From 2024 to 2025, I worked at Google Research as a Student Researcher on enhancing the use of LLMs (details confidential).

[My role in this project]
Student Researcher

Working with Gromit Chan at Adobe Research, I developed a system that assists with poster design by simulating real-time collaboration with persona-driven agents.

This work was done during my internship at Adobe Research. (UIST '25 paper)

Keyscreen of PosterMate

[My role in this project]
Leading the project

Advised by Lucy Lu Wang and Gary Hsieh, I built a system that automatically tailors design implications from HCI papers to individual designers' needs with generative AI. (DIS '25 paper)

Pipeline for generating customized design cards using our system

[My role in this project]
Leading the project

Working with Young-Ho Kim at Naver AI, I built an interface that helps users to generate evidence-baesd exercise plans using an LLM chatbot.

This work was done during my internship at Naver AI. (CUI '25 paper)

Keyscreen of our system

[My role in this project]
Leading the project

Advised by Lucy Lu Wang and Gary Hsieh, I built a system that automatically transforms design implications from HCI papers into design cards with generative AI. (CHI '24 paper)

Pipeline of generating design cards using our system

[My role in this project]
Leading the project

Working with Ruican Zhong, Gary Hsieh, Lucas Colusso, and Kaiser Permanente research team, we built an LLM-infused Miro plugin that helps designers to generate causal pathway diagrams (CPDs) to inform their theory-driven designing. (CHI '24 paper)

An example of causal pathway diagram (CPD)

[My role in this project]
Building a system

Working with Taewan Kim and Hwajung Hong, we explored the design space of generative language models (e.g., GPT-3) in supporting a reflective writing process. (CHI '24 paper)

Keyscreen of our system

[My role in this project]
System development, design ideation

Advised by Gary Hsieh, I built a chatbot that helps familiarization of ad hoc teammates. Plus, I revealed the efficacy of such familiarization in ad hoc team performance. (CHI '23 paper)

Overview of the IntroBot's system components

[My role in this project]
Leading the project

In this project, we investigated how older adults use online videos for learning. More specifically, we adopted a mixed-method approach by conducting (1) an analysis of 41.8M logs extracted from K-MOOC, a Korean MOOC (Massive Open Online Course) platform and (2) in-depth interview sessions with older adults.

Led by Seoyoung Kim, this project was advised by Juho Kim. (CHI '23 paper)

Example of the data analysis from our study

[My role in this project]
Running interview sessions, taking part in thematic analysis, analyzing log data, writing

To guide the design of explanation for assessing damages from satellite imageries, we characterized how humans generate explainations on the AI detection and their effects.

Working with Katelyn Morrison, this project was co-advised by Ken Holstein and Adam Perer. This work was accepted at CSCW 2023 and NeurIPS '21 Workshop (AI for Humanitarian Assistance and Disaster Response).

Examples of the explanation in damage detection tasks from satellite imageries

[My role in this project]
User study, open-coding responses, system development

In this project, we investigated methods to assist emotional support processes in online mental health community, particularly by leveraging AI.

To support their community interactions, we developed and deployed (i) multi-class sentiment detection model using BERT to support the affectionate feedback process and (ii) keyword extraction based on the saliency of each keywords to facilitate the experience-sharing processes.

Co-advised by Hwajung Hong and Jinwook Seo, this work was published in CHI 2022 Extended Abstracts (Late-Breaking Work) and CHI 2022 Workshop (Future of Emotion in HCI).

Keyscreens of our system

[My role in this project]
Leading the project, preliminary interview, storyboarding, writing, system development

The course registration process at Seoul National University, often referred to as a "war," is notoriously competitive due to its first-come, first-served nature.

During this process, students occasionally refresh the registration website by mistake. When this happens, the system immediately displays an error message and forces the user to log in again. While this might be a measure to prevent the use of macros (though this remains unconfirmed), it poses a significant problem, since logging in again during such a high-pressure moment wastes precious time.

To address this issue, I developed a browser extension that allows users to bypass the error message with a single click, eliminating the need for a full log-in. This extension is compatible with both Chrome and Whale browsers.

