Jeonbuk National University · Department of Computer Science and Artificial Intelligence
We study how code runs, and how to make it run faster — across web engines, LLM-based systems, and AI compilers. The work is empirical: we profile real engines, find the bottlenecks, and upstream the fixes.
Open source we contribute to
| Project | Description |
|---|---|
| Escargot | Lightweight JavaScript engine for memory-constrained devices, supporting modern ECMAScript in a small footprint. We analyze its internals and work on performance and memory optimizations. |
| Walrus | Lightweight WebAssembly runtime with full standard support, built around an interpreter with JIT compilation underway. We work on execution performance. |
| lwnode | Memory-efficient Node.js implementation running on top of Escargot, targeting consumer devices such as phones, watches, and TVs. |
Built in the lab
| Project | Description |
|---|---|
| Escargot Review Bot | Self-hosted AI reviewer for Escargot pull requests. Four LLM passes — defect, refactoring, compiler, and style — feed a judge stage that merges overlapping findings into inline comments, running entirely on a local LLM. |
| JCodeQuest | Gamified platform for algorithm practice, with LLM-generated problems, three-judge ensemble grading, real-time code battles, and automatic plagiarism checks. |
| demucs-lite | Compressing the Demucs source-separation model to run on mobile. FP16/INT8 quantization and audio chunking bring inference for one second of audio to roughly 100 ms on a Qualcomm NPU. |
| Iron Device Simulator private |
Web dashboard for a speaker protection library, visualizing speaker temperature and diaphragm displacement in real time from audio files or live microphone input. Built as an industry collaboration project. |
Public contributions to lab and upstream project repositories since January 1, 2026. Private repositories and private contributions are excluded.
In 2026 the lab made 228 commits and 47 pull requests across 6 public repositories.