Hyunjun Kim
AI Researcher
Hyunjun Kim
// Specializing in LLM Research & Retrieval-Augmented Generation
About Me
// An undergraduate researcher's journey to become a competent AI researcher in both industry and academia.
I am an undergraduate researcher in the School of Computing at KAIST. Currently, I am actively conducting research in LLM Security andRetrieval-Augmented Generation (RAG). Despite being an undergraduate, I have already authored five AI research papers.
Fueled by a burning passion for AI research and a relentless spirit of inquiry, I aim to become a distinguished AI researcher recognized in both industry and academia. To achieve this, I plan to pursue a PhD to delve deeper into my research.
Transcending the limitations of an undergraduate, I am involved in research activities at various domestic and international institutions, including KAIST, the University of Adelaide, and AIM Intelligence. I have presented papers at top-tier international conferences such as ACL and ICLR. With youthful vigor and boundless potential, I aspire to open new horizons in AI research.
Research Vision & Aspirations
future_goal = "Building on my experience of authoring 5 papers as an undergraduate, I aim to grow into a world-class AI researcher recognized in both industry and academia through a PhD program, contributing to humanity with innovative AI technology."
Undergraduate Researcher
Authored 5 AI research papers while at KAIST School of Computing
PhD Aspirant
Aiming for a PhD to delve deeper into AI research
International Collaborator
Research experience at KAIST, University of Adelaide, and AIM Intelligence
Industry & Academia Goal
Aspiring to be a competent AI researcher recognized in both fields
Undergraduate Achievements
// A young AI researcher's challenging journey
Education
Korea Advanced Institute of Science and Technology (KAIST)
Bachelor of Science, Computer Science
• All courses conducted in English
• Planning Ph.D. in AI Research and Retrieval-Augmented Generation (RAG)
Work Experience
MLAI@KAIST AI Graduate School
Research Intern, LLM jailbreak red team
- •Developing retrieval-augmented generation (RAG) for AI-driven drug discovery
- •Integrating multi-document summarization, molecular pathway analysis
- •Multi-LLM collaboration ("Chain of Agents") and long-context summarization
- •Accelerating research through advanced AI methodologies
AIM Intelligence
Research Intern, LLM jailbreak red team
- •Investigated advanced LLM jailbreak techniques (achieved up to 95.9% success)
- •Co-first author of "One-Shot is Enough..." (ACL 2025 main)
- •Introduced M2S (Multi-turn-to-Single-turn) adversarial methods
- •Co-authored "Breaking the Guardrails with Personality..." (ACL 2025 under review)
- •Exploring personality-driven jailbreak techniques
CREST - The University of Adelaide
Research Intern
- •Led a 36-participant study on LLM-based software engineering tasks with multi-class dependencies and iterative refinement
- •Analyzed GPT logs, test outcomes, and screen recordings to create data-driven guidelines
- •Boosted developer productivity by ~20% through empirical analysis
- •Received strong recommendation letter acknowledging first-author-level contributions
Research Publications
// Contributing to cutting-edge AI research across multiple domains
(*: equal contribution, †: Corresponding author)
Published Research
One-Shot is Enough: Consolidating Multi-Turn Attacks into Efficient Single-Turn Prompts for LLMs
Junwoo Ha*, Hyunjun Kim*, Sangyoon Yu, Haon Park, Ashkan Yousefpour, Yuna Park, Suhyun Kim†
ACL main, 2025
95.9% jailbreak success rate, 80%+ time reduction
Introduced M2S method for efficient LLM jailbreaking (95.9% success). Reduced LLM security testing time by 80%+ by consolidating multi-turn attacks.
View PaperOptimizing Retrieval Strategies for Financial Question Answering Documents in Retrieval-Augmented Generation Systems
Sejong Kim*, Hyunseo Song*, Hyunwoo Seo*, Hyunjun Kim*†
ICLR Advances in Financial AI Workshop, 2025
86% accuracy improvement across 7 datasets
Achieved 86% improvement in retrieval accuracy across 7 real-world finance QA datasets. Reduced LLM hallucinations and accelerated compliance-critical insights at scale.
