博客
思考与写作
关于编码、产品构建和数据分析的思考。
A Trend Line Is Not a Trend
Drawing a line through some points always produces a trend. Whether that movement is real, or just noise dressed up as a finding, is a separate question, and the one that should decide whether anyone acts on it.
Put AI Governance on the Request Path, Not in a Document
Most AI governance is a document written once and filed away. The more useful version makes every AI-assisted answer carry its own audit trail and compliance evidence, produced as the work happens.
A Forecast Is a Statement About Uncertainty, Not the Future
People want a forecast to tell them what will happen. The useful thing a forecast actually does is tell you how sure you are allowed to be, which is what lets someone make a careful decision.
Analytics Tells You What Happened. Intelligence Tells You What To Do.
Most data work stops at describing the past. The harder and more valuable step is turning that description into a recommendation someone can act on, framed by the risk they carry and the call they have to make.
Mentoring Made Me a Better Analyst
I started mentoring data science students to give something back. It turned out to sharpen the exact skill my day job depends on: making something complicated clear enough for someone else to act on.
Building Something Workable When There Is No Playbook
The hardest projects are not the technically hard ones. They are the ones with no infrastructure, no brief, and no precedent. What I learned about getting moving when clarity is low.
When the Categories Change Underneath You
The flashy part of data work is the model. The part that actually decides whether you can trust the answer is the quiet taxonomy work nobody puts on a slide. A story about a category that changed.
The Dashboard Is Not the Deliverable
Early on I thought the chart was the work. It took a few years of building analysis for people who had to act on it to learn that the real output is a decision, not a dashboard.
The Case for the Deliberate Generalist
Why I built my career across seven technical domains on purpose, and why breadth held to a consistent method is a strategic choice rather than a lack of focus.
Reproducibility Is a Feature, Not a Chore
Automating manual data workflows taught me that being able to run something again, the same way, on demand, is not housekeeping. It is the feature that makes everything else trustworthy.
208 Applications. One Yes. What the Data Says About Job Hunting.
A data-driven retrospective on 208 job applications, a 9.91% interview rate, 20+ resume versions, and what the numbers actually tell you about landing a role in Australia as an international graduate.
Academia, Internships, Industry: Three Worlds, Three Lessons
One month into industry after years in academia and internships. What I thought would transfer, what actually did, and the one skill that carried me through all three worlds.
I Let Claude Build My Blog. Here's What Happened.
94 pages, 6 months of iteration, and one weekend where an AI agent built the blog, RSS feed, newsletter backend, and dark mode. A reflection on building with AI.
My Inbox Had 3,000 Unread Emails. So I Built an AI Agent to Triage It.
How Claude Cowork + Gmail MCP became a side project: an agent skill that labels email by provider, builds sender rules, and archives the noise — without ever deleting a message.
I Let AI Build a Business App. It Worked. But Was I Thinking?
Building a CRM for my mum's food business with pure vibe coding. 10 hours of planning, zero manual input, and one uncomfortable question I still can't answer.
Hello World — Knowledge Compounds
Starting a blog. A reflection on why knowledge compounds, how first-principles thinking keeps it growing, and what you can expect to find here.
I Built an AI Fact-Checker for Climate Claims. Here's What I Learned.
Building a Transformer-based fact-checking system for climate change claims from scratch — no pretrained models, no BERT, no shortcuts. A COMP90042 NLP project retrospective.
2024, New Year, New Milestone
A heartfelt reflection on starting a master's degree, interning at HEX, rebuilding a portfolio, and the goals I carry into 2024 — the year I turn 25.