I focus on building practical, secure, and human-centered solutions at the intersection of Artificial Intelligence, Cybersecurity, Linux Infrastructure, and Open Source Software.
- 🏅 2024 Golden Bell Award (廣播金鐘獎) — Innovation & R&D Application Award
- Project: AI Hakka Speech Generation Platform (AI 客語語音生成平台)
- Dedicated to minority language preservation and speech AI applications through TTS/ASR technologies.
Lightweight, SEO-focused, performance-tuning, and cloud storage optimization plugins.
輕量、專注 SEO、效能調校、端點收斂與雲端儲存卸載的 WordPress 外掛套件。
- omni-webmaster-seo-suite — All-in-one WordPress SEO & Webmaster toolkit for metadata, schema, and site optimization.
全功能 WordPress 效能優化與 SEO 工具箱，涵蓋結構化資料、中繼標記與全站體質調校。 - omni-wp-perf-hardening — Single-file performance hardening plugin that tames expensive endpoints with CDN-friendly cache headers.
收斂 WordPress 高成本端點的單檔效能防護外掛，提供 CDN 友善快取標頭以減輕主機負載。 - smart-file-renamer — Automatically sanitize and rename media uploads with accents and special characters for better SEO.
在上傳過程中自動重命名含有變音符號與特殊字元的檔案，生成乾淨且利於 SEO 的檔案名稱。 - smart-image-upload-resizer — Lightweight WordPress plugin that automatically resizes uploaded images, converts to WebP/AVIF, and optimizes speed.
輕量圖片上傳自動縮圖外掛，支援 WebP/AVIF 格式轉換，顯著提升網站載入速度。 - just-s3-offload — Lightweight, zero-dependency Media Library offload plugin for Amazon S3 & S3-compatible cloud storage.
輕量、零依賴的 WordPress S3 與相容雲端儲存卸載外掛，無痛轉移媒體庫靜態資源。 - just-gcs-offload — Seamless, dependency-free Media Library offload plugin for Google Cloud Storage (GCS).
輕量無依賴的 WordPress GCS 媒體庫儲存空間卸載外掛，支援直接上傳與公開讀取優化。 - zh-to-en-slug — Automatically translate Chinese post titles to clean English slugs using Google Translation API.
自動將文章中文標題翻譯為語意清晰英文 Slug 的外掛，大幅改善網址結構與可讀性。 - disable-all-thumbnails — Prevent generation of unnecessary thumbnail formats in WordPress to conserve server storage.
停用非必要縮圖格式自動生成，節省寶貴的主機儲存空間並減少磁碟 I/O。 - disable-all-comments — Completely disable comment functions and spam attack surfaces across WordPress.
全面停用留言功能與相關端點，徹底杜絕垃圾留言騷擾與潛在攻擊面。 - modern-rss-image-feed — Enhanced RSS feed support with modern image formats (WebP, AVIF) via Media RSS (
<media:content>) tags.
為 RSS Feed 增添現代圖片格式（WebP、AVIF）支援，相容各類新式閱讀器。
Clean, privacy-first, and utility-focused browser extensions and web tools.
純前端運算、隱私優先、無伺服器依賴的實用瀏覽器擴充套件與網頁工具。
- just-tab-reloader — Minimalist cross-browser extension that auto-refreshes tabs within randomized intervals and watchdog recovery.
支援隨機區間與防呆看門狗的分頁自動重新整理擴充套件，防止監控後台與登入工作階段逾期斷線。
- just-a-new-tab — A minimalist, glassmorphic new tab dashboard featuring dynamic backgrounds, clocks, weather, and daily quotes.
拾光新分頁 (Just a New Tab)：極簡美觀的毛玻璃風新分頁儀表板，支援高畫質動態背景、時鐘天氣、每日金句與快捷導航。
- just-image-video-downloader — Quick, clean one-click high-resolution image and video downloader browser extension.
簡潔俐落的跨瀏覽器一鍵高畫質圖片與影音下載擴充套件，快速擷取網頁多媒體資源。
- just-ad-blocker — Compact, high-speed, and privacy-focused ad and tracker blocker built on Manifest V3.
