Situation: everyone collects travel recommendations — a saved list here, a screenshot there, a friend's tip you half-remember. None of it is queryable, none of it is yours in any structured way, and when you actually plan a trip you start from scratch.
The method applied: put the whole thing in a structured brain. A DuckDB database holds the sources (guides, articles, my own trips), a taste rubric (what I love, what I'd never touch), and the evidence behind every recommendation — including my own GPS-matched photos. A scoring model ranks places against my preferences, not a generic "popular now". Then the site is derived from that database at build time: no CMS, no runtime database, no drift. Every recommendation on the page can point back to why it's there and, where possible, to a photo I actually took nearby.
Why it's a use case, not just a demo: it shows the full Structure-Beats-Magic loop on real, personal data. Curated sources feed a scored model; the machinery is shown, not hidden ("here's why this place ranks"); and the output is a published artefact you can throw away and regenerate. Swap the domain from travel to a client's product catalogue or knowledge base and the shape is identical.
What changes when structure is added: recommendations stop being a pile and become a defensible, personal, queryable system — one that gets better every trip instead of starting over.