Building Hospital Information Systems and ERP systems at @tzuchi（慈濟）. Background spans finance/stock web apps and travel platforms, currently focused on bringing AI engineering into internal enterprise tooling. 在慈濟從事醫療資訊系統與 ERP 系統開發,過去經歷橫跨金融/股市網站、旅遊平台,目前重心放在把 AI 工程能力導入企業內部工具。

• C# / .NET
• SQL Server — 資料庫設計
• Odoo

• Azure OpenAI (AOAI)

• Anthropic / Claude

• vLLM

• Model Context Protocol (MCP)

• Deploying AOAI and on-prem vLLM models for internal automation features (e.g. meeting-minutes summarization) — implementation details are internal/NDA 導入 AOAI 與地端 vLLM 模型,支援公司內部自動化功能(如會議記錄整理),部分實作細節屬公司內部資訊,無法公開詳述

• Exploring LLM Agent design / MCP, Prompt Engineering, RAG, and LiteLLM / proxy gateway patterns for cost governance & observability 平常持續鑽研 LLM Agent 設計、MCP、Prompt Engineering、RAG,以及 LiteLLM/Proxy Gateway 的成本治理與可觀測性

• Claude Code
• GitHub Copilot
• ChatGPT · OpenAI Codex (no working open-source icon available for OpenAI's brand — shown as text)
• Antigravity CLI (Google)

gooaye-agent contains full copyrighted podcast transcripts and stays private — described here without a link. The rest are public. gooaye-agent 因含完整版權逐字稿維持 private,僅文字描述不附連結;其餘皆已開源。

gooaye-agent — AI knowledge-base agent built on a 689-episode podcast archive (private) Gooaye(股癌)知識庫與投資心態 AI Agent Skill

• Scraped and transcribed 689 episodes (540+ hours) of the Gooaye finance podcast, then extracted evidence-grounded notes per episode 抓取並轉錄 689 集(540+ 小時)Gooaye 股癌 podcast,逐集萃取有逐字稿佐證的重點筆記
• Packaged as a dual-mode Claude Agent Skill: archive lookup mode (cites episode + transcript evidence) and a persona-driven investment-mindset review mode 封裝為雙模態 Claude Agent Skill:模式 A 做歷史觀點查證(附集數與逐字稿引用),模式 B 模擬主持人語氣做投資心態健檢
• End-to-end pipeline design: scrape → transcribe → evidence extraction → synthesis → agent persona packaging 完整展現「爬取 → 轉錄 → 證據萃取 → 摘要合成 → Agent 人格封裝」的 pipeline 設計能力

video_content_capture (vcc) — Local-first CLI for transcription & grounded reporting

• Python 3.13 CLI: probe (ffprobe) → audio extraction → transcription (on-device MLX Whisper or AssemblyAI) → Claude-generated report with evidence links back to transcript segments 依序執行 probe(ffprobe)→ 音訊抽取 → 語音辨識(本機 MLX Whisper 或 AssemblyAI)→ 產生具逐字稿證據連結的報告
• Explicit safety boundaries: speakers anonymized, timestamps always derived from canonical segment IDs (never trusted from model output), predictive content labeled as "speaker's opinion, not verified fact" 明確劃出安全邊界:講者匿名化、時間戳一律由程式依 segment ID 產生(不採信模型輸出的時間)、預測性內容標註為「講者觀點,非獨立查證」
• The generalized transcription pipeline behind gooaye-agent 是 gooaye-agent 背後轉錄 pipeline 的通用化版本

cc-token-statusline — Claude Code statusline tool, published via npx

• Single-line display of context-window usage %, weekly rate-limit remaining, and per-conversation token usage (input/output/cache) 單行顯示 context 用量百分比、週限額剩餘、單次對話累積 token(input/output/cache)
• Zero npm dependencies, installable directly via npx -y cc-token-statusline — no clone or build step 零 npm 依賴,npx -y cc-token-statusline 即可安裝,無需 clone/build

ai-daily-digest — End-to-end AI content pipeline (Expo + Supabase + LLM)

• Aggregates Substack, YouTube, and Hacker News sources on a daily pg_cron schedule, summarizes new items with an LLM, and pushes one daily digest notification 以 pg_cron 排程每日聚合 Substack、YouTube、Hacker News 等來源,用 LLM 摘要新內容,推播單則每日摘要通知
• Spec-driven: requirements and design tracked as OpenSpec changes, implemented test-first 採 OpenSpec 規格先行、測試先行的開發流程

stock_operation_system — Personal stock portfolio management system

• Vite + React + TypeScript SPA on an Express + Prisma + PostgreSQL backend Vite + React + TypeScript 前端,搭配 Express + Prisma + PostgreSQL 後端
• Uses Decimal.js for financial precision and property-based testing (fast-check) to verify calculation logic 使用 Decimal.js 處理金融數值精度,並以 Property-Based Testing(fast-check)驗證計算邏輯

agent-voice — Voice notifications for AI coding agents

• Tiny cross-platform (macOS/Windows) CLI that lets Claude Code, Codex CLI, and similar agents speak when they finish a task, need input, need authorization, or hit an error 跨平台輕量 CLI,讓 Claude Code、Codex CLI 等 agent 在完成任務/需要輸入/需要授權/發生錯誤時用語音通知使用者
• v1 is deliberately minimal: system-native TTS only, no cloud services, no API keys, no GUI v1 刻意精簡:純本機 TTS,無雲端服務、無 API Key、無 GUI

• Email: hlchimeilawrence@gmail.com
• GitHub: @huijoson