Case study · Redesign
Gemfind.
Visit the live siteA faithful modern rebuild of GEMFIND — my uncle-in-law Michael Paine's 1997 Windows program that mapped Australia's gemstone fossicking spots. Same data, same idea, better cartography.
01 The brief
In 1997, my uncle-in-law Michael Paine wrote GEMFIND — a Visual Basic 3 shareware program that plotted Australia's gemstone fossicking spots on a scalable map of the country. It ran on Windows 3.1, shipped as a ~500 KB self-extracting executable, and drew its data from a 1987 reference book. I wanted to bring it back: the same data and the same idea, but web based and built using modern tech.
02 What I made
Gemfind is a modern web app that plots the same localities Michael compiled (34 gemstone types, 316 localities, 634 gemstone-to-locality cross-references) on a real basemap, with the bidirectional gemstone↔town filter the original was built around. It's optimised for mobile and desktop.
03 Highlights
Before and after
The starkest part of the project is the jump: a grey Windows 3.1 window with a black-outline map becomes a clean, content-first web app on a real basemap. Same information, thirty years of design and tooling later. The redesign was the point.
Honouring the original
Faithfulness mattered more than tidiness. I kept the bidirectional gemstone↔town filter that made the original tick, and translated Michael's iconic multicoloured concentric circles (richness at a site) into modern petal clusters. Five gemstones with zero records still appear, greyed out, because they were in the original table. I didn't add anything that wasn't there, and I didn't drop anything without flagging it.
Data & provenance
Behind the map is a Python pipeline that rescued the 1997 Access/Excel data, cleaned it (every spelling fix and merge logged), and geocoded it against GeoNames, while keeping Michael's original coordinate grid as a sanity check. I used AI to help organise the data. I'm not proficient in Python but Claude did a great job of parsing all of the old data and making it usable for the web.
Same data. Same idea. Better cartography. The about page credits Michael in full and frames the site as an unaffiliated tribute, with the history shown openly as part of the design.
04 What it looked like
Thirty years apart — the same map of Australia.
05 What I learned
The real lesson here was about working with AI. I designed and built the whole project in a single morning at a café, entirely with Claude Code: the data pipeline, the cartography, and the front end. I'm happy to be upfront about that. Typing the code isn't the hard part anymore. The judgement is, knowing what to ask for and what to throw away. Taste is the part that doesn't commoditise.