Quiet Confidence
Rigorous Execution 沉穩自信
嚴謹執行

2025/12 — Present現在

Freelance自由接案

Contracted to independently design, build, and ship five production-grade AI systems end-to-end — an agentic research engine, a financial-report automation pipeline, an enterprise ERP/portal integration, a document-intelligence platform, and a real-time market monitoring agent. Everything under "Work" on this site was built solo during this engagement. The ERP/portal alone spans ~8,000 lines of code across 5 production pages; the report pipeline processes 100–200 broker PDFs daily on an automated schedule. 受託獨立從零開始設計、開發並上線五套生產級 AI 系統 —— 自主研究引擎、金融報告自動化流程、企業 ERP／入口網站整合、文件智能平台，以及即時市場監控代理。這個網站「專案案例」底下的所有作品都是這段接案期間獨立完成的。光是 ERP／入口平台就有約 8,000 行程式碼、5 個正式頁面；報告流程每天自動處理 100–200 份券商 PDF。

2025/7 — 2025/12

Lvwang Digital (旅汪數位股份有限公司)

Though officially an AI Software Engineer, my responsibilities spanned AI, full-stack, database, and DevOps — built a voice-guided tour system on Azure TTS, a real-time airport flight-info crawler, an async FastAPI + PostgreSQL backend, and the full Docker / Harbor / Nginx deployment pipeline. This cross-disciplinary work strengthened my engineering foundation; wanting to focus more deeply on core AI development, I chose to specialize further in that direction. 職稱雖是 AI 軟體工程師，實際工作橫跨 AI、全端、資料庫與 DevOps —— 開發基於 Azure TTS 的語音導覽系統、機場即時航班資訊爬蟲、非同步 FastAPI + PostgreSQL 後端，以及完整的 Docker／Harbor／Nginx 部署流程。這些跨領域的工作讓工程底子更扎實；而為了更專注在 AI 核心開發，選擇往這個方向繼續深耕。

2024/11 — 2025/7

Surglasses (台灣骨王生技股份有限公司)

Sole AI engineer at a medical device company — designed and shipped AsclepiusAI, a RAG-based anatomy Q&A system, from architecture to deployment. Shipped as the world's first AI-powered anatomy table, covered by international press. After launch, wanted to keep building — proposed a customer-data pipeline for continuous model improvement, but as a small team, my day-to-day shifted to product-direction discussions and cross-department coordination; left to stay hands-on. 在醫療器材公司擔任唯一 AI 工程師 —— 從架構設計到部署，獨立打造 AsclepiusAI（基於 RAG 的解剖學問答系統）。產品上線為「全球首款 AI 解剖桌」，獲國際媒體報導。產品上線後想繼續做技術，提案建立客戶資料 pipeline 持續改善模型，但團隊規模小，日常工作轉為產品方向發想與跨部門溝通協調；為了留在第一線工程工作而離開。

2024/1 — 2024/6

Epistar Corporation (晶元光電股份有限公司)

Retrained a wafer-defect detection model with data augmentation, lifting accuracy to 99%, and rewrote the data pipeline for a 24x speedup (12m30s → 30s). 導入資料增強技術重新訓練晶圓瑕疵檢測模型，準確率提升至 99%，並重寫資料處理流程，效能提升 24 倍（12 分 30 秒 → 30 秒）。

2023/6 — 2023/9

AU Optronics (友達光電股份有限公司)

Built AI-TalentSearch, a natural-language résumé matching system — screening 1,000+ résumés against a role description in under a minute. 打造 AI-TalentSearch 履歷媒合系統，用自然語言描述需求，一分鐘內從上千份履歷中找出最匹配人才。

2023/2 — 2023/6

ITRI (財團法人工業技術研究院)

Built a shelf-removal detection model and automated stability testing for a computer-vision pipeline. 建立 AI 模型判斷商品是否從貨架取出，並執行電腦視覺流程的自動化穩定性測試。