I’m Jun — I build across ML research, Go systems, and the web. Less polish-for-show, more things that earn their keep.
A minimal, educational implementation of Neural Architecture Search. Three search strategies on toy datasets (MNIST, CIFAR-10): uniform random baseline, population-based evolutionary search with mutation, and a REINFORCE RL controller that generates architectures autoregressively as 13-token sequences.
A real-time terminal dashboard for your local dev environment — kill the
ghost process behind EADDRINUSE without leaving the terminal. Auto-discovers
active ports, maps each to its process & PID, tracks port lifecycle and uptime,
exports snapshots to JSON/CSV, and kills hung processes with a single keystroke.
Cross-platform, refreshes every 3s.
A zero-disk-footprint CLI for ephemeral notes. Every note carries a time-to-live — when the timer expires, it’s gone. Supports burn-after-read secrets, clipboard copies that self-wipe after 10s, and a kill switch that erases everything, permanently. A scratchpad that never outlives the moment you needed it.
An interactive playground for data structures & algorithms — built to study without burning through YouTube tutorials. Press run, then watch everything play out at once, or step ahead one operation at a time with each move explained. Live and deployed.
An end-to-end sports-analytics pipeline for the 2026 FIFA World Cup. A current-Elo strength anchor, a Dixon-Coles-style Poisson goal model, and a calibrated gradient-boosted classifier are blended at the cross-validated optimum, then fed into a 10,000-run Monte Carlo tournament simulator that estimates every team’s odds of reaching each stage — served through a FastAPI backend and an interactive bracket dashboard.
Researched the growth and piezoelectric properties of Rochelle salt and KDP crystals under microgravity, comparing them to Earth-grown samples to test whether microgravity yields larger, more uniform crystals with fewer defects — and how that crystal structure influences performance in sensors, transducers, and electro-optic devices. Helped build the software infrastructure and testing process behind the experiment.
Developed the frontend GUI for HADES (Hazard Analysis Designer), a computational safety tool built under NASA’s System-Wide Safety Project and the In-time Aviation Safety Management System (IASMS). Built interactive diagramming interfaces and safety-assessment table generation, and researched best-practice packages for hazard detection in aviation.
Co-founded Billion and run operations, R&D, and development. 3rd place winner of the Congressional App Challenge.
Helped lead the mechanical design and assembly for the 2025/26 Navigator ROV.
I’m drawn to the small surface area — a CLI that does one thing well, a search loop you can actually read, a visualizer that makes the abstract click.
Most of my projects start as a personal annoyance or a thing I wanted to understand, then turn into something tight enough to ship. gaze and ephm came from everyday terminal friction; NAStoy from wanting to feel how architecture search actually behaves; dsavisualizer from being bad at picturing algorithms in my head.
I care about tools that are honest — fast, legible, and free of the cruft that makes software feel heavier than it is.
Alongside building, I’ve worked on real systems: a frontend GUI for NASA Ames’ HADES aviation-safety tool, microgravity crystal-growth research as a team lead at Quest for Space, and co-founding Billion (3rd place, Congressional App Challenge).