技能圖譜
10 年全端開發 + AI 產品經驗的完整技術版圖
Full-Stack 開發
React + FastAPI + PostgreSQL，從原型到生產的端到端交付
前端開發
8+ production projects
Career Creator, Xian, Personal Site
All modern projects, PostCSS
Personal site, landing pages
Career Creator real-time sync
後端開發
7+ projects, Pydantic v2, SQLAlchemy async
Poetry, ruff, pre-commit, pytest
100+ endpoints (Med Vision)
Vercel serverless functions
Self-identified gap, deepening
資料庫
5+ production DBs, async SQLAlchemy
Duotopia multi-env, Career Creator
Chinese Literacy Platform
Chinese Literacy, Vercel KV
Career iOS, Career Creator
測試與品質
All FastAPI projects
Chinese Literacy, Career Creator
CI integration, load testing
Skill + agent system enforced
LLM & AI 應用
從 API 整合到 Agent 架構，15 個 AI 專案的實戰累積
LLM 整合與開發
Vertex AI, GPT-4, Gemini, Claude
15 production AI projects
SDM system, FHIR, MaiAgent
P0 gap for agent reliability
AI Agent 架構
30 agents, 48 skills, 17 hooks
Skills, hooks, agent teams, MCP
Job Manager: 15 parallel projects
Understand protocol, not yet built
檢索與 RAG
Career iOS in production
Implementing 2026 Q1
Cohere Rerank planned for TEMC
語音與多模態
Duotopia, Xian voice PDCA
Voice-first UX design
Research phase
AI 治理與優化
Strong intuition, building framework
3-tier: Haiku/Sonnet/Opus
P0 gap for medical/finance
Planned for TEMC Phase 2
Cloud & DevOps
GCP 為主，GitHub Actions CI/CD，支撐 $200K+/月營收的基礎設施
Google Cloud Platform
5+ production services, auto-scaling
Multi-env, backup, connection pooling
Docker images, cleanup policies
Gemini API, embeddings
CI/CD automation, multi-project
Multi-env, VPN (Suhul POC)
CI/CD 自動化
11+ workflows, multi-stage deploy
Auto Cloud Run per PR/issue
20+ Dockerfiles across projects
Global hook system, secret scanning
部署平台
Med Vision, Personal Site, Henrytz
Suhul Tai'an migration POC
Not yet used, GCP via gcloud CLI
維運與可觀測性
Implicit with Cloud Run
GitHub Actions minutes tracking
No Prometheus/Grafana/Datadog
Single-region, no DR plan
All sync architecture currently
學習路線圖 2026
Production RAG：Career iOS 升級 L1→L2（pgvector + BM25） Structured Output：下 2 個專案加入 Pydantic V2 / Zod 驗證 GCP 日誌強化：Med Vision 加 structured logging
TEMC 學科 RAG（L2→L3 + Cohere Reranker） AI Observability：建立 audit trail 儀表板 TypeScript 深度補強：重構 1 個 FastAPI→TS 專案
VCA 課程上線：整理 Prompt Engineering 方法論 自建 MCP Server（GitHub Issues / GCP 整合） 監控系統：Cloud Monitoring + alerting policies
LLM Fine-tuning（LoRA）：TEMC 教育模型 Pub/Sub 非同步架構 災難復原計畫 + 多區域部署評估
實戰能力在台灣接案圈已經頂尖。下一步不是學更多技術，而是把現有技能從「能用」推到「精通」— RAG 深度、AI 可觀測性、Cloud 維運是最高 ROI 的突破點。