The AI personal stylist that styles people from the clothes they already own — and earns money the moment it recommends what they genuinely lack.
Karaoglu LTD · Investor briefing · 2026
of global carbon emissions come from fashion — more than aviation and shipping combined (industry estimates).
the average garment is worn before being discarded (industry estimates) — closets are full of unused value.
the "what do I wear?" decision is made under time pressure, and usually answered by panic-buying more of the same.
Fashion e-commerce optimizes for selling more clothes. Nobody optimizes for the wardrobe people already have — the place where the daily decision actually happens.
Photograph any garment — the background is cut on-device, then the AI names and classifies it (Google's model through our own server, the on-device engine as the fallback) and learns from every correction.
Scored, explained outfits for today's weather and occasion — color theory, formality, freshness. Transparent, never a black box.
Every swipe, save and wear trains a personal taste model. The stylist visibly gets smarter — users can inspect what it learned.
Style Circles: friends' closets become borrowable. When a friend owns the missing piece, the app says borrow — €0, zero new clothes.
Only when a real calendar moment meets a real gap does the app recommend a purchase — through commission-tracked partner links.
Make everyone feel excellently dressed every day using what they already own — advice first, always explained, free where it costs us nothing. Commerce appears only where a genuine need exists.
The operating system for the world's wardrobes: every closet digitized, every garment worn to its full life — and advertising replaced by advice, with brands paying to answer real needs, never to interrupt or to buy data.
In-browser background removal (self-hosted models, no cloud cost), silhouette classification trained on 54K labeled garments (MIT-licensed data), and a correction loop that makes it smarter for a lifetime.
Deterministic seven-axis scoring engine — color harmony & undertone depth, formality, weather, learned taste, proportion, pattern play, freshness — plus learned this-goes-with-that pair memory, with human-readable reasons on every look.
Conversational styling that remembers sizes ("my size is M"), answers sizing and color questions, and honestly declines what it can't do.
Borrow-before-you-buy across friends' closets, with real impact math (CO₂e, water, €) from saved looks.
Generative AI fitting (state-of-the-art image model) paints the outfit onto the user's own photo. Metered — the one feature that costs us money — and a session is charged only when a fitting is actually delivered.
Consumer app, Partner Studio for brands, Stylist Studio for human stylists — plus Mission Control for the operator, KPIs from a real event ledger.
Quality bar: 1582 automated engine tests · 7 end-to-end persona acceptance suites (including a real scan-pipeline test and an offline-resilience test) · zero-backend architecture that deploys to any static host in minutes (cloud accounts ship built-in, activated by connecting a free Supabase project).
Recommendations come from a deterministic engine stack that applies the same craft a trained stylist does — and explains every score it gives, in plain words, on the card.
Hue-wheel color harmony (tonal, analogous, complementary zones) with warm/cool undertone cohesion · dress-code coherence against the occasion · a warmth-vs-forecast weather model · learned personal taste · silhouette proportion (volume balanced against fit) · pattern play (statement prints anchored by solids) · wardrobe rotation. Weighted, summed, explained.
Every swipe, wear, save and shuffle nudges a decaying affinity model — including which exact pieces are worn together, not just which pieces are liked. A saturating confidence curve decides how far to trust it: neutral until proven personal, and old preferences fade unless re-confirmed, so the model tracks who the user is becoming.
On-device: background removal, silhouette classification against prototypes trained on 54K labeled garments, dominant-color reading and geometric stripe/check detection — all in the browser at zero marginal cost, improving for a lifetime from every user correction. By default a cloud eye (Google's model through our own server, fenced per device) reads category, fabric and the brand label; the on-device tier is the fallback whenever it cannot be reached.
No stochastic output: same closet, same day, same answer. Every constant is unit-guarded, every axis surfaces in the user-facing "Why?" breakdown, and the Stylist Brain screen shows exactly what the model consumed and learned. A language-model layer can sit on top for polish — without ever touching the decisions.
The full algorithm specification — formulas, constants, update rules — lives in the design document (§4), maintained to due-diligence grade.
Every commerce surface in updresser is designed to prove we're not chasing the user's wallet — which is exactly why the purchase suggestions convert.
When the closet lacks nothing, the shopping section says so and sells nothing. An app that tells you not to spend is the one you believe when it says buy.
Every gap offers the €0 borrow from a friend's closet before any shop link. Sustainability isn't a page — it's the default path.
The Stylist Brain shows users every signal the model consumed and what it learned. No black box, no dark patterns.
