雲端架構與 AI Agent，
從設計做到上線。Cloud architecture and AI agents,
from design to production.
我是 Tim Chiang，目前擔任解決方案架構師，主要做金融業的雲端架構、網路、資安與 AI Agent 專案。工作之外，我用 AI 開發自己的平台與工具。I'm Tim Chiang, currently a Solutions Architect working mostly on cloud architecture, networking, security and AI agent projects for financial institutions. Outside work, I build my own platforms and tools with AI.
獨立作品Featured projects
工作之外持續打造的個人工具與平台，把對 AI 開發紀錄、跨工具記憶與自動化的想法做成實物。Independent tools and platforms exploring AI development records, unified agent memory, and workflow automation.
Telos
企業 AI 控制平面：模型、金鑰、預算、護欄與稽核集中管理，客戶的業務可以用擴充接進來一起管。目前由我一人開發。An enterprise AI control plane for models, keys, budgets, guardrails and audit, where customers plug in their own business logic as extensions. Currently built solo.
Taimoe
記錄 AI 開發從開工、分級、提問、審查退回到放行的完整履歷。只記錄、不核准；所有數字皆由 Git、CI 與只增不改的事件表算出，不採信 AI 自述。A complete record of every AI-written change, from kickoff to release: risk level, questions asked, review pushbacks, and merge decisions. Records, never approves—computed from Git, CI, and an append-only log.
Axiom
自架的跨工具長期記憶伺服器，讓 Claude、ChatGPT 與 Gemini 共用同一顆大腦，底層結合 PostgreSQL 向量檢索、中日韓 Trigram 與時間衰減。Self-hosted AI memory exposed as an MCP server. Gives Claude, ChatGPT, and Gemini CLI a shared long-term brain powered by Postgres hybrid search (vector + CJK trigram) and recency decay.
Webs
台灣生活工具站：油價、菜價、空品、發票對獎等即時資訊，加上所得稅、電費等在地計算機，資料都來自官方來源。Everyday tools for Taiwan: live fuel and produce prices, air quality and receipt-lottery results, plus local calculators for income tax and electricity bills, all from official sources.
Homelab
用兩台實體主機架起自己的開發基礎設施：Git 與 CI、Kubernetes、資料庫與監控，也包括 AI 開發用量的觀測。A self-hosted development stack on two physical servers: Git and CI, Kubernetes, databases and monitoring, including observability for AI-assisted development.
PhilAI
早期實驗作品——第一個從零到正式上線完全由 AI 撰寫完成的獨立專案。An early side project and milestone—the first standalone web application shipped end-to-end entirely written by AI.
Graholder
股東會紀念品代領與訂單物流自動化平台，整合多據點收單、庫存批次處理與配貨追蹤流程；2026 年停止營運，作為完整做過一輪的營運案例保留。Shareholder gift logistics and order automation platform streamlining multi-site collection, batch inventory, and fulfillment workflows. Wound down in 2026; kept as a complete, shipped operations case.
技術筆記精選Field notes
從實際專案抽出來的架構模式與取捨，文章以英文撰寫。Architecture patterns and trade-offs distilled from real projects.
The Minimum Viable Landing Zone
GCP Landing Zone 第零天到底需要什麼：依「事後改起來多貴」排序每個決定，哪些可以安全延後。What a GCP landing zone actually needs on day zero, organised by how expensive each decision is to reverse, and what you can safely defer.
Multi-Cloud Cost Data Platform
把 AWS、Azure、GCP 帳單整進同一個可查詢模型：以 FOCUS 為正規化目標，帳號對應決定準確度，本質是批次問題。Unifying AWS, Azure, and GCP billing data into one queryable model: FOCUS as the normalisation target, account mapping as the accuracy layer.
Data-Loss Guardrails for an LLM Assistant
在受監管產業的對客助理前面做個資護欄：三種結果而非兩種、分層偵測，以及讓你不自欺的評測紀律。Lessons from building a personal-data guardrail in front of an assistant in a regulated industry: three outcomes instead of two, layered detection, honest evaluation.
