About
I'm a first-year Master's student at KAIST's School of Computing, working in the Collaborative Social Technologies Lab under Prof. Joseph Seering. I hold a B.S. in Electrical Engineering from KAIST (Magna Cum Laude, GPA 3.88/4.3). I work on human-centered NLP, multi-agent systems, and the computational analysis of group interaction.
I co-led TIDES, a longitudinal bilingual dataset for modeling multi-party social dynamics, accepted to COLM 2026 (co-first author). My current work asks whether internal model controls can manufacture useful cognitive diversity within a single base model, and whether that diversity survives extended multi-agent conversation without converging.
Before graduate school, I spent 7 months at Samsung Electronics' Device Solution Division working on LLM-enhanced RAG systems. Earlier, I did computer vision projects: GAN-based image manipulation and audio anomaly detection.
Research Interests
Current
TIDES & Computational Analysis of Group Interaction
Co-led TIDES, a longitudinal bilingual corpus for modeling multi-party social dynamics: 12 teams, 104 meetings, and 76K+ utterances annotated with speaker roles, developmental stages, and interaction-process categories. Accepted to COLM 2026 (co-first author).
Manufacturing Cognitive Diversity via Internal Model Controls
Can internal model controls manufacture useful cognitive diversity within a single base model? I'm building a systematic falsification record with token-matched baselines and effect-size analysis to find out.
Diversity Persistence in Multi-Agent Conversation
Does diversity survive extended multi-agent conversation, or do agents inevitably converge? I study this with multi-turn debate among stance-seeded agents.
Previous
LLM & RAG System Optimization
At Samsung Electronics' Device Solution Division: Prompt-Induced Reasoning based query decomposition improving search quality by 28% and reducing zero-hit queries by 43%, a query–document relevance check reaching 98.1% score stability, a RAG evaluation system applied to 3,000+ real-world evaluations, and Korean-optimized LLM prompt engineering.
Computer Vision
StyleCLIP-based localized image editing in GAN latent space, audio anomaly detection for industrial machinery (DCASE Challenge), and CNN architecture optimization.
Education
2026 — now
Master of Science in Computer Science
Korea Advanced Institute of Science and Technology (KAIST)
- Collaborative Social Technologies Lab
- Advisor: Prof. Joseph Seering
2020 — 2026
Bachelor of Science in Electrical Engineering
Korea Advanced Institute of Science and Technology (KAIST)
- GPA 3.88/4.3 (95.80/100)
- Magna Cum Laude
Skills & Languages
Programming
python · c
Frameworks
pytorch · tensorflow · hugging face transformers
AI·ML
llm · rag · prompt engineering · gan · clip · object detection · segmentation
Languages
Korean (native) · English (fluent · TOEIC 975 · TOEFL 110)
Contact
The fastest way to reach me is email — jeonggyumarkk@kaist.ac.kr. I'm happy to talk about research or potential collaborations.