軟體工程師

負責 AI 影片自動生成平台（OVideo，含旅遊業客戶專案）後端開發。平台需編排多項高延遲 AI 任務（LLM 腳本生成、image-to-video、TTS、lip sync），並整合多家 AI 供應商。重構影片生成管線與異步處理架構，將單支影片端到端生成時間由 45 分鐘縮短至 15–20 分鐘，效能提升約 60%。

1. Worker / 影片生成管線

• 設計並建置 PostgreSQL-based 自製異步佇列系統（8 張 per-task queue table + 6 個獨立 worker polling）：採 per-task 分表隔離而非單表加 task_type 欄位，避免任務洪峰互相阻塞（noisy-neighbor），並讓各 worker 可獨立設定並發上限（如 preview 50 / auto_export 2）；經評估後選用 PostgreSQL 而非 Redis / RabbitMQ broker，以降低維運複雜度。
• 實作 exponential backoff 自動重試（[30s, 2min, 5min]）與 dead-letter queue（DLQ），確保長時 AI 任務失敗可追蹤、可恢復。
• 建置端到端影片生成管線：LLM 腳本生成、image-to-video、TTS 語音合成、FFmpeg 影片合成、唇形同步（lip sync）、Whisper 字幕對齊，全程由佇列驅動自動化完成。
• 設計快取策略，重用已生成的影片片段，降低重複呼叫 AI 供應商的成本與生成時間。

2. API / 架構重構

• 將約 2,000 行的單體 main.py 重構為 9 個 FastAPI router 模組（約 40 個 endpoints），並抽出 service layer 分層（API / Service / Worker / Infrastructure 四層架構），提升可維護性與測試性。
• 將手動 SQL schema 管理導入 Alembic migration，實現版本化資料庫遷移與容器啟動時自動升級（PostgreSQL 15, SQLAlchemy 2.0）。
• 將 AI 輔助開發工具（Claude Code, ChatGPT）導入開發流程，用於大型重構、prototyping 與測試生成，並建立 CLAUDE.md / AGENTS.md / ADR 架構決策紀錄，沉澱開發規範與設計脈絡。

3. LLM / AI 整合

• 以 Factory pattern 實作 multi-provider LLM 抽象層（OpenAI GPT-4o / Google Gemini），支援文字生成與視覺理解（vision），可依設定切換供應商。
• 以 Provider / Adapter 模式整合 image-to-video 供應商（Runway ML / Google Vertex AI Veo），透過環境變數切換，並同時支援 sync / async 介面。

4. 測試 / 品質

• 以 pytest 建立約 670+ 測試案例（24 個測試模組），涵蓋 API endpoint、worker 與 service 層，搭配 pytest-cov 量測覆蓋率。
• 建立 GitHub Actions CI，於 push / PR 自動執行測試與覆蓋率檢查，確保重構不破壞既有功能。

5. 可觀測性 / 監控

• 自架 Prometheus + Grafana + Loki + Promtail 監控堆疊：整合 node / docker_stats / postgres exporter 監控系統與容器指標，以 Loki + Promtail 聚合 6 個 worker 的日誌，並建立擴線容量規劃 dashboard，支援問題根因分析與容量決策。

6. CI/CD 與部署

• 以 Docker Compose 編排 9 個服務（API + 6 workers + 排程任務），搭配 Nginx 反向代理，建立 staging / production 雙環境部署流程。
• 維運 GCP 與雲端儲存（Google Cloud Storage / Azure Blob Storage）整合。

技術棧：Python, FastAPI, SQLAlchemy, Alembic, PostgreSQL, FFmpeg, OpenAI GPT-4o, Google Gemini, Vertex AI, Runway ML, Whisper, pytest, pytest-cov, GitHub Actions, Prometheus, Grafana, Loki, Promtail, Docker, Nginx, GCP

軟體工程師

任職於醫療健康照護科技公司，擔任後端工程師並兼任行動端開發；於 CTO 直屬的後端團隊中擔任 mentor、帶領 2 位工程師，並與 PM、設計及醫療專業團隊跨職能協作交付產品。參與 15+ 個 Django 微服務（主導其中 8 個）與多個 Flutter 行動應用，產品線涵蓋口腔肌力健康管理（UOral2）、數位認知行為治療（ICBT）、長照訂閱管理（UanSubscribe）、多模型對話 AI（UanGPT）。

