SAMI: AI-Mediated Social Interaction in Online Learning project• Designed a learnersourcing system using an AI social agent to enhance both intrinsic and extrinsic motivation of learners.

StupidTutor: Understanding users’ comprehension ability from inconsistent large language models (LLMs)• Identified five types of output inconsistencies and designed a user study to understand users’ interaction with LLMs. • Spearheaded data analysis and visualization of a user study with 250+ participants using Pandas with Python and R. • Developed the back-end for a user study interface using SQLite and Flask, collaborating with a front-end developer.Grinder: Designing human-AI collaboration-based object detection in video commerce• Led 20+ remote unmoderated usability tests encompassing both qualitative studies and quantitative benchmark studies.

Pioneering virtual reality and explainable AI for neurodiversity in remote work scenarios• Compared and evaluated multiple AI models, and created different styles of explanations to display AI predictions.

Developing a real-time background segmentation model for the video conference platform• Developed a PyTorch model achieving 1.5x speed, 34% reduced loss, and 14% fewer parameters than the base model.• Leveraged unlabeled dataset training by knowledge distillation, improving accuracy by transfer learning and fine-tuning.• Integrated the developed model into the Samsung video conference platform, Knox Meeting.

Data analysis of Key Performance Indicators (KPIs) from LTE eNBs statistical data for KPI Modeling• Extracted and performed correlation analysis between KPIs from LTE eNBs data using Pandas with Python.• Optimized the LTE eNB KPI modeling based on identified correlations in collaboration with machine learning developers.