Multi-Agent Enterprise Assistant
橫跨多個知識領域的企業助理模式：路由分派、每個資料擁有者一個專責 Agent、依群組授權。A pattern for enterprise assistants that answer across many knowledge domains: router-based dispatch, one specialist per data owner, group-based authorisation.
實戰重點Selected experience
在高度監管的金融環境裡，把雲端基礎設施與生成式 AI 從概念驗證推進到正式環境。Taking cloud infrastructure and generative AI from pilot to production inside regulated financial environments.
金融業 AI Agent 落地AI agents for finance
為金融機構設計並交付 AI Agent 前導與正式方案，支援文件審查、客戶開戶作業與內部知識檢索流程。Designed and delivered AI agent pilots and workflows for financial institutions across document review, customer onboarding, and internal knowledge retrieval.
多 Agent 架構Multi-agent architecture
設計由總控 Agent 分派問題給多個專責 Agent 的架構，每個領域各自維護知識與權限。Designed multi-agent architectures where a router agent dispatches each question to specialist agents, each owning its own knowledge and permissions.
多平台 AI 助理AI assistants in everyday chat
把 AI Agent 接進 Teams、LINE、Telegram 等大家每天在用的通訊工具，不必另外裝 App 就能使用。Brought AI agents into the chat tools people already use every day, such as Teams, LINE and Telegram, with no extra app to install.
對客 AI 助理Customer-facing AI assistants
設計直接面對客戶的 AI 助理，從對話流程、敏感資料保護到業務系統串接，讓 AI 能在受監管的環境裡上線服務。Designed customer-facing AI assistants end to end, from conversation flow and sensitive-data protection to business-system integration, so AI can serve customers inside a regulated environment.
文件辨識 AIDocument understanding
用多模態模型辨識各類文件：自動分類、拆頁、擷取欄位並交叉核對，每個數字都能點回原始頁面。Multimodal document understanding: classification, page splitting, field extraction and cross-checks, with every number traceable back to its source page.
地端個資防護On-premises data protection
自建地端個資偵測與遮罩服務，規則加上自己訓練的小型模型，只用 CPU、離線運作，資料不出機房。A self-built on-premises service that detects and masks personal data, combining rules with small self-trained models. CPU-only and offline, so data never leaves the data center.
雲端防火牆導入Cloud firewall rollout
為既有雲端環境導入高可用防火牆，在不影響線上服務的前提下，分批把流量納入管控。Brought an existing cloud environment under a high-availability firewall layer, moving traffic over in phases without disrupting live services.
技術版圖How I build
穩固的雲端網路底座、可控的 AI Agent 執行層，加上可重現的自動化維運。Resilient cloud foundations, governed agent runtimes, and reproducible operations.
雲端底座CLOUD FOUNDATION
- Google CloudVPC、IAM、Landing ZoneVPC, IAM, landing zones
- Hybrid NetworkingBGP、專線與防火牆高可用BGP, interconnect, HA firewalls
- KubernetesGKE 叢集與工作負載隔離GKE clusters & workload isolation
AI 與 AGENT 層AI & AGENTS
- Vertex AI & ADK多代理協作與工具串接multi-agent orchestration & tools
- AI Gateway & MCP流量治理、長期記憶與開發紀錄routing, memory & dev records
- Guardrails敏感資料邊界與合規控管data boundaries & policy controls
交付與維運DELIVERY & OPS
- Terraform模組化基礎設施即程式碼modular infrastructure as code
- CI/CD & Cloud Run自動化建置與無伺服器部署automated builds & serverless
- Observability & SIEM用量追蹤、Prisma Cloud、Splunktelemetry, Prisma Cloud, Splunk
聊聊Get in touch
想討論雲端架構、AI Agent，或有想法與機會，歡迎直接聯絡我。Want to talk cloud architecture, AI agents, or an opportunity? Reach out.
關於這個網站。目前任職於伊雲谷（eCloudValley）擔任解決方案架構師。這裡整理我在 Google Cloud、Kubernetes 與 AI Agent 實作上的 技術筆記 與 文章；想進一步認識我的背景，歡迎看看 關於我。About this site. Currently a Solutions Architect at eCloudValley. I keep my field notes on Google Cloud, Kubernetes, and AI agents in Notes and Blog—or read more about my background on About.