Hi, I'm Polo. A Data Engineer building scalable pipelines & AI systems.
Recent Articles
View All學習筆記 - 初探 Kafka, CDC 實作
初探 Kafka，了解 Kafka 的基本概念和使用方法，並實作 CDC。
Mar 21, 2026GitHub Pages 自訂部署：從 Build 到上線的完整流程
拆解 GitHub Pages 自訂部署的實作細節——GitHub Actions 如何知道要去哪抓你 build 好的靜態檔案？這篇用實際的 workflow 設定來說明整個流程。
Mar 15, 2026AI 工具比較：為什麼我選擇 Claude Code
2024-2025 年 AI 編程工具百花齊放。這篇文章整理我在評估各類工具後，最終選擇 Claude Code 的思考過程。
Jan 7, 2026Selected Projects
View AllAutonomous Code Repair Agent
AI / Developer Tooling
An autonomous coding-agent system that diagnoses reported bugs, implements fixes in isolated worktrees, runs deterministic verification and independent review, and delivers validated GitLab merge requests with resumable execution and safety gates.
LLM Compliance Agent
AI / LangGraph
Graph-based multi-agent system using LangGraph to automate regulatory notice processing, product delisting, and alert generation. Handles 200+ daily government emails with 99.5% accuracy.
Similar Product Search Platform
ML / Data Pipeline
Large-scale product similarity search serving 15M products using multi-modal embeddings (VGG + Sentence-BERT) with Faiss ANN indexing, enabling 10s-level batch retrieval.
Internal Backend Platform
Backend / FastAPI
RESTful APIs built with FastAPI serving 500+ users across finance workflows. Features Google OAuth SSO, RBAC, Docker/K8s deployment, and ELK-based centralized logging.
AI SRE
AI / SRE / LangGraph
A multi-agent Kubernetes incident response system that correlates alerts, performs RCA and safety analysis, routes human approvals, and executes policy-gated remediation with post-check verification.
BSMI & LLM Fine-Tuning
AI / Qwen2.5-VL / LoRA
An end-to-end vision-language pipeline that LoRA fine-tunes Qwen2.5-VL-7B to match product images against BSMI certification text, then scales GPU inference to produce threshold-based pass, ban, and review decisions.