Internship Inside Out 3
Explore Research, Challenges, and Growth
KAIST 전산학부 인터랙티브 컴퓨팅 연구실(ICLab)은 유비쿼터스 컴퓨팅을 통한 개인 맞춤형 건강 및 웰빙 증진을 목표로 하는 실험실로, 모바일, 웨어러블, IoT 기술을 기반으로 한 컨텍스트 인식 컴퓨팅, 머신러닝 등의 다양한 기술 및 응용 서비스에 대한 연구를 수행하고 있습니다.
2025년도 여름방학 동안, ICLab에서는 8개의 인턴십 프로젝트를 공모하여 16명의 인턴학생을 선발하고 8주간의 연구 프로그램을 진행했습니다. 각 프로젝트는 대학원생 또는 교수님의 멘토링 아래 진행되었으며, 연구의 흐름을 점검하고 발전시키기 위해 1주 차(시작), 4주 차(중간), 8주 차(최종) 총 세 차례의 연구 발표회를 가졌습니다.
이번 인턴십을 마무리하며, 4 명의 학생들이 직접 경험한 연구 과정과 느낀 점을 공유드리고자 합니다.
학부생이지만 연구는 하고 싶어
written by. 금강산
안녕하세요. Simulator-Based Behavioral Analysis of Alcohol-Induced Behavioral Changes Using Multi-Modal Data라는 주제로 2025년 여름 ICLab 인턴십에 참여한 금강산입니다.
2026년 이후 미국에서 판매되는 모든 신차에 음주운전 방지 기술(DADSS) 탑재가 의무화되면서, 음주운전 탐지 연구의 필요성은 더욱 커졌습니다. 그러나 실제 차량을 통한 실험은 현실적으로 불가능하기에, 시뮬레이터 기반 연구가 중요한 대안이 되었습니다. 이번 프로젝트는 이러한 문제의식에서 출발했습니다.
그동안 HCI 분야 학습을 하며 주로 실험 설계와 결과 분석에 집중해 왔습니다. 하지만 학부생으로서 처음으로 제대로 된 연구를 직접 진행해보며 연구는 기획, 즉 첫 단추가 중요하다는 사실을 배웠습니다. 수많은 선행 연구를 읽고 비교하며, 우리 연구만의 방향과 새로움을 찾는 과정에서 진정한 연구의 의미를 체감했습니다. 연구의 동기 설정에서부터 실험 설계와 데이터 분석, 그리고 결과 해석까지 전 과정을 직접 고민하고 실행하며, 어렵지만 밀도 있는 배움을 얻을 수 있었습니다.
처음부터 새로운 연구를 시작한 만큼, CARLA 설치와 환경 세팅, 시나리오 설계, 데이터셋 구축까지 모든 과정을 밟아야 했습니다. 그 과정 속에서도 처음 세운 목표와 필요한 변수를 잃지 않고 나아간 경험이 큰 자산이 되었습니다. 인턴십 기간 동안 연구의 확장성과 개선 방안에 대해서도 깊이 생각할 수 있었고, 기회가 된다면 이를 더욱 발전시켜 나가고 싶습니다.
이번 인턴십은 대학에 입학한 이래 어쩌면 가장 치열하면서도 희망찬 시간이었습니다. 정말 오랜만에 꿈을 잃고 방황하던 오랜 무기력에서 벗어나 다시 목표를 향해 몰입할 수 있었고, 연구실에서 보낸 무수한 시간들이 소중한 자양분이 되었습니다. 연구실 선배님들과 교수님의 지도 하에 많이 배웠고, 또 겸손해졌습니다. 많은 가르침을 주신 멘토 박소민 선배님, 장수은 선배님, 그리고 팀원 Merey에게도 감사 인사를 전합니다. 무엇보다 소중한 기회를 주신 교수님, 그리고 저를 따뜻하게 맞아주신 ICLab 구성원분들께 감사하다는 말씀 드리고 싶습니다!
Fitting a Giant Brain in Your Pocket: My Internship with On-Device LLMs
Written by. Himani Paudayal
On-Device LLM Project OverviewAs a fourth-year undergraduate double majoring in Aerospace Engineering and Computer Science, I had the privilege of spending my first KAIST internship immersed in the challenge of making LLMs run efficiently on devices. Working alongside my teammate under the guidance of Professor Uichin Lee, I was encouraged to think critically at every stage, from defining the problem to validating our results. His feedback pushed us to justify our choices, refine our setup, and approach problems with both technical depth and practical awareness.
