Ambassador
Selected as an Upstage AI Ambassador
Joined a national ambassador cohort to prototype AI workflows with Document Parse, Embeddings, and Solar Pro APIs.
Sumin Im
M.S. Student, KAIST EE
Hello, I'm Sumin Im, an M.S. student in KAIST School of Electrical Engineering and a member of the Mobile Intelligence & Interaction Lab, advised by Prof. Sung-Ju Lee. I am particularly interested in LLM efficiency.
During my undergraduate years at Kyung Hee University, I worked on language and multimodal models, Korean agent evaluation, O-RAN simulation, and on-device AI products.
If technology connects people and the world, I want to stand at the center of that connection. I am interested in building AI systems that remain useful under real constraints: limited data, privacy, latency, and human context.
- Efficient Multimodal Intelligence
- AI Systems
- On-device AI
- Agent Evaluation
News
Admission
Starting M.S. research at KAIST EE MIIL
I will begin research at the Mobile Intelligence & Interaction Lab on efficient multimodal intelligence and human-centered AI systems.
Graduation
Graduation! Completed Computer Engineering at Kyung Hee University
I completed the B.S. program in Computer Engineering at Kyung Hee University.
Publication
2025 / Korean Society for Geospatial Information Science
Improving Equity in Public Sports Facility Accessibility Using K-means-based Adaptive Gaussian 2SFCA and Multi-Objective Greedy Optimization
Sumin Im2024 / Korean Institute of Information Scientists and Engineers
O-RAN and ns-O-RAN: A Study on Architecture and Interface Analysis
Sumin Im et al.Patent Filings
2026 / Korean patent application, pending
Real-Time QR Code Threat Detection System and Method Using Context-Based On-Device AI
2026 / Korean patent application, pending
On-Device AI-Based Voice Impersonation Real-Time Detection System and Method
Timeline
Chungnam Science High School
Scientific foundation
Built the curiosity and quantitative base that later shaped AI and systems research.Kyung Hee University
B.S. Computer Engineering
Studied software, AI, networks, and product engineering while maintaining a 4.27 / 4.5 GPA.
Automobile Network Lab
KHU EE, advised by Prof. Haneul Ko
Researched O-RAN and 6G network systems, including ns-O-RAN simulation and AI-based conditional handover.
LinkedSpaces Inc.
Silicon Valley software engineering intern
Built on-device semantic image search, POI ranking, and album intelligence for private photo systems.
Efficient Multimodal Intelligence Lab
KHU EMIL, advised by Prof. Sungha Choi
Researching efficient reasoning for LLMs while writing my undergraduate thesis, leading paper study groups, and preparing top-tier submissions.Mobile Intelligence & Interaction Lab
KAIST EE, advised by Prof. Sung-Ju Lee
Beginning M.S. research on efficient multimodal intelligence and human-centered AI systems.Selected Projects
Mar 2026 - Present / KHU EMIL research
Evidence-Routed Deep Thinking
Training-Free Attention Routing Diagnostics for LLM Reasoning, studying how attention routes toward evidence sources during deep thinking.Python, PyTorch, Attention Diagnostics, LLM Reasoning
Dec 2025 - Present / KAIST MIIL research
Personalized Health-LLM via Test-Time In-Context Learning
A metadata-guided example selection framework for personalized wearable health prediction without model fine-tuning.Python, PyTorch, LLM APIs, MOMENT, Sleep-EDF
Aug 2025 - Dec 2025 / Open Source Contribution Academy
Ko-AgentBench: Korean AI Agent Evaluation Framework
A Korean agentic tool-calling benchmark for evaluating LLM tool use, reasoning, API reliability, and leaderboard reproducibility.Python, LangChain, Naver/Kakao APIs, evaluation pipeline
Mar 2024 - Sep 2024 / Automobile Network Lab
AI-based Conditional Handover Optimization in 6G O-RAN
An AI xApp simulation for conditional handover decisions in O-RAN environments using radio quality indicators.Python, TensorFlow, NS-3, LSTM
Mar 2025 - Jun 2025 / Kyung Hee University, Prof. Eui-Nam Huh
Federated Learning-based Fraud Detection with Caching Strategy
A vertical federated learning system with caching and feature sparsification for lower-cost fraud detection.TensorFlow, gRPC, Docker
Jan 2025 - Present / Product project
BalJach: Self-living Preparation App
A Play Store app that supports rental fraud prevention with deposit verification, room inspection, and self-living preparation flows.Flutter, Firebase, Google CloudAwards & Scholarships
President Award
AI MCP Hackathon Season 2
Ministry of Science and ICT, NIA
Built an AI video-to-music generation service using Model Context Protocol and placed in the top 4%.2nd place out of 47 teams
Grand Prize
UNI-DTHON
National Software University Hackathon
Led a Flutter-based real-estate safety solution for foreign residents and won the grand prize across five universities.1st place among 140 participants
Grand Prize
Wanted x Kyung Hee Prompton
Prompt Engineering Competition
Developed LLM prompting strategies for education and automation workflows.1st place among 50 participants
Grand Prize
KHU'DATA DATATHON
KHUDA and Lovable, Kyung Hee University
Built Re:Zip, a data-driven housing platform with CPTED-based route scoring.1st place out of 11 teams
Excellence Award
KHUTHON
Kyung Hee University AI Hackathon
Developed Pruning, an AI-driven agricultural optimization platform with ML inference and a FastAPI backend.5th place out of 48 teams
Finalist
AI_TOP_100
Kakao Impact Foundation and Brain Impact
Selected as a national finalist in an AI problem-solving competition.Top 100 out of 3,000 participants
Finalist
AWS x Codetree Programming Contest
Algorithm Competition
Recognized for algorithmic problem-solving performance.Top 100 nationwide
Funding
Kyung Hee University Valley Start-Up
LINC 3.0 Student Entrepreneurship
Received funding and recognition for a rental fraud prevention platform.KRW 2.5M project funding
Around the World
13 cities · 8 countries · 4 continentsHover a destination to trace the flight from Seoul.
Notion Blog