感谢您的访问 😃

本博客旨在分享我在探索计算机图形学、硬件架构、XR（扩展现实）及移动应用开发等不同技术领域时积累的研究经验与项目成果，以及由此获得的独特见解。我致力于从多个维度对技术进行剖析——从理论基础到实际开发场景的应用案例，再到由此引发的社会议题。

目前，我特别关注分布式渲染与虚拟现实（VR）开发，致力于探索如何利用人类知觉特性来提升系统的效率与沉浸感。我的研究重点在于分布式计算环境下的模型部署优化技术，专注于以性能为核心的渲染管线设计。

Variable rate compression based on the scene gist / Colorization of the perception of scene gist (VRGaze)

我希望通过技术为解决社会问题做出贡献，并分享在此过程中获得的知识，创造新的价值。

期待在这里与大家分享最新的技术趋势与研究经验，成为一个活跃的交流空间。:)

Profiles

LinkedIn GitHub GitLab (Personal Projects)

Projects

TinyDB

Repository Rust

  • Simple database engine written in Rust
  • Supports file-based record management, query processing, indexing, SQL parsing, and materialization

KENSv3-go

Repository Golang

  • Golang port of the KAIST KENS network simulator
  • Implementing Reliable Transfer Protocol on top of IPv4, MAC protocol, and corresponding syscalls

Pathfinder

Repository React, Three.js, Typescript, Vite

  • Real-time 3D navigation clone, inspired by Openpilot
  • Integrated OpenStreetMap (OSM), OpenRouteService (ORS), Overpass API, and GeoJSON for interactive, animated map visualization on the fly

CRAYON

Repository React, Typescript, Gin, Python, Mediapipe, ffmpeg

  • Real-time video tutoring platform for kids living in countryside, focuing on the guidance of building rapport
  • Implemented an emotion recognition pipeline using Media PipeFace Mesh, processing video frames to detect facial landmarks and CV-based sentiment analysis, classifying student engagement levels for immediate feedback
  • Designed an interactive node-based visualization cockpit to feedback session timelines in real-time, providing tutors with actionable, data-drived pedagogical suggestions (inspired by Foundry and n8n)

VRGaze

Repository Unity, Blender, Python, RenderDoc

  • Unity plugin for block-designed psychophysics VR experiments, featuring customizable block structures, stimulus control, and comprehensive data logging for Meta Quest Pro
  • Deployed the system on an HDRP-based rendering framework with cloth simulation, and optimized real-time performance through RenderDoc and Unity Profiler to ensure a stable 90 FPS for responsive user experience

projects.hyeondg.org

Repository React, Three.js, Typescript, Vite

  • New landing page for projects showcase (WIP)

GAP8-Transformer

Repository C/C++, Python, Rust, GGUF

  • Memory-safe DMA API and transformer inference engine on GAP8 RISC-V IoT microcontroller
  • 2‑phasing triple‑buffered DMA: Three on‑chip buffers that continuously rotate through fetch, process, and commit roles, so while one tile of data is being processed (MAC), the next tile is being fetched from DRAM and the previous tile is being written back, largely hiding memory latency for CNN/Transformer layers

hazardflow-kaist

Repository Rust

  • Part of KAIST Microarchitecture Design coursework
  • Hazardflow HDL port of riscv-sodor processor, implementing 5-stage processor design with I-cache and M extension

pintos-kaist

Repository C

  • Part of KAIST Operating Systems coursework
  • Educational operating system on x86_64, written in C

nvim-config

Repository Lua

  • Neovim plugins for Java, Python, Clang, Rust, Golang, Typescript and Lua IDE support
  • Preconfigured plugin manager, keymap, and plugins for LSP, autocompletion, and syntax highlighting, giving you out‑of‑the‑box language support similar to a lightweight IDE

DT

Repository Swift (iOS)

  • Complete functional clone of CASIO JS-40B
    • Grand Total (GT): Stores and recalls the running total of multiple separate calculations.
    • Independent Memory (M+, M-, MR, MC): Keys to add, subtract, recall, or clear a stored number independent of the current calculation.
    • Constant Calculation: Sets a number as a constant (by pressing × or ÷ twice) to easily multiply/divide a series of values by it.
    • Profit Margin (%), Tax Calculation (TAX+, TAX-), and Currency Conversion

Dual-n-Back

Repository App Store Swift (iOS)

Sharpen focus, memory, and attention with Dual-n-Back, an advanced cognitive training app designed around the scientifically proven n-back exercise—now extended up to quad-n-back for deeper mental challenge.

  • Intelligent Cognitive Training: Strengthen working memory and mental flexibility with synchronized visual and auditory tasks that adapt to your performance in real time.
  • Fully Customizable Sessions: Personalize your training environment with adjustable colors, voice styles, and display sizes to match your preferences or accessibility needs.
  • Adaptive Difficulty System: Stay in your optimal challenge zone with an algorithm that automatically adjusts difficulty as your accuracy improves.
  • Daily Motivation & Tracking: Build consistency through optional daily reminders and challenge modes that keep your practice engaging and progress visible.
  • Insightful Progress Analytics: Discover how your attention, memory retention, and multitasking evolve over time through detailed performance insights.

Dual-n-Back transforms daily training into a personalized mental workout designed to help unlock your full cognitive potential.