# Kos：AI Agents 與後端工程師

# Kos: AI Agents & Backend Developer

Canonical: https://www.aikos.cc/

專注金流可靠性、帳務一致性與異常復原。曾任支付、錢包與內容平台技術負責人，帶領 13 人以上團隊，以冪等、分散式鎖與自動對帳避免重複扣款、確保帳務一致；現以 Laravel、Golang、Python 與 Node.js 建構 AI Agent 工作流程與後端服務。

## 正在找新機會
- 職位: 資深後端工程師、技術主管
- 型態: 全職、約聘、接案
- 地點: 台灣、海外或遠端
- 語言: 中文和英文

## 序章 (2008–2013): 從客戶的需求開始寫起

2008 年在知識科技入行，同時扛好幾個企業網站、活動網站與客製化平台。之後用 ASP.NET 與 C# 寫企業網站與內部管理系統；在易透網，用 Delphi 寫 ActiveX 接上晶片卡讀卡機，為政府機關與企業做身分驗證與安全登入。

## 第一章 (2014–2018): 第一次帶人，第一次對帳

在幸福家不動產擔任資訊主管，帶 4 人以下的團隊，從架構、開發到 Linux 部署都自己來。接著在聲聯網重構多層通路分佣與帳務結算系統：可配置的分佣規則引擎、帳務流水與資金稽核。從這裡開始，算錢這件事不能出錯。

## 第二章 (2019–2024): 規模與速度

在傲創科技從零建置新聞採集平台與全文搜尋系統，每日處理十萬級新聞資料。可配置式採集框架讓新站點接入從約 2 天縮到半天內；搜尋回應從約 2 秒優化到 300 毫秒內。

## 第三章 (2024–2026): 一筆錢只能入帳一次

擔任支付、錢包與內容平台的技術負責人，管理 13 人以上團隊。回呼重送、網路重試都可能讓同一筆款項被處理兩次，於是用冪等、分散式鎖與自動對帳，讓同一筆交易只入帳一次，會員餘額與資金流水隨時對得上。

## 現在: AI Agents

現在以 Laravel、Golang、Python 與 Node.js 建構 AI Agent 工作流程與後端服務。工作中也把 AI 用在實處：分析分佣規則、檢測帳務異常、解析文件，以及輔助開發與監控。下一章，正在找。

---

Focused on payment reliability, ledger consistency, and failure recovery. Former tech lead for payment, wallet, and content platforms, leading a 13+ person team and using idempotency, distributed locks, and automated reconciliation to prevent double charges and keep the books consistent. Now building AI agent workflows and backend services with Laravel, Golang, Python, and Node.js.

## Open to new roles
- Roles: Senior backend engineer, tech lead
- Type: Full-time, contract, freelance
- Where: Taiwan, overseas, or remote
- Languages: Chinese and English

## Prologue (2008–2013): Learning on client projects

Kos started in 2008 at ZhiShi Technology, running several corporate, campaign and custom sites at once. Then came ASP.NET and C# corporate sites and internal tools, and at YiTou Net, Delphi ActiveX controls wired to smart-card readers: ID verification and secure login for government and enterprise clients.

## Chapter 1 (2014–2018): First team, first ledger

At Happy Home Realty, Kos became engineering manager of a team of up to four, owning everything from architecture to Linux deployment. At ShengLian Network came a rebuild of the multi-tier commission and settlement system: a configurable rule engine, an account ledger and fund audits. From here on, the money had to add up.

## Chapter 2 (2019–2024): Scale and speed

At AoChuang Technology, Kos built a news crawling platform and full-text search from scratch, processing news at the 100K-per-day scale. A configurable crawling framework cut new-site onboarding from about 2 days to under half a day; search went from about 2 s to under 300 ms.

## Chapter 3 (2024–2026): Every payment counted once

As tech lead for payment, wallet and content platforms, Kos managed a team of 13+. Callback resends and network retries can process the same payment twice, so idempotency, distributed locks and automated reconciliation make each transaction post once and keep balances and ledgers matched.

## Now: AI Agents

Today Kos builds AI agent workflows and backend services with Laravel, Golang, Python and Node.js, and puts AI to work on the job: analyzing commission rules, detecting accounting anomalies, parsing documents, and assisting development and monitoring. The next chapter is open.