Email-to-Wardrobe pipeline.

During the war, I was stuck at home, reorganizing my closet, and I had this realization: I didn't know what I owned. What I paid, when I bought things, whether I'd actually worn them.

The numbers backed it up. 53 items purchased per year on average, 4x more than in 2000. 40% never worn. ~100 hours a year spent in wardrobe panic.

Every closet app on the market (Whering, Acloset, Cladwell, Alta) requires you to photograph every item before you get any value. Nobody finishes that.

But receipts already live in email. And after COVID, even physical stores started pushing emailed receipts. The coverage is there now in a way it wasn't a few years ago.

You sign in with Google. The system searches your inbox, filters out noise, sends real purchase emails to Claude for extraction, saves the results.

You get a browsable wardrobe: brand, category, size, color, price, order date, product images. Purchases come in four currencies. Prices are stored as-is and converted at display time with cached exchange rates, so you can see total spend in ILS even if you bought from Revolve in USD and ASOS in GBP.

Claude is the expensive step. It should only see emails that have a real chance of being fashion purchases. The entire pipeline is designed around that constraint.

The first version was an empty table and a PARSE button. I loaded 50 emails and immediately hit the first problem: Temu alone generated 10+ emails per order, all noise. I cut it. Get signal from the easy retailers first.

Then patterns started showing up. Small indie brands were invisible because they all send through Shopify. One query addition caught Tiger Mist, Free People, Reformation, Sugarfree. International brands route through Global-e. The search strategy shifted from “list retailers” to “find platforms.”

After the first full scan, I ran a cost analysis. 80% of Claude calls were wasted. Not on expected noise, but on things I wouldn't have predicted: medical bills from Israeli invoicing platforms (morning.co, ezcount), courier updates, Hebrew parking confirmations matching as order signals because “אישור הזמנה” means both “order confirmation” and “reservation confirmation.”

I traced every wasted call to specific senders, simulated fixes, deployed the ones that cut cost without losing fashion items.

The product would work fine as “Gmail Closet Scanner.” But I named it Kasane, Japanese for “to layer,” from the art of layering kimono fabrics in deliberate color harmonies.

Picked a terracotta and sage palette because it should feel like a closet, not a dashboard. Chose an editorial serif. Designed a logo with two leaves meeting at the stem of the K: one for raw input, one for refined output.