Household finance dashboard

This one is personal, and deliberately so — it's a finance dashboard I built for my own household, replacing a sprawling spreadsheet, and I use it every day. It consolidates income, spending, property, pensions, investments, insurance and net worth into one view.

I'm describing the architecture and the constraints, not the contents. The actual numbers live only on my own machine — that's the whole point — so nothing financial appears here. What's worth showing is how it's built, because the design decisions are the transferable part.

The decisions that matter

• No database — a YAML data layer. All of it sits in thirteen plain YAML files, edited inline and saved straight back. For a single-household tool, a database would be weight without benefit; flat files are inspectable, versionable, and trivially portable. Choosing the simplest thing that genuinely fits is a judgement, not a shortcut.
• Real data never leaves the machine, and never enters version control. The data layer is local-only by rule. The repository carries the app, not the figures.
• A demo mode that cannot leak. There's a fictional dataset for screenshots and walkthroughs, and a hard rule that demo and real data are never mixed in the same response. Names are driven by profile variables rather than hard-coded, so the demo is genuinely separate, not a lightly-edited copy of reality.
• An AI advisor that reads but never decides. There's a Claude-powered advisor over the figures — but it's framed as decision support. It surfaces and explains; it doesn't move money or take actions. Money is exactly the place to keep AI on a short leash.

Why it's here

Not for the numbers — for the discipline. The same instincts that keep a household's finances safely behind a boundary are the ones a business needs before it puts AI anywhere near its own sensitive data. This is that boundary, built and lived with daily rather than described.