1. AI / LLM Systems

• 主導開發 UanGPT 多模型對話系統（團隊第一貢獻者），以 LangChain 統一介面整合 GPT-4o / Claude / Gemini / Azure OpenAI 與 AWS Bedrock，支援動態切換模型供應商，並落地輔仁大學醫學系臨床場域試點。
• 設計 RAG pipeline（Qdrant Vector DB + Bedrock Embeddings + Rerank），建置醫療領域文件檢索與知識問答能力。
• 主導開發診療評估單 OCR 辨識服務（PaddleOCR + Google GenAI），自動化紙本醫療表單數位化流程。
• 開發 DDK / MPD 口腔音訊檢測 API（librosa + WebRTC VAD 訊號分析），將語音生理檢測整合進臨床評估流程。

2. Backend Architecture

• 參與設計 Django REST Framework API 架構（50+ ViewSets、100+ data models），同時支援 Web / App / IoT 多端存取。
• 開發多階段臨床評估功能（前測/中測/後測 DDK 量測整合與問卷回呼 API），參與 ICBT 自動化治療流程系統開發。
• 獨立開發統一認證授權服務（uauth / uaccounts），建立 RBAC 權限、JWT / OTP 與 API token 認證機制。
• 主導開發健康量測 IoT 數據接入：建立 HRV 生理量測裝置與量測機台的數據寫入 API 及 LINE LIFF 量測介面。
• 主導開發推播通知服務（UNotify）、健康好生活（UGoodhealthlife）、居家照護（Uanaculife）等多個獨立 Django 服務。
• 以 Django TestCase / pytest 為多個服務建立單元與整合測試，確保 API 行為正確性與回歸防護。

3. Mobile App（Flutter）

• 主導開發 3 個 Flutter 行動應用（iOS / Android 雙平台）：口肌訓練 App（UOral2M 及其重構版）、復健追蹤 App（Resil）。
• 整合 Firebase Messaging 推播、TensorFlow Lite 端上推論，並自製 WebRTC VAD 語音偵測 Flutter plugin。

4. System Integration

• 維護與優化 TapPay / Line Pay / Apple Pay 金流流程：修正 Line Pay 卡號記憶與 3D 驗證、手機載具條碼驗證、統一第三方付款前端驗證邏輯。
• 開發 pgvector 向量檢索應用，支援 LLM embedding 與人臉特徵比對。

5. Infra / DevOps

• 建置 Docker 多階段構建 + Supervisord 多進程編排（Django + Celery Worker + Beat）部署流程，搭配 Nginx 與 Drone CI/CD。
• 維運 AWS（RDS / S3 / SES / Bedrock）與 GCP（Compute Engine / Vertex AI）雲端服務。
• 獨立撰寫公司 RD 技術規範文件（Django / Python / Docker / 資安規範），建立團隊開發標準。

技術棧：Python, Django, Django REST Framework, FastAPI, PostgreSQL, pgvector, Redis, Celery, RabbitMQ, LangChain, AWS Bedrock, Qdrant, RAG, Whisper, pytest, Django TestCase, Flutter, Firebase, Docker, Nginx, AWS, GCP

技術支援工程師

參與職安卡專案技術支援，派駐於職業安全衛生署，負責系統教學與署內工具開發。

• 支援職安卡課程及系統操作教學，協助培訓超過 300 名學員，確保學員短時間內熟悉系統操作流程。
• 開發 2 支 LINE Bot，提升用戶互動與服務效率。
• 協助建立署內法規函釋系統（公文管理平台），整合並管理超過 133 筆公文，提升公文流程效率。
• 開發資料匯入工具，維護法規網站資料完整性。
• 自學期間完成 CodeWars 題庫 75 題，並以獨立接案工程師身份交付 3 個專案（個人翻唱網站、B2C 電商平台、線上點歌系統），使用 Python (Django) / HTML / CSS / JavaScript / PostgreSQL。