小巧、高速且注重隱私的廣告與追蹤器攔截擴充套件，完整支援 Manifest V3。
- unmark-web — Browser-first client that strips invisible Unicode marks and C2PA/EXIF/XMP/ID3 metadata locally.
100% 瀏覽端隱私優先 AI 浮水印、隱形標記與多媒體中繼資料清除工具。 - md-to-wordpress-converter — Client-side utility converting Markdown formatting directly into native WordPress Gutenberg blocks.
將 Markdown 語法轉換為 WordPress Gutenberg 原生區塊並支援一鍵複製貼上的排版工具。 - image-aspect-ratio-calculator — Privacy-first 100% client-side aspect ratio calculator, cropper, and high-resolution image exporter.
圖片與影片比例智慧計算、裁切與高畫質匯出工具 (Aspect Ratio & Crop Pro)。
Applied AI/LLM applications, model selection matrix, agent evaluation, and video generation.
前瞻大型語言模型應用、開源模型決策選型、Agent 評測體系與視訊生成架構。
- h3-ui — Lightweight web frontend and job queue for vLLM-Omni video generation servers (single-file, stdlib only).
適用於 vLLM-Omni 視訊生成伺服器的輕量網頁前端與任務佇列工具，支援隨機種子與結果復現。 - open-model-selector — Open-source AI model decision matrix and hardware pairing guide (Arena.ai Coding x CanIRun.ai).
AI 開放模型選擇決策與硬體搭配矩陣，結合選型方法論與硬體門檻量測，協助精準選型。 - llm-zhtw-agent-exam — Traditional Chinese LLM Agent benchmark: machine-verifiable test questions and evaluation scripts.
繁體中文 Agent 能力小考卷：機器可驗證的繁中 Prompt、推理邏輯與指令遵循測試題庫與自動化評測腳本。 - comfy-video-bench — ComfyUI HTTP API driver for MiniMax-H3 & LTX-2.5 with graph flattening, latent upscale, and thermal guards.
MiniMax-H3 與 LTX-2.5 視訊生成效能基準測試與 ComfyUI HTTP API 驅動工具，含圖形扁平化與散熱保護。 - Hakka Speech AI (客語語音生成平台) — 2024 Golden Bell Award-winning Hakka TTS/ASR speech models and minority language digitization research.
榮獲 2024 廣播金鐘獎之客語語音合成 (TTS) 與語音辨識 (ASR) 核心技術研發與本土語言數位典藏。
Humanist typography, minority language support, and modern webfont optimization.
本土語言字元支援、人文襯線風格與現代網頁字型切片優化。
- changming-serif-tc — ChangMing Serif TC: a humanist serif webfont for Traditional Chinese with Taiwanese Hokkien and Hakka coverage.
昌明體 (ChangMing Serif TC)：以思源明體為骨架的人文明體，柔化襯線圓角，完整涵蓋台客語用字，提供分片 Webfont 與外掛。
Enterprise on-prem infrastructure, GPU thermal guards, and containerized QNAP NAS packages (.qpkg).
企業級地端架構、GPU 熱管理看門狗與 QNAP NAS 容器化開源應用套件 (.qpkg)。
- gb10-ops — Thermal and memory guard daemon for keeping an NVIDIA DGX Spark (GB10) alive under sustained unified memory load.
NVIDIA DGX Spark (GB10) 守護與採樣監控程式，在持續高溫與統一記憶體 (Unified Memory) 負載下進行自動保護。 - onprem-ops-30days — 30-day on-premise infrastructure engineering: articles, architecture blueprints, and operational scripts.
地端維運三十天：企業地端基礎設施架構、系統維運系列專文與實務自動化程式碼索引。 - qnap-comfyui-qpkg — Install ComfyUI on QNAP NAS as an App Center package running on Container Station with NVIDIA GPU support.
QNAP NAS 上的 ComfyUI App Center 容器化安裝套件，支援 NVIDIA RTX GPU 硬體加速與即時生成。 - open-webui-ollama-qpkg — One-click Open WebUI + Ollama LLM environment on QNAP NAS Container Station with auto GPU detection.