Partner links are labeled, commissions disclosed ("never at extra cost to you"), cloud-gated features say so instead of simulating.
| Rail | Mechanic | Margin profile |
|---|---|---|
| Affiliatelive in product | Style Radar gap cards → Zalando (via Awin) & Amazon links, typically ~8% of sale value (the Shop-My-Feed shelf is cold until the Stage-6 shop — D139) | ~100% — costs the user nothing |
| Premium · €/$/£ 9 per monththe same clean number in every market — honest pricing | 10 AI try-ons/week and perks via per-currency Stripe Payment Links (DE/US/UK live, worldwide fallback) | High — meters the one costly feature |
| Margin add-onsannual · packs · resale | Annual Premium at €/$/£ 90 (two months free, cash up front) · one-time try-on packs — 10 for €/$/£ 5 (~87% gross margin vs AI cost) · resale prepared in-app from her real wear record — she sells in her own marketplace account and keeps everything; the affiliate rail behind it is a labeled planning model (industry estimate €0.50–3 per completed sale), unsigned today | High — priced against unit cost, no new fixed cost |
| B2B challengesPartner Studio built | Brands host sponsored style challenges (reference price €10,000/mo), hyper-targeted on-device — age group, gender, country, learned style — with live KPIs from the event ledger — recurring card billing via Stripe subscriptions; lapsed campaigns pause automatically and reactivate on payment. Plus pay-per-click sponsored picks: one labeled product card in matched users' shopping feed at €0.50/click from fixed budgets — the click ledger is the invoice, and a pick can be need-matched: served only to wardrobes that genuinely lack the piece, paired on-device with clothes the user already owns | Pure software margin |
Founder rule, enforced in code: unlimited wardrobe, scanning, styling, learning and Style School stay free forever. Users are never squeezed — the model monetizes moments of genuine value.
Top of the daily feed. Anchored to the calendar, grounded in a real gap, borrow-first — so a purchase suggestion reads as planning, not advertising.
Christmas is in 8 weeks — one white shirt covers every dinner, service and family photo.
⇄ Borrow Mia's White Poplin Shirt — €0
or buy once, buy right: Shop at Zalando ↗ · Shop at Amazon Fashion ↗
Every click-out is a tracked affiliate event; every suggestion exists only because the closet genuinely lacks the piece and a real date is approaching. When nothing is missing: "Radar clear — nothing to buy."
global apparel market (industry estimates) — updresser sits on top of every purchase decision, without holding inventory.
European fashion e-commerce (industry estimates) — the affiliate pool the Style Radar taps with intent-qualified clicks.
consumers, human stylists, and brands — each side makes the others more valuable, on one codebase.
Styling apps proved demand (Stitch Fix, Zalon, Whering); resale apps proved the conscious shift (Vinted, Vestiaire). updresser combines both instincts at the moment of daily decision — before the purchase, where influence is highest and inventory risk is zero.
Come for the daily outfit, stay for the learning stylist. Style Circles make the product better with every friend who joins — borrowing needs a network.
Self-serve studio: calendar, bookings, client threads. Stylists bring their client books with them — acquisition that pays us.
Sponsored challenges inside the most trust-first styling context there is, with live cost-per-join KPIs. Brands fund reach; users get culture, not banners.
One link → installable app on any phone. No 30% platform cut on subscriptions, no review gatekeeping, instant iteration.
The landing page runs the real engine on a demo closet — visitors experience the product before creating an account (capped, fenced, upsell-ready).
Borrow-before-you-buy only works with friends inside — the core feature is the referral mechanic, and every invite is rewarded in-app (points + The Connector badge).
Every stylist promotes their own booking page; every brand challenge is a co-marketing campaign.
Full consumer product (588 shipped versions), cloud backend live (Supabase: accounts, roaming closets, the social Lounge with follows, profiles and private chat), the daily ritual with offline guarantee (one visit, then the U-Bahn test passes — asserted in CI), brand + stylist portals, operator console, revenue rails coded end-to-end, German-localized acquisition surface, public Style-School pages feeding SEO, EU AI Act Art. 50 machine-readable marking on every AI fitting (shipped three months ahead of the 2026-12-02 deadline), 1582 automated tests + 7 persona acceptance suites, due-diligence-grade design documentation.
Deployed on Vercel, installable PWA, nightly verified database backups. Stripe activates with account links — no engineering between here and first revenue.
No invented numbers: the KPI system reports only real ledger events, and demo entities are excluded from revenue by design. What we sell today is the machine, not a vanity metric.
The styling engine runs on-device at near-zero marginal cost per user; the two model calls — generative try-on and the cloud scan read — are metered and fenced, and an automatic server-side circuit breaker caps weekly spend, degrading quality before availability and never the other way around.
The app is live; the first cohort walks the daily ritual and retention is measured from the real ledger — the numbers every later step is gated on.
Stripe Payment Links first (the checklist is written), embedded Stripe after — tier buttons simulate honestly until the links exist, then revenue flows with zero rebuild.
Second vision provider chosen by bake-off; LLM stylist conversation behind the existing brain interface; the try-on refinement path tuned on real usage.
Capacitor wrapper once retention clears its bar (the evaluation is already written), stylist payouts (Stripe Connect), brand self-serve campaigns.
Every step is additive behind an existing interface — no rewrites. Full technical detail: the design document ships with the codebase.
The product is built and tested; the rails are coded. Capital goes to the three things software alone can't do:
Accounts and sync are live on free-tier rails; capital buys production-grade capacity, point-in-time backups, and the privacy-reviewed org-wide analytics build.
Landing the first sponsored-challenge partners and the affiliate volume deals.
Filling the demo funnel and the first thousand circles.
Let's talk about dressing the world from its own closets.
updresser — your closet, upgraded. · Live product demo on the landing page of this domain.