資訊操作員 (OP)

負責政府機關資訊室的 IT 現場支援服務，處理資訊設備故障、帳號問題與日常維運。

• 處理超過 338 件 IT 支援案件，涵蓋帳號解鎖、故障排除、設備重啟等情境。
• 嚴格依照政府單位標準作業流程（SOP）操作，維持資訊安全與運作穩定。

Software Engineer

Backend development for OVideo, an AI video auto-generation platform (including projects for travel-industry clients). The platform orchestrates multiple high-latency AI tasks (LLM script generation, image-to-video, TTS, lip sync) and integrates multiple AI providers. Refactored the video generation pipeline and async processing architecture, reducing end-to-end generation time per video from 45 minutes to 15–20 minutes (~60% improvement).

1. Workers / Video Generation Pipeline

• Designed and built a custom PostgreSQL-based async queue system (8 per-task queue tables + 6 dedicated polling workers): chose per-task table isolation over a single table with a task_type column to avoid noisy-neighbor blocking and allow per-worker concurrency limits (e.g. preview 50 / auto_export 2); chose PostgreSQL over a Redis / RabbitMQ broker to reduce operational complexity.
• Implemented exponential-backoff retries ([30s, 2min, 5min]) and a dead-letter queue (DLQ) to make long-running AI task failures traceable and recoverable.
• Built the end-to-end video generation pipeline: LLM script generation, image-to-video, TTS synthesis, FFmpeg composition, lip sync, and Whisper subtitle alignment — fully queue-driven and automated.
• Designed caching strategies to reuse generated video segments, cutting AI provider costs and generation time.

2. API / Architecture Refactoring

• Refactored a ~2,000-line monolithic main.py into 9 FastAPI router modules (~40 endpoints) with an extracted service layer (API / Service / Worker / Infrastructure four-tier architecture), improving maintainability and testability.
• Migrated manual SQL schema management to Alembic migrations, enabling versioned database migrations and auto-upgrade on container startup (PostgreSQL 15, SQLAlchemy 2.0).
• Integrated AI-assisted development tools (Claude Code, ChatGPT) into the workflow for large-scale refactoring, prototyping, and test generation, and authored CLAUDE.md / AGENTS.md / ADR records to capture development conventions and design rationale.

3. LLM / AI Integration

• Implemented a multi-provider LLM abstraction layer with a Factory pattern (OpenAI GPT-4o / Google Gemini), supporting text generation and vision, with configurable provider switching.
• Integrated image-to-video providers (Runway ML / Google Vertex AI Veo) via a Provider / Adapter pattern, switchable through environment variables and supporting both sync / async interfaces.

4. Testing / Quality

• Built ~670+ test cases with pytest (24 test modules) covering API endpoints, workers, and the service layer, with pytest-cov coverage measurement.
• Set up GitHub Actions CI to run tests and coverage checks automatically on push / PR, ensuring refactors don't break existing behavior.

5. Observability / Monitoring

• Self-hosted a Prometheus + Grafana + Loki + Promtail monitoring stack: integrated node / docker_stats / postgres exporters for system and container metrics, aggregated logs from 6 workers via Loki + Promtail, and built a capacity-planning dashboard for root-cause analysis and scaling decisions.

6. CI/CD & Deployment

• Orchestrated 9 services with Docker Compose (API + 6 workers + scheduled tasks) behind an Nginx reverse proxy, establishing staging / production dual-environment deployment.
• Operated GCP and cloud storage integrations (Google Cloud Storage / Azure Blob Storage).

Tech Stack: Python, FastAPI, SQLAlchemy, Alembic, PostgreSQL, FFmpeg, OpenAI GPT-4o, Google Gemini, Vertex AI, Runway ML, Whisper, pytest, pytest-cov, GitHub Actions, Prometheus, Grafana, Loki, Promtail, Docker, Nginx, GCP

Software Engineer

Backend engineer (with mobile development) at a healthcare technology company; served as a mentor leading 2 engineers in the CTO's backend team, collaborating cross-functionally with PM, design, and medical professionals. Contributed to 15+ Django microservices (led 8 of them) and multiple Flutter apps, across oral-muscle health management (UOral2), digital cognitive behavioral therapy (ICBT), long-term care subscription management (UanSubscribe), and multi-model conversational AI (UanGPT).

