AI ENGINEERING · PRODUCTION AI AGENT SYSTEMS
把 AI Agent 從 PoC 推向 Production
我是 Bobo,一個有 20 年經驗的 cloud / backend technical lead,現在專注在一件事:把 LLM agent 當成真正的企業系統來工程化——不是跑得動的 demo,而是企業敢接到客戶資料、財務、製造流程上的 production 系統。
For English readers
I'm a cloud-native backend technical lead transitioning into production-grade LLM and agentic AI platforms. My background combines ~20 years of backend and cloud architecture, hands-on technical leadership, and deep practical work with modern AI-agent tooling (Claude Code, self-built MCP servers, Hermes Agent, OpenAB, multi-step agent workflows).
I design AI agent systems the way I design distributed systems: with attention to RAG architecture, vector retrieval & embedding strategy, tool-calling / MCP, agent memory, multi-agent orchestration, evaluation & observability, governance, and the latency / reliability / cost tradeoffs that decide whether an agent ever leaves the demo stage. I'm especially interested in taking agents from PoC to enterprise production — with eval, observability, access control and governance built in from day one — and in leading small (3–8 person) AI engineering teams to deliver them reliably.
The flagship series below is in Traditional Chinese; the competency map links each JD area to the deepest write-up.
能力地圖 Competency Map
每一塊都對應一篇深入的系統工程文章。
旗艦系列 Flagship Series
《從 PoC 到 Production:企業 AI Agent 系統工程》· 12 篇
市面上的 AI agent 教學多半停在「會呼叫 API、跑得動 demo」。這個系列補上 demo 到 production 之間最難的那一段:RAG、向量檢索、治理、可觀測性與團隊落地。
- 01. 為什麼企業 AI Agent 卡在 PoC?
- 02. 企業 AI Agent 系統架構藍圖
- 03. RAG 架構實戰
- 04. 向量資料庫與 embedding 策略
- 05. 權限感知檢索
- 06. Tool use 與 MCP
- 07. Agent memory 與狀態管理
- 08. 多代理協作
- 09. 生產級 LLM 可觀測性與評估
- 10. 延遲、可靠性、成本的系統權衡
- 11. Agent 治理框架
- 12. 帶領一支 3–8 人的 AI 工程小隊
延伸閱讀 Selected Writing
同一套能力,在這些既有系列裡有更多 hands-on 紀錄。
聊聊 Get in touch
如果你在找一個能把企業 AI agent 從 PoC 帶到 production、又能帶團隊的人,歡迎聯絡。