Keyscreen of SNU CRS Refresher

[My role in this project]
Independent project

Have you ever wished a robot to send romantic messages to the other on behalf of yourself? Having a conversation in a romantic relationship is often difficult. Sometimes we ask our friends for help, but such help is limited.

In this project, we assumed that an AI agent in chat settings became so smart that it perfectly analyzes the counterpart's emotion and suggests responses in a chat setting. Along with a prototype app LOVEGORITHM that helps such a romantic relationship, we designed and developed an LOVEGORITHM online user community that illustrates possible feedback and behaviors of users. Check out the link above!

This work was submitted as a team project assignment to the Spring 2021 Creative Research Practice course.

Keyscreens of LOVEGORITHM

Keyscreen of LOVEGORITHM online user community

[My role in this project]
Web development, prototype design, setting up dialogues, ideation

We aimed to develop a system that facilitates users' conversation on political issues and developed a high-fidelity prototype: PoliNewsBot.

This work was submitted as a team project assignment to the Spring 2021 HCI Theory and Practice course.

Keyscreens of PoliNewsBot

[My role in this project]
System development, ideation, setting up dialogues

We explored the role and behavior of entities surrounding translation algorithms, in a situation where users trick and game them.

This work was publised in CSCW 2021.

Entities of reviewing communities where translation algorithm is applied

[My role in this project]
Writing, designing figures

Due to the irreversible characteristics and disastrous effects of Age-related Macular Degeneration (AMD), a precise diagnosis of the symptoms is extremely important. Yet, paper-based Amsler Grid, the most prevalent testing method, is highly limited in that it requires the indirect report of patients and quantitative reporting is difficult.

In this project, I designed and developed a touch-based Amsler grid web app. As such, I aimed to support patients' exact report of their regions of symptoms.

What made this project special to me is that, I made a LaTeX template for writing this paper! I found that my friends preparing for their thesis always felt it difficult to produce a thesis format, since our department only gave us a 'scanned' version of template, not a real document. Thus, I generated a template which can be found here (this template was built upon previous Ph.D. dissertation template under license)

This study was conducted for my B.S. thesis, and later published in HCI Korea 2022.

Keyscreen of AmslerTouch

[My role in this project]
Independent project

Prolonged time of computer use increased the prevalence of ocular problems including eyestrain, tired eyes, irritation, redness, blurred vision, and double vision, collectively referred to as computer vision syndrome. We aimed to explore the interface elements of computer-based intervention for computer vision syndrome to set design guidelines based on pros/cons of each interface element.

In this project, I developed LiquidEye, an intervention application targeted to MacOS. To understand which design elements of intervention app do users prefer, we collected each user's time-series system preferences through the study, which were jointly analyzed with the self-reported satisfaction to identify the significance of each element statistically. (JMIR '21 paper)

Keyscreen of LiquidEye

[My role in this project]
System development, running interview sessions, writing

It is a prevalent behavior of having a chat with strangers in online settings where people can easily gather. Yet, people often find it difficult to initiate and maintain conversation due to the lack of information about strangers.

Hence, we aimed to facilitate conversation between the strangers with the use of machine learning (ML) algorithms and present BlahBlahBot, an ML-infused chatbot that moderates conversation between strangers with personalized topics. Based on social media posts, BlahBlahBot supports the conversation by suggesting topics that are likely to be of mutual interest between users.

A user study with three groups (control, random topic chatbot, and BlahBlahBot; N=18) found the feasibility of BlahBlahBot in increasing both conversation quality and closeness to the partner, along with the factors that led to such increases from the user interview. Overall, our preliminary results imply that an ML-infused conversational agent can be effective for augmenting a dyadic conversation. (CHI '21 Late-Breaking Work)

Framework of BlahBlahBot

[My role in this project]
System development, running interview sessions, writing

According to a recent survey conducted by American Psychological Association [1], academic stress is cited as the number one stressor in students' daily life. This issue is specifically considered serious in South Korea, where half of Korean teenagers experienced a thought of suicidal and academic stress is reported as a top-ranked attribute to such phenomenon [2].