QNAP NAS 的 Open WebUI + Ollama 一鍵部署環境套件，具備 GPU 自動偵測與輕量容器外殼。 - roon-qpkg — High-fidelity Roon Server deployment package for QNAP NAS with containerized status UI and SSD tuning guides.
QNAP NAS 專用之 Roon Server 高音質音樂串流容器化部署套件，含狀態監控 UI 與 SSD 優化指引。 - changedetection-qpkg — Pack changedetection.io website monitoring service into a native QNAP QPKG via Container Station.
將 changedetection.io 打包成 QNAP QPKG 之網頁變更監控容器套件，隨時掌握網頁更新。 - homepage-qpkg — Highly customizable self-hosted application dashboard packaged as an App Center QPKG for QNAP NAS.
QNAP NAS 專用之 Homepage 自託管個人應用儀表板 (Dashboard) 部署套件。 - qpkg-template — Ready-to-use template and scaffolding for packaging applications into QNAP App Center packages (.qpkg).
QNAP NAS App Center 套件 (.qpkg) 快速封裝範本與骨架專案，便於快速建立專屬套件。
Interactive Washi-style notebook platforms for footprint tracking, life milestones, and share card generation.
結合個人足跡打卡、人生里程碑收集與手帳風分享圖卡生成器的沉浸式 Web 應用系列。
- travel-achievement-list — Earth Travel Achievements: A Washi-style notebook platform for tracking global travel footprints and generating share cards.
🌍 地球旅行成就系統：結合全球旅遊足跡打卡、里程碑收集與手帳風分享圖卡生成的手帳平台。 - taiwan-travel-achievement-list — Taiwan Travel Achievements: Footprint tracking across Taiwan townships and attractions with custom share cards.
🌍 台灣旅行成就系統：深度走訪台灣各鄉鎮市區景點的足跡打卡手帳與分享卡生成器。 - japan-travel-achievement-list — Japan Travel Achievements: Track 47 prefectures footprints, unlock regional badges, and export travel cards.
🇯🇵 日本旅行成就系統：日本 47 都道府縣足跡探索、名所成就解鎖與和風手帳圖卡分享平台。 - developer-achievement-list — Developer Achievement System: Track software engineering tech stacks, career milestones, and dev footprints.
💻 軟體開發工程師成就系統：專為工程師打造的技術版圖、職涯里程碑與開發足跡手帳平台。 - life-achievement-list — Life Achievement System: Notebook platform for recording personal milestones, life footprints, and memories.
📖 個人人生成就清單系統：記錄個人生命重要歷程、生活足跡與暖心記憶的手帳風平台。 - gamer-achievement-list — Gamer Achievement List: 135+ unlockable gaming achievements, domain radar, and gamer title badges.
🎮 遊戲玩家解鎖成就清單：橫跨各大遊戲領域的 135+ 項成就打卡、領域雷達圖與玩家專屬稱號。 - anime-achievement-list — Anime Achievement List: Otaku achievement checklist, gacha cards, and shareable anime fan cards.
🎌 動漫成就打卡與稱號系統：動漫迷必備的成就清單、領域宅力稱號與專屬動漫分享卡。 - couple-achievement-list — Couple Achievement List: Relationship milestone checklist, intimacy radar, and sweet memory cards.
💑 伴侶與親密關係成就系統：情侶與夫妻的專屬甜蜜打卡、親密關係雷達圖與回憶圖卡。
Curated open-source projects, ported packages, and actively tracked repositories.
深度研究、引進移植與持續追蹤之開源生態專案：
🤖 AI Agent, LLM Harness & Prompt Engineering (智慧代理與開發增強)
- everything-claude-code — Agent harness performance optimization system, skills, memory, and research workflows for AI coding assistants.
Agent 效能最佳化系統、技能與記憶機制（支援 Claude Code、Codex、Cursor 等工具）。 - academic-research-skills — Academic research skill modules for Claude Code (research → write → review → revise → finalize).
Claude Code 學術研究技能模組（涵蓋文獻研究、撰寫、審閱到最終定稿之完整流程）。 - skills — Public ecosystem repository for Claude Code and agent skills.