View PaperHumanity's Last Exam
Long Phan et al. (including Hyunjun Kim)
arXiv:2501.14249, 2025
Multi-institutional collaborative effort
Contributed to large-scale collaborative evaluation of AI capabilities.
View PaperUnder Review
Breaking the Guardrails with Personality: Character-Driven LLMs Influencing Jailbreak Success
Yuna Park, Yujin Kim, Sangyoon Yu, Jiyoung Park, Garam Kim, Won Woo Ro, Junwoo Ha, Hyunjun Kim, Suhyun Kim
(Under review) ACL, 2025
5-institution collaboration on novel jailbreak methods
Explores personality-driven jailbreak techniques in collaboration with KIST, KAIST, SNU, Yonsei, and University of Seoul.
Empirical Analysis on Effective Prompting Strategies on Coding Tasks: A Controlled Experiment
Sangwon Hyun, Hyunjun Kim, Jinhyuk Jang, Hyojin Choi, and M. Ali Babar
(Under review) ICSE, 2026
36-participant controlled study with ~20% productivity boost
Investigating how different prompting strategies optimize LLM performance in real-world software engineering.
Research Timeline
// Journey through groundbreaking AI research and innovations
KAIST Computer Science - Bachelor's Degree
Entrepreneurship & Innovation Course Excellence
Dean's List Achievement
Teaching Assistant - CS101 Introduction to Programming
CREST - University of Adelaide - Research Intern
AIM Intelligence - Research Intern (LLM Jailbreak Red Team)
Multiple Excellence Awards
Excellence Award - Youth SW Mentorship Essay Contest
Gold Award - 4th UNIST-KAIST-POSTECH Data Science Competition
ACL 2025 Main - One-Shot is Enough Publication
ICLR Workshop Publication - Financial RAG Optimization
CoE Leadership Award - KAIST College of Engineering
MLAI@KAIST AI Graduate School - Research Intern
PhD Research Goals
Timeline Navigation
// Drag the slider or click anywhere on the track to explore timeline
Awards & Honors
Gold Award (out of 200+ teams)
4th UNIST-KAIST-POSTECH Data Science Competition
Led development of an advanced retrieval-augmented generation (RAG) system for financial data analytics and multi-agent collaboration. Expanded into a research paper published at ICLR Advances in Financial AI Workshop, 2025.
View DetailsCoE Leadership
KAIST College of Engineering
Recognized for outstanding achievements beyond academics and research.
Excellence Award (out of over 200 teams)
Generative AI Application Contest
Honored for innovative application of generative AI technologies.
Excellence Award
Korea Tourism Data Lab Best Practices
Awarded for "Transforming Daejeon via Tourism Data Analysis," validated by a 288-student survey.
Excellence Award
Youth SW Mentorship Essay Contest
Mentored middle/high school students in coding (Fall 2024) and authored a recognized essay on their growth.
Dean's List
KAIST
Achieved top 2% ranking among freshmen.
Additional Experience
Special Issues: Entrepreneurship & Innovation <Core Skills for Entrepreneurs>
Grade: A0 | Instructor: Prof. Jongchul Kim (Founding Member, McKinsey Seoul)
Analyzed early-adopter segments for "Answerock" using McKinsey's seven-step framework. Delivered a top-graded report recommending high-impact feature enhancements and a phased rollout strategy.
View PPTLet's Collaborate
// Open for research opportunities, AI research discussions, and innovative projects
Research Collaboration
I'm passionate about advancing AI Research and LLM Security. Whether you're interested in collaborative research, discussing innovative approaches to AI development, or exploring RAG systems, I'd love to connect.
Available for
Research discussions, conference meetings, PhD opportunities
Current Focus
AI Research frameworks, LLM security and defense mechanisms
Response Time
Usually within 24-48 hours for research inquiries
if (interested_in_AI_research) {
reach_out()
}