This, however, cannot be solely dealt with reducing academic affair. Since such academic phenomenon is multi-faceted with lots of stakeholders involved (e.g., homeroom teacher, parents, and friends), we should not naïvely understand the issue. For example, high schools often focus on increasing the number of students entering so-called 'prestigious universities.' However, by solely focusing on the name of university rather than a major, this often undermines the agency of students' decision on their future career, and leading students to suffer from subsidious academic stresses as well as a stress on the study itself.

Since the academic curriculum drastically changes over time and policy-makers and relatively older people cannot easily understand such a situation, we aimed to show how difficult it is for high school students to 'survive' under such factors that often lead to academic stresses. By having users choose a choice in each crossroad of choices and differentiating the following storieline, we, though indirectly, let them feel how stressed high school students would be in the midst of a choice.

Keyscreens of Easier said than done

REFERENCE
[1] American Psychological Association. (2014). Stress in America: Are teens adopting adults’ stress habits. Washington, DC: Author.
[2] Kang, Y. (2014). Poll shows half of Korean teenagers have suicidal thoughts. The Wall Street Journal.

[My role in this project]
System development

In this project, we designed an application to support students who had been studying abroad but returned home due to the pandemic. Many of them struggled with irregular sleep patterns caused by time zone differences, which affected both their health and academic performance. To deeply understand their needs, we conducted a series of research activities, including in-depth interviews, observational studies of YouTube vloggers, diary studies, and web-based surveys.

Based on our findings, we developed 夜-planet, an application tailored to students navigating these challenges. The character '夜' means 'night' in Chinese, and the name reflects the app’s focus on supporting nocturnal lifestyles.

If you're interested in exploring our prototype, please check out the link above.

Prototype of 夜-planet

[My role in this project]
User study, ideating design elements

At the end of my sophomore year, I came across a CHI paper by chance: “Blocks4All: Overcoming Accessibility Barriers to Blocks Programming for Children with Visual Impairments” by Milne and colleagues. The paper explored how to make block-based programming tools more accessible for children with visual impairments. I was fascinated by how HCI researchers deeply engaged with specific user groups and translated design implications into tangible systems.

Around that same time, my supervisor asked if I’d be interested in leading a project to develop an AAC (Augmentative and Alternative Communication) application. I immediately said yes; at that point, I was eager to get involved in research, but opportunities were scarce for undergraduates without prior experience. From then on, I spent countless nights developing an iOS app, conducting user studies, reading papers, and designing interfaces. Everything was new to me—from receiving research funding to navigating the IRB process for our study. As luck would have it, we wrapped up the project just in time for the CHI 2020 deadline. We submitted our work, and it was accepted with an Honorable Mention Award.

In our paper, we emphasized that AAC stakeholders shouldn’t be limited to just the child, but must include caregivers as essential partners in communication. With that in mind, we developed Talkingboogie, a mobile AAC system designed to support children with developmental disabilities and their caregivers in collaboratively shaping effective communication strategies.

Usage flow of TalkingBoogie system

TalkingBoogie system consists of two mobile apps: TalkingBoogie-AAC for caregiver-child communication, and TalkingBoogie-Coach supporting caregiver collaboration.

Keyscreens of TalkingBoogie-AAC

Keyscreens of TalkingBoogie-Coach

[My role in this project]
System development, writing, user study, running interview sessions, data analysis

I believe that one's favourite set of music in a specific period represents own emotion at that period. To prove this, I listed all my favourite songs by each semester. Then, I analyzed the emotion of each song using Cyanite, and I visualized the results with D3.js.

Well, I was super shocked by the result. For my depressing time, I used to listen to melancholic songs. For my highest period, my playlist was full of happy and energetic songs. Why don't you try with your own playlist?

This design work was submitted as an assignment to Spring 2020 Design Thinking for Communication course.