Our project focused on sLLMs like LLaMA 3B, Qwen 2B, and DeepSeek 1.5B, optimized for resource-constrained environments such as smartphones and edge devices. This involved experimenting with aggressive quantization in 4-bit and 8-bit formats, as well as structured pruning to remove redundant weights without sacrificing too much accuracy. We spent countless hours testing, logging latency, measuring memory usage, and checking accuracy benchmarks. Some days the models ran smoothly; other days, they refused to cooperate entirely. Each challenge deepened my understanding of trade-offs and strengthened my problem-solving skills.
At the start, I was both excited and a bit nervous, convinced that I needed to show clear “progress” every week. It did not take long to realize that research does not follow a straight line. Some weeks bring breakthroughs, others bring only new questions, and both are equally valuable.
With the guidance of our Professor and TAs, I have completed this internship with sharper technical skills, a clearer understanding of the research process, and a lasting appreciation for persistence over perfection.
When Curiosity Meets Code
Written by. Merey Makhmutova
I have always been curious about digital twins, simulations, and autonomous vehicles, and this internship gave me a chance to turn my curiosity into action. During my internship, I worked on a driving simulation-based project focused on detecting and analyzing drunk driving, which I believe is a great stepping stone for my future career.
My work began with extensive literature review on drunk driving detection, driving simulator studies, and task detection algorithms. And after reading numerous papers, I designed a set of behavioral and physiological metrics for drunk driving detection, ensuring they were both academically grounded and practically feasible. Using the CARLA simulator integrated with multiple sensors, I designed and developed algorithms to automatically detect and analyze driving tasks such as lane changes, turns, stops, speed changes, and peripheral device operations.
The project taught me how to combine domain knowledge with practical coding skills for data processing, signal analysis, and sensor fusion. I learned how to think critically about algorithm design, balance precision with robustness, and validate results against both theoretical models and real-world behavior.
One of the most valuable lessons I learned is that research rarely goes exactly as planned - equipment can malfunction, sensors may behave unpredictably, and experiments often require redesign. These challenges taught me to adapt quickly, embrace creative problem solving, and pivot when needed.
The meaningful highlight of my internship was the supportive and inclusive environment in the lab. The professor and lab members were very kind and welcoming, especially to interns and international students. I am deeply thankful to my mentors, who shared their expertise, offered guidance, and gave valuable feedback. This culture of teamwork, mentorship, and open communication made the experience both academically enriching and personally meaningful.
Opening (Norman) Doors to HCI
Written by. Renz Samuel Gutierrez
Education tool which is developed for learning design thinking at summer session.During the Fall 2024 semester, I came across a course taught by Professor Uichin Lee named Introduction to HCI (CS374). Back then, I neither knew about nor thought about the field of HCI. However, week by week, the course slowly made me realize that behind every project and system, the core of design is not just about building something functional, but about creating for people. That single idea shifted my perspective and planted the seed that would eventually lead me to ICLab.
Fast forward to the summer of 2025, and now I am working on a project for the very same course I once took. My teammate Jihoon and I explored how AI could support the design thinking process. It was fascinating to discover that industry UX professionals already use AI in similar ways, and numerous studies have highlighted its potential in education. However, we also recognized its limits. As Donald Norman quotes in The Design of Everyday Things, “Design is really an act of communication, which means having a deep understanding of the person with whom the designer is communicating.” Technology may continue to evolve, but the human understanding and empathy at the heart of HCI will always remain irreplaceable.
This internship taught me that research is more than just building a project; it’s about exploring, questioning, and discovering. Working with Professor Lee was an incredible opportunity, as his guidance continually pushed our work forward and expanded our perspective. I’m also grateful to the lab members for fostering such a welcoming atmosphere where ideas and insights could be freely shared.
I leave this experience not only with stronger technical skills, but also with a deeper curiosity for the questions still unanswered. As the saying goes, “Once you stop learning, you start dying.” There is so much left to discover in the world, and often the most rewarding insights come when you step into unfamiliar territory.
About Us
