The situation
BoxDivvy is a food co-op that delivers members' orders every week. Packing starts with a spreadsheet exported from the order system. The organiser used to look up every product in a hand-kept master list of about 1,100 items, run spreadsheet macros, and print four sheets: Pantry, Bags, Bread and Produce. The macros worked. The pain was the 3 to 10 new products that arrived every single week, each needing to be classified by hand before anything would print properly.
What I did
I built a web app that fetches the weekly order spreadsheet itself and turns it into the four sheets, laid out exactly the way he already printed them: vertical headers, cells coloured by delivery route, landscape. Produce is sorted by how hard it is, hardest at the bottom of the box, then by size. Weighed goods come out as weights, like 10.5 kg.
When a new product appears, the run stops and asks. It doesn't guess, because a wrong guess quietly puts an item on the wrong sheet and the wrong box. But it helps: it suggests near matches (that '900g' is probably last week's '800g' renamed), shows the product photo, and makes it one click.
Claude writes the short names that fit on the printouts. Doing them in batches brought the cost of renaming the entire catalogue down to about $1.30.
A separate tracker records Woolworths, Coles and Aldi prices, so the co-op can compare what members paid with the supermarkets at the moment orders closed, and it flags which comparisons are genuine equivalents and which are stand-ins.
How it turned out
Built from May 2026 and in real weekly use. It's been through the usual life of a real tool: the co-op's login page moved and broke the fetch for a week, and a page that claimed to show twelve weeks of orders turned out to be showing one. Both found, both fixed.