開發者 Agent Skills 公開生態套件庫，提供豐富的可擴充技能腳本。 - notebooklm-mcp — MCP server for NotebookLM with grounded, citation-backed answers from Gemini.
NotebookLM 的 MCP 伺服器，讓 AI 代理直接調用具來源引用的 Gemini 知識庫。 - deepseek-harness — Plugin-based intelligent agent harness architecture for DeepSeek models.
DeepSeek Harness 插件式智慧代理架構，具備高度擴展性與模組化能力。 - moltbot — Cross-platform personal AI assistant running anywhere (the lobster way 🦞).
全平台個人 AI 智慧助理，支援多種作業系統與聊天平台。 - docforge — Documentation-driven development template and scaffolding for AI-assisted projects.
文檔驅動開發 (Doc-driven development) 之 AI 專案範本，加速規格制定。 - smolaideveloper — Lightweight autonomous developer agent designed for long context windows.
適用於長上下文的輕量自託管 AI Developer 代理，自主編寫代碼。 - shell_gpt — Command-line productivity tool powered by ChatGPT for rapid shell workflows.
終端機命令列 ChatGPT 高效生產力工具，加速 Linux/macOS 日常維運。 - SalesGPT — Context-aware AI sales outreach agent with autonomous conversation management.
具情境感知能力的自動化 AI 銷售與外展代理，支援多輪自主溝通。 - AI-Products-All-In-One — Curated collection of battle-tested ChatGPT and generative AI products and tools.
實用 ChatGPT 與生成式 AI 產品與工具彙整庫，精選優質開源專案。 - ChatGPT-Line-Bot — Turn ChatGPT into a versatile conversational bot on LINE messaging platform.
將 ChatGPT 整合至 LINE 通訊軟體之機器人，支援智慧對話與自動回覆。 - ChatGPT-Tinder-Bot — Discord and automated conversational agent integration utilizing ChatGPT.
Discord 與智慧對話整合之 Bot 工具，探索多渠道人機互動。
⚡ LLM Serving, Inference & Distributed Training (大型語言模型推理與分散式訓練)
- vllm & vllm-omni — High-throughput, memory-efficient LLM and omni-modality serving and inference engines.
高吞吐、高顯存效率的 LLM 與全模態 (Omni) 推理引擎，廣泛用於生產環境。 - DeepSpeed — Deep learning optimization library by Microsoft for extreme-scale distributed training and inference.
微軟深度學習分散式訓練與高擴展性推理優化庫，支援 ZeRO 記憶體優化技術。 - llama.cpp — High-performance inference of LLaMA models in pure C/C++ with minimal dependencies.
純 C/C++ 高效率地端 LLaMA 模型推理實現，支援各類硬體架構量化運行。 - stanford_alpaca — Stanford Alpaca instruction-following model training recipe and synthetic dataset.
史丹佛 Alpaca 指令微調模型訓練代碼與資料集，開創指令微調典範。 - taide-bench-eval — Benchmark evaluation suite for Taiwanese localized TAIDE language models.
台灣在地化 TAIDE 繁體中文大型模型基準評測框架與能力測試驗證。 - evals — OpenAI framework for evaluating LLM capabilities and open-source benchmark registry.
OpenAI 模型能力評估框架與開源基準測試登錄庫，標準化評測工具。 - Olive — Hardware-aware model compression, optimization, and compilation toolkit by Microsoft.
微軟硬體感知模型壓縮、優化與編譯工具，針對指定晶片進行極致加速。
🎬 Generative Media, Diffusion & Video Models (影像與視訊生成前沿)
- MiniMax-H3-DGX-Spark — Measured SM121 compatibility recipe for MiniMax H3 FL2VA on NVIDIA DGX Spark with vLLM-Omni and online FP8.
在 NVIDIA DGX Spark 上透過 vLLM-Omni 與在線 FP8 執行 MiniMax H3 FL2VA 之 SM121 相容配方。 - LWM — Large World Model (LWM): World models trained on million-length video and language sequences with RingAttention.
