Kloset

About

Kloset turns an outfit photo into a shoppable list. Upload a picture - a Pinterest look, a street-style shot, a friend's fit - and the app breaks it into individual garments, then finds real, buyable products for each one from Indian retailers, tiered Budget / Mid / High / Premium.WHAT'S BUILT AND WORKING- Garment identification via Claude with a structured output schema. It separates layered pieces correctly (a shirt worn open reads as outerwear, not a shirt) and infers audience, so a women's query doesn't come back with boys' jeans.- Visual product matching through Google Lens, plus a hand-built colour-correction layer of around 344 lines that exists because Lens is effectively colour-blind. This was the hardest part of the app and it is the reason the results are actually usable.- Streaming analysis: garments appear one at a time as they are identified. Measured on the live production backend, the first piece lands at about 5 seconds, where the path it replaced showed nothing at all until every piece was done.- Every identified piece is then searched automatically, through a bounded queue rather than a fan-out, so one scan fills the whole screen without anyone tapping. Search spend is a deliberate one-line lever: setting AUTO_SHOP_ALL to false restores tap-to-load, which charges only for the pieces a user actually opens and is a large saving at volume.- Caching throughout: Lens results, product-link resolution, and self-hosted image crops, the last added after a third-party image host went down and silently disabled Lens on every scan.- Affiliate router mapping merchant domains to affiliate networks, with live VCommission campaign IDs for around 12 Indian retailers (Ajio, Levi's, Uniqlo, Bearhouse, Neeman's, Lavie, Salty and others).- Monetization: RevenueCat SDK integrated, a 5-scan free tier, and server-side entitlement middleware so a patched client cannot grant itself Pro.- Deployed and running on Railway. Around 7,200 lines of TypeScript across 48 commits, plus an eval harness that runs a folder of outfit photos through the real analyze and search path and writes an HTML report, so you can measure a change instead of guessing at it.NOT LOCKED TO INDIAThe matching pipeline is geo-agnostic - garment identification and Lens visual matching do not care where you sell. Retargeting to another market means changing the Google locale parameters, swapping the merchant allowlist, and switching the currency test and formatter. Those sit in one server file and one small mobile helper, so it is a contained job rather than a rewrite.HONEST STATUS - PLEASE READ- No users and no revenue. The app has never been publicly released. I have device-tested it and it works, but there is zero real-world traction.- Not on the App Store. Everything needed to submit is in place (RevenueCat wired, paywall built, backend live), but I never made the submission.- In the demo video the free-scan counter reads 999 rather than 5. That is only because the limit was raised while recording, so the whole flow could be shown without hitting the paywall. The shipped allowance is 5.WHAT YOU GET- The full private GitHub repo (React Native / Expo mobile app plus Node, Express and TypeScript backend), transferred to you.- The affiliate router with all merchant-to-campaign mappings.- The eval harness and the test outfit methodology.- A written handover of what cost me weeks to learn: why Lens needs a colour layer, why the latency floor is the provider, and what I tried that did not work, namely parallel query broadening, fashion-embedding re-ranking and a cheaper text-search provider.- A walkthrough call if it is useful.WHAT YOU WILL NEED TO SET UPAPI keys are not transferable, so you will need your own Anthropic, SearchAPI.io, RevenueCat and Railway accounts. All four are self-serve signups. Every environment variable is documented in .env.example, including what reads it and what happens if you leave it blank, so setup is filling in a form rather than reverse-engineering the code. The VCommission affiliate account is registered to me, so you would re-apply under your own; approval is routine.WHY I AM SELLINGI have moved on to other projects and this one has sat untouched for a month. The hard part, making outfit-photo-to-buyable-product matching actually work, is solved and verified on a real device. It deserves someone who will ship it, and that is not going to be me.SUPPORT THROUGH THE PURCHASEI will stay with you for the whole handover, not just the repo transfer. That means a walkthrough of the codebase and how the matching pipeline fits together, help getting your own API keys wired up and the backend deployed under your account, guidance on the affiliate re-application, and answers to your questions while you get it running. I would rather you end up with something that actually works than close a sale and disappear.