Draped
Buy less, keep longer. An app that tells you whether a piece of clothing is you before you buy it.
Draped started because my sisters and I wanted a better way of browsing Vinted. The idea is that it should feel like going round the charity and secondhand shops with the friend who has an eye for finding a gem, the one who holds something up and says that's so you.
It began as a colour-season shopping platform with an affiliate model. I killed that, because a product whose whole point is buying fewer clothes cannot earn money every time you buy new.
Now you do a short colour quiz, then Draped sits over the shops you already use, ASOS and Vinted first, and marks each piece as a keeper or not for you, on colour, on size and on what the fabric is made of. If you are about to buy new, it looks for the same thing secondhand first. One scoring engine, written once in plain JavaScript, runs in three places: the website, a Chrome extension on desktop and an iPhone app that hosts the shop's own site and injects the engine into it.
Decisions I would make again
Colour is measured in a perceptual Lab colour space, so two shades that look the same distance apart to a person are the same distance apart to the engine.
The engine clusters the colours in each garment photo with k-means. For a while the clustering was non-deterministic, and the same jumper could change verdict between page loads on about half the rack. Fixing how the clusters are seeded fixed the verdicts.
It is scored for precision over recall, because if the engine misses a keeper it costs you nothing, whereas a wrong keeper costs you a return. So the metric is F0.5 against a labelled set of 1,236 ASOS items, and the engine sits at around 72 on it.
Garment segmentation runs on the device. On desktop that is a SegFormer model in the extension. On the phone it runs natively through CoreML, which took one card from 396 to 120 milliseconds.
Phones get warm. The app keeps a per-minute heat ledger and slows its own pacing when the phone is hot, rather than draining the battery to keep up.
Where it is, honestly
The site and quiz are live and public. The extension is built and runs loaded from source; it is not on the Chrome Web Store. The iPhone app is an internal build on two phones with over-the-air updates. There are no users beyond my family, no price and no billing yet. That is where I think a consumer product should be before strangers see it.
The idea that started all of this was Fold Studios, an app for finding the cushions and rugs that go with your room. It is parked while Draped gets the attention.