利用 RingAttention 在超長百萬上下文視訊與語言上訓練的大型世界模型。 - CogVideo — Text-to-video generation architecture featuring CogVideoX and foundation video models.
清華開源 CogVideoX 文字生成影片 (Text-to-Video) 模型架構與推理實現。 - ToonCrafter — Generative cartoon interpolation and keyframe animation synthesis framework.
卡通插畫動畫生成與畫面補幀 (Cartoon Interpolation) 前沿研究。 - Paints-UNDO — Generative framework modeling human digital painting behavior and stroke history.
模擬人類數位繪畫過程與筆觸歷程的生成式模型，再現創作步驟。 - OOTDiffusion — Outfitting Over Time: High-fidelity virtual try-on diffusion models.
高擬真虛擬試衣 (Virtual Try-on) 擴散模型，保留服裝細節與自然穿著姿態。 - StreamDiffusion — Pipeline-level solution for real-time interactive generation and live diffusion streaming.
管道層級之即時互動式圖像生成解決方案，實現毫秒級串流生成。 - StableCascade — High-efficiency text-to-image architecture based on Würstchen multi-stage compression.
基於 Würstchen 架構之高效高品質文字生圖模型，具備極快潛空間壓縮速度。 - MiniGPT-4 — Enhancing vision-language understanding by aligning advanced visual encoders with LLMs.
整合先進大語言模型增強視覺語言理解能力，具備圖像細部理解與對話推論。 - 3d-to-photo — Virtual photo studio combining ThreeJS and Stable Diffusion for product photography.
結合 ThreeJS 與 Stable Diffusion 的虛擬商品攝影棚，任意場景即時渲染。
🎨 ComfyUI & Diffusion WebUI Ecosystem (AI 繪圖節點與工作流生態)
- ComfyUI — Modular, node-based Stable Diffusion and generative AI GUI and backend.
強大且模組化的節點式 Stable Diffusion / Flux 工作流 GUI 與後端架構。 - stable-diffusion-webui & stable-diffusion-webui-forge — Classic AUTOMATIC1111 SD WebUI and the memory-optimized Forge branch.
SD WebUI 經典操作界面與 Forge 低顯存加速分支，社群核心繪圖工具。 - sd-webui-controlnet — ControlNet extension for Stable Diffusion WebUI enabling precise structural guidance.
ControlNet 精準控制生成擴充套件，精確約束姿勢、邊緣與深度。 - AIGODLIKE-ComfyUI-Translation — Multilingual interface translation plugin for ComfyUI menus and nodes.
ComfyUI 多國語言繁中/簡中介面翻譯插件，消除節點操作語言隔閡。 - kohya_ss — Training GUI for LoRA, Dreambooth, and fine-tuning diffusion models.
LoRA、Dreambooth 擴散模型訓練 GUI 工具箱，支援豐富調參選項。 - multidiffusion-upscaler-for-automatic1111 — Tiled diffusion upscaler with VAE VRAM optimization for ultra-high-resolution rendering.
平鋪式超解析度放大與 VAE 顯存優化擴充套件，低顯存生成億級像素大圖。 - a1111-sd-webui-tagcomplete — Tag autocompletion and Booru tag suggestion extension for SD WebUI.
提示詞自動補全與標籤聯想插件，大幅提升提示詞編寫效率。 - openui — Describe user interfaces using imagination and render them live in real time.
用自然語言想像 UI 並即時渲染呈現的前端生成工具，顛覆前端原型設計。
🗣️ Speech Synthesis & Molecular Biology (語音技術與分子預測)
- ChatTTS-ui — Local web interface and API wrapper for the ChatTTS conversational speech synthesis model.
對話式擬真語音合成 ChatTTS 之本機網頁操作介面與對外 API 服務。 - YourTTS — Zero-shot multi-speaker text-to-speech synthesis and voice conversion.
零樣本多說話人語音合成 (Zero-Shot TTS) 與語音轉換模型，快速複製聲音特徵。 - alphafold3 — Inference pipeline for AlphaFold 3 biomolecular complex structure prediction.
