Jimmy
Software engineer · systems & automation · Taiwan
About
I'm Jimmy, a software engineer based in Taiwan. Most of my work lives near the command line: Linux systems, Python and shell automation, more and more time pairing with LLM coding agents, and the occasional JVM web app for old times' sake. I run a small home lab to stay honest about how much I actually understand — servers, backups, monitoring, and the discipline of writing things down.
我是 Jimmy,在台灣工作的軟體工程師。大部分的時間在命令列附近:Linux 系統、Python 與 shell 自動化,越來越多時間在跟 LLM coding agent 並肩工作,偶爾回味 JVM 世界的網頁應用。家裡養了一台小伺服器,用來提醒自己對系統的理解還差多少——備份、監控,以及把事情寫下來的紀律。
Experience
LLM-assisted engineering
Working daily with agentic coding tools — curating context, building verification loops, comparing models, and keeping a human in charge of the diff.
Personal infrastructure & home lab
Designing and maintaining a small Linux-based home lab: automated backups, monitoring, DNS, and a documentation-first approach to operations. The lab is where half of my real learning happens.
Google IT Automation with Python — Professional Certificate
Coursework and practice files: Python, Git and GitHub, Linux troubleshooting, and using automation to stop doing the same thing twice.
JVM web development with Grails
Studied the Grails framework by working through a Google Calendar integration — my introduction to convention-over-configuration and the Groovy language.
Where the commits live, on and off, ever since.
Skills
Languages
Python daily Bash daily Groovy nostalgicAI & LLM tooling
Claude Code Codex CLI OpenCode prompt & context design agent workflows LLM APIsSystems
Linux (Ubuntu / Debian) Git & GitHub cron ZshPractices
automation-first plain-text documentation backups you actually test boring, reliable toolsProjects
Exercise files from the Google IT Automation with Python certificate — scripting, data handling, and Linux system administration drills.
A Grails application integrating with Google Calendar, used as a way to learn the framework and the JVM ecosystem.
Agentic coding notes
Working notes on LLM-assisted development: context curation, verification loops, when to trust the agent and when to read every line, and the etiquette of reviewing machine-written diffs.
Notes
Plain text ages well. Notes I wrote a decade ago in plain text still open today. Formats come and go; text stays.
rsync and a cron job. For a single machine, a boring rsync script has outlived every backup product I have tried.
Read the RFC once in a while. Reading a specification end to end is underrated. Half the mysteries of the web clear up on their own.
Certifications teach vocabulary; labs teach judgment. The certificate tells you what to call things. The 2 a.m. outage teaches you what they actually do.
Context beats clever prompts. What you let a model see matters more than how you ask. Curate the input and the output mostly takes care of itself.
Review the diff, not the vibes. An agent that writes ninety percent of the code still leaves you owning one hundred percent of it.