1. AI / LLM Systems

• Led development of UanGPT, a multi-model conversational AI system (top contributor), unifying GPT-4o / Claude / Gemini / Azure OpenAI and AWS Bedrock behind a LangChain interface with dynamic provider switching; piloted in the clinical setting of Fu Jen Catholic University's School of Medicine.
• Designed the RAG pipeline (Qdrant Vector DB + Bedrock Embeddings + Rerank) for medical document retrieval and knowledge Q&A.
• Led development of an OCR service for clinical assessment forms (PaddleOCR + Google GenAI), automating digitization of paper medical forms.
• Developed DDK / MPD oral audio assessment APIs (librosa + WebRTC VAD signal analysis), integrating speech-based physiological testing into clinical evaluation flows.

2. Backend Architecture

• Co-designed the Django REST Framework API architecture (50+ ViewSets, 100+ data models) serving Web / App / IoT clients.
• Developed multi-stage clinical assessment features (pre/mid/post DDK measurement integration and questionnaire callback APIs); contributed to the ICBT automated therapy workflow system.
• Independently built unified authentication/authorization services (uauth / uaccounts) with RBAC, JWT / OTP, and API token authentication.
• Led IoT health-measurement data ingestion: data-write APIs for HRV devices and measurement kiosks, plus a LINE LIFF measurement interface.
• Led development of multiple standalone Django services: push notification service (UNotify), UGoodhealthlife, and Uanaculife home-care.
• Built unit and integration tests with Django TestCase / pytest across multiple services to ensure API correctness and regression protection.

3. Mobile App (Flutter)

• Led development of 3 Flutter apps (iOS / Android): oral-muscle training apps (UOral2M and its refactored version) and a rehabilitation tracking app (Resil).
• Integrated Firebase Messaging and on-device TensorFlow Lite inference; built a custom WebRTC VAD voice-detection Flutter plugin.

4. System Integration

• Maintained and optimized TapPay / Line Pay / Apple Pay payment flows: fixed Line Pay card memorization and 3D verification, mobile carrier barcode validation, and unified third-party payment front-end validation logic.
• Developed pgvector vector-search applications supporting LLM embeddings and facial feature matching.

5. Infra / DevOps

• Built Docker multi-stage builds + Supervisord multi-process orchestration (Django + Celery Worker + Beat) deployments with Nginx and Drone CI/CD.
• Operated AWS (RDS / S3 / SES / Bedrock) and GCP (Compute Engine / Vertex AI) cloud services.
• Independently authored company R&D technical guidelines (Django / Python / Docker / security), establishing team development standards.

Tech Stack: Python, Django, Django REST Framework, FastAPI, PostgreSQL, pgvector, Redis, Celery, RabbitMQ, LangChain, AWS Bedrock, Qdrant, RAG, Whisper, pytest, Django TestCase, Flutter, Firebase, Docker, Nginx, AWS, GCP

Technical Support Engineer

Onsite technical support for the Occupational Safety Card project at the Occupational Safety and Health Administration, responsible for system training and internal tool development.

• Conducted Occupational Safety Card system training for over 300 trainees.
• Developed 2 LINE Bots to improve user engagement and service efficiency.
• Helped build the internal regulatory interpretation system (official document management platform), consolidating 133+ documents.
• Developed data import tools to maintain regulatory website data integrity.
• Self-taught: completed 75 CodeWars katas and delivered 3 freelance projects (personal cover-song site, B2C e-commerce platform, online song-request system) using Python (Django) / HTML / CSS / JavaScript / PostgreSQL.

IT Operator (OP)

Onsite IT support for a government agency's IT office, handling equipment failures, account issues, and daily operations.

• Resolved over 338 IT support cases covering account unlocks, troubleshooting, and equipment restarts.
• Strictly followed government SOPs to maintain information security and operational stability.

私立華夏科技大學

Hwa Hsia University of Technology