DeepMind AlphaFold 3 生物分子複合體結構預測推理管道，涵蓋蛋白質、核酸與配體。
🔒 Cybersecurity, Privacy & Reverse Engineering (資訊安全、逆向工程與隱私清理)
- watermarks-remover — Python service stripping AI provenance marks, Unicode marks, and metadata from media and docs.
清除跨廠商 AI 浮水印、Unicode 隱形文字標記、C2PA 與多媒體中繼資料的後端服務。 - ghidra — Comprehensive software reverse engineering (SRE) suite developed by the NSA.
美國國家安全局 (NSA) 開源之軟體逆向工程與反組譯分析框架。 - ettercap — Comprehensive suite for man-in-the-middle attacks and network security analysis.
經典中間人攻擊 (MITM) 與網路資安分析嗅探工具，支援封包過濾與注入。 - emux — Firmware emulation framework for analyzing and debugging embedded system binaries.
韌體模擬分析與安全測試框架 (Firmware Emulation Framework)，支援物聯網安全研究。
🛠️ System Infrastructure, Fonts & WordPress Tools (系統維運、字型與開源工具)
- Jellyfin-QPKG — QNAP QPKG deployment package for Jellyfin media server using Container Station.
QNAP NAS 原生 Container Station 封裝之 Jellyfin 影音媒體串流伺服器套件。 - jellyfin.org — Source code for the Jellyfin project website, user documentation, and blog.
Jellyfin 官方網站、技術文件與部落格專案原始碼。 - source-han-sans & source-han-serif — Open-source Pan-CJK fonts developed by Adobe and Google (Source Han Sans & Serif).
思源黑體與思源宋體開源字型專案，涵蓋東亞漢字完整字形。 - tauhu-oo — Traditional Chinese font supporting indigenous Taiwanese characters and romanization.
支援本土語言漢字與羅馬字的「豆腐烏」烏體字型，衍生自思源黑體。 - afdko — Adobe Font Development Kit for OpenType font creation and inspection.
Adobe OpenType 字型開發工具套件 (Font Development Kit)，專門用於字型轉碼與構建。 - snicco — Modern, secure enterprise WordPress architecture and development framework.
企業級 WordPress 安全與現代化架構開發框架，遵循嚴謹工程實踐。 - Synology_HDD_db — Script to add non-certified HDD, SSD, and NVMe drives to Synology compatible database.
擴充群暉 NAS 相容硬碟與 NVMe SSD 清單資料庫，解除未認證硬碟警示。 - panasonic_smart_app — Panasonic Smart App appliance integration for Home Assistant home automation.
Home Assistant 之 Panasonic 智慧家電整合套件，實現智慧家庭聯動。 - chatgpt-retrieval-plugin — Semantic search and document retrieval plugin for ChatGPT using vector databases.
OpenAI 語意檢索與向量資料庫外掛整合，提供外接知識庫增強。 - How-To-Ask-Questions-The-Smart-Way — Traditional Chinese translation of Eric S. Raymond's classic hacker guide.
經典黑客指南：《提問的智慧》繁體中文版，引導高效與專業的技術溝通。
| Category | Technologies & Tools |
|---|---|
| AI / Machine Learning | Python, PyTorch, Transformers, OpenVINO, Whisper, Hakka TTS/ASR, Ollama, vLLM |
| Cybersecurity | ISO 27001, ISO 27701, SIEM, Wazuh, Cloud Security, Security Automation |
| Infrastructure | Linux (Debian/Ubuntu), Docker, QNAP NAS, VMware, Nginx, Apache, Redis, GCS, AWS |
| Development & CI/CD | Python, PHP, JavaScript (WebExtensions API), Bash, WordPress Plugins, GitHub Actions |
"Practical. Secure. Open. Human-centered."
- 🐙 GitHub: github.com/ivanusto
- 📝 Blog: 優格網 (yblog.org)
- 𝕏 X (Twitter): @ivanusto
- 🧵 Threads: @ivanusto
- 📸 Instagram: @ivanusto
- 📘 Facebook: ivanusto
⭐️ Feel free to check out my repositories and drop a star if you find them helpful!