[My role in this project]
Independent project

It's always fun to imagine things that will never happen XD

In this project, I assumed that well-known tech companies run their own football team in a 'Tech Champions League.' In this tournament, each team competes using own market cap. For this project, I mainly focused on designing small multiples with the following question: What comes to mind immediately when I think of the company?

This design work was submitted as an assignment to Spring 2020 Design Thinking for Communication course.

[My role in this project]
Independent project

In this collaborative project among Samsung Research, SNU HCI+d Lab, and Yonsei university, we aimed to design a coherent visual identity for Bixby: a smart agent deployed to Samsung appliances.

During the study, I took part in developing prototypes and conducting user experiment.

Example screen we used during the user study

[My role in this project]
Prototype development, running experiment, writing

I worked as an iOS developer at a local micro-mobility sharing startup for two years. During this time, I realized how crucial it is to understand potential users for a service to succeed.

To deepen this understanding, I took a 'Digital Ethnography' class, where students explored user groups around a chosen topic. I focused on why users hesitate to ride shared mobility devices and investigated this through ethnographic methods.

I conducted observational and interview studies with six college students. Although the scope was limited due to the pandemic, the experience gave me valuable insight into how HCI researchers apply qualitative methods to study real-world user behavior.

[My role in this project]
Independent project

Although text-based conversational agents (chatbots) hold great potential for journaling diets and managing personal health, they suffer from low user acceptance. To address this issue, we propose incorporating the persona of real-world caregivers into the chatbot.

Working with Youjin Hwang, we conducted a series of studies identifying several important linguistic features that make users perceive a chatbot as a real person. (MobileHCI '20 Late-Breaking Result)

[My role in this project]
Writing

Although text-based conversational agents (chatbots) hold significant potential for diet journaling and health management, they often face low user acceptance. To address this challenge, we propose incorporating the personas of real-world caregivers into the chatbot design.

Led by Youjin Hwang, we conducted a series of studies comparing user perceptions of a chatbot across different personas: a doctor, a family member, and no persona. Although the study was carried out using a Wizard-of-Oz methodology, we believe the findings can be extended to automated chatbot systems powered by NLP.

This study was presented in CHI 2020 Workshop (Conversational Agents for Health and Wellbeing).

Three personas applied to a chatbot

[My role in this project]
System development, running experiment, writing

Juvis, a Korean company specializing in diet management, needed a way to deliver key information and assist users with reservations. To address this, we designed a chatbot that guided users through the process. This experience deepened my understanding of chatbot design and later inspired a follow-up project on integrating personas into chatbot interactions.

Keyscreens of Juvis chatbot

[My role in this project]
System development

For more than 2 years, I worked as an iOS developer at Elecle, an electric mobility-sharing startup based in Korea.

Because every step of renting and returning a mobility device is handled through a mobile app, app development is a critical component of mobility sharing services. Therefore, we have placed strong emphasis on enhancing the app’s usability and usefulness for end-users, and have designed various interaction methods to improve the overall experience.

Elecle is currently available in more than 5 cities in Korea, including Seoul, Suwon, Bucheon, Gimpo, and Sejong.

Me with an Elecle bike XD

Keyscreens of Elecle iOS app

[My role in this project]
System development, design ideation

I found it incredibly difficult to keep track of the courses I took in college, and I was surprised to learn that many others faced the same problem.

To simplify the process, I developed an iOS app called ECE Grad. This app helps ECE undergrads easily track the courses they've completed and see how many credits they still need. It also automatically visualizes their course history and provides encouraging messages to motivate them to stay on top of their academic progress.

However, some users mentioned that manually entering courses into the system was a bit time-consuming. Based on their feedback, I added a crawling function to streamline the process: by simply entering their SNU credentials, the system automatically retrieves the relevant course data.

Keyscreens of ECE grad

[My role in this project]
Independent project

When I began my internship at the HCI+d Lab, I noticed that the website was somewhat outdated. I collaborated with my labmates to redesign it using a previously created template and updated it with the lab members' profiles.

It was a great experience contributing to the lab with my skillsets!

My profiile here:)

[My role in this project]
Development, design