoul National University

Data Science

• GPA: 4.08/4.3
• Advisor: Prof. Yohan Jo
• Research interests: Multi-modal Foundation Models, Voice Assistants, Tool-augmented Agents, Post-training & Reinforcement Learning

B.S.

Korea University

Industrial Management Engineering

• GPA: 4.08/4.5 (Major: 4.14/4.5)
• Dean’s List in College of Engineering (GPA: 4.5/4.5; 2023 Fall)
• Graduated with Great Honors (with one semester early graduation)

Don't Adapt Small Language Models for Tools; Adapt Tool Schemas to the Models

ACL 2026 Main (acceptance rate: 19%)

Proposed PA-Tool, a training-free method that adapts tool schemas to align with models’ pretrained knowledge, improving tool-use performance by up to 17% and reducing schema misalignment errors by 80%.

SpeakerSleuth: Can LALMs Judge Speaker Consistency across Multi-turn Dialogues?

ACL 2026 Main (acceptance rate: 19%)

Introduced a benchmark evaluating whether Large Audio-Language Models can reliably judge speaker consistency across multi-turn conversations, revealing significant biases in prioritizing text over acoustics.

SpokenUS: A Spoken User Simulator for Task-Oriented Dialogue

Under review; Machine Learning for Audio Workshop @ ICML 2026

Developed a spoken user simulator that jointly generates text and speech tokens, modeling realistic spoken behaviors (cross-turn slots, barge-in, disfluency, emotion-aware speech) for task-oriented dialogue systems.

SimuHome: A Temporal- and Environment-Aware Benchmark for Smart Home LLM Agents

ICLR 2026 Oral (acceptance rate: 1.13%)

Developed a time-accelerated smart home simulation environment with 600 benchmark episodes, revealing that even top models struggle with temporal scheduling and state verification.

Quantifying Data Contamination in Psychometric Evaluations of LLMs

EACL 2026 Findings (acceptance rate: 36.2%)

Proposed a framework to systematically measure data contamination in psychometric evaluations of LLMs, providing evidence of strong contamination in popular inventories.

Tool-Augmented Agents: Evolution from Autonomy to Interaction

Korean Institute of Information Scientists and Engineers, Vol. 43, No. 11, pp. 14-25

Comprehensive survey examining the evolution of tool-augmented agents, focusing on the shift from autonomous capabilities to interactive paradigms in human-centered interaction.