Submitting to the machine

KURB's spiral into the jaws of AI

Believe it or not, toiling endlessly with expanding the KURBiverse to solve problems for a 15 year old version of yourself does come with more moments of bliss than you would expect. But today is not one of them, and neither was the Christmas period. Luckily, my personal flavour of neural divergence allows me to enjoy arguing with ESL Dota2 players for 12 hours a day (non-incidentally also catering to my 15 year old self) as much as I do wrangling with code that makes up the most efficient clothing-grail hunter on the planet in equal parts. One monolithic and intrinsic problem across the entire secondhand clothing system is, say it with me, d a t a h a r m o n i s a t i o n. And what I mean by this is, for example: a Small in Japan is a substantially different size to a Small in the US. Even an Ann Demeulemeester Small in Japan is a different size to a Yohji Yamamoto Small in Japan. EVEN a Ann Demeulemeester SS25 Small in Japan is a different size to a Ann Demeulemeester AW12 Small in Japan. In short, it’s utterly incoherent. But how can we fix this within the KURB systems where we routinely collect listings from every secondhand clothing vendor across the planet for over 500 brands? The answer lies in A(l)i.

This festive season, I submitted myself to the machine and sacrificed whatever precious opportunities for relaxation I had developing models that offer us baby steps towards solving critical KURB search quality problems. This challenge doesn’t stop at sizing; noise and data discrepancies exist across every feature associated with a secondhand clothing listing, think color, material, condition, item category, etc. Unfortunately, we can’t employ an army of labour-exploitative mammals to carefully label the 100,000’s of items KURB indexes on the daily. So instead, we used just one highly trained labour-exploitative mammal to augment a battalion of deep-learned AI models to solve these problems, making sure to also do our part in the AI labour moral panic. And I hate to say it but it’s kind of working out. Who would have thought that devoting a substantial amount of the world’s free capital into a bunch of over-engineered toasters would enable us to precisely systematise the categorisation of clothing based on image and text. Though the implementation didn’t come for free - I have personally sifted through 10,000’s of listings to manually train the AI to differentiate between an asian man’s well manicured big toe and a Lemaire bag. It worked so well in fact that at the end of January I deployed the fruits of that psychosis-inducing science into our item indexing for us all to enjoy!

Now you can rest assured that we have quashed the flood of Baseball Cards and anime figurines which routinely slid into the Craig Green and 14th Addiction sections. Our search interface, the K-Hole if you will, is finally a safe space for all us fashion nerds. Now we can, woefully gaze into our monitors as countless priceless archival secondhand designer pieces glide by our financially irresponsible eyes at rates never before seen, with sphincters slightly less tensed. But this is only the first of many sieges on my mental well being I will take for the sake of improving KURB industries. The upcoming object of my neurotic obsession will be refining the color, material and size harmonisation filters for all applicable listings. Seldom do marketplaces offer this information in a usable format, making this task particularly tedious, but luckily we did at least conceive a conversion matrix for how we will harmonise all clothing based on either a) their item category and b) ideally their direct measurements or otherwise entirely contrived alphanumeric sizing and c) whatever other scant features we feel will equip our forthcoming AI beast. Then there are 3 more models in the “improve core KURB item quality” pipeline which works in slightly different ways. In the end, all these tedious machinations are fueled by the belief that we will create the one true secondhand designer clothing search engine so that anyone can fully realise the potential of this incredible ecosystem. But ulterior to this, it will also enable us to offer a kick ass overview of the entire secondhand clothing ecosystem- something that’ll we will tease out over the coming months.

Consider this as me answering a bunch of questions no one ever asked for and as another announcement of our commitment to making KURB sick as hell. To complement that mission, I also want to invite you all to tell us how you think KURB can be better for you. We’re doing a big push to talk to all of you, so if you’ve got 15 minutes and want to tell us what you want from KURB now’s your shot. We are at your mercy! That said, I am going back to developing systems that systematise all the little KURB tasks I am feverishly obsessed with!