Projects

Built a mobile IDE for Android and iOS that lets you work on your own machines from a phone: terminals, a file tree with inline Git status, an editor, GitHub pull requests and the coding agent of your choice, Claude Code or Codex, running either on the device itself or on remote hosts over SSH. Laid out for a single thumb, and as a desktop-style IDE on tablets and iPad. A companion Go server keeps terminals and agent chats running on remote hosts and shares them across devices.

Investigated an agentic framework for video moment retrieval by dissecting temporal queries using AST-driven structures and semantic operators. Evaluated LLM planners for reasoning over temporal relations and multi-event dependencies. Studied Adaptive Keyframe Sampling as a relevance-aware alternative to uniform sampling for long-video grounding. Developed hierarchical subqueries via spaCy dependency parsing to isolate core action predicates, attributes, and contextual modifiers for improved localization.

Benchmarked the K-EmoPhone mobile and wearable dataset for in-situ emotion and stress classification. Achieved 68.3% accuracy using a soft voting ensemble of XGBoost, and TabNet models. Focused on recent sensor-based physiological and self-reported data fusion.

Built K-means and Random Forest codebooks for bag-of-words representation, developed CNN models, and benchmarked against ResNet-50 using the Caltech101 dataset.

Compared PCA, low-dimensional PCA, PCA-LDA ensemble, and Random Forest for face recognition tasks. Evaluated subspace learning methods for joint reconstruction and discrimination.

Analyzed real customs data from Korea Customs Service and built a machine learning model to detect fake import declarations with over 91% accuracy.

Developed a simple calculator using JavaScript for web basics

Implemented and visualized a Perceptron classifier in action and utilized it for spam detection, comparing its performance with traditional ML models.

Created an animated eps of a soccer player losing their leg while kicking the ball. One of my earliest coding projects, but it was fun!