Online Wedding Dress Shop
Wedding-Hochzeit is an online-only bridal store that puts several dress manufacturers, an in-house couturier, and a network of local tailors behind one storefront - so a bride orders a made-to-measure wedding dress directly from the people who make it, fits it in person nearby, and sees it on herself with AI before a stitch is cut.
Tech stack:
The client
Kagama s.r.o
Wedding-Hochzeit is an online-only bridal store operated by Kagama s.r.o. It is not a boutique with its own rack - it is a marketplace. Several dress manufacturers, a single in-house couturier, and a network of local tailor partners all work behind one storefront, so a bride can order a gown made to her own measurements directly from the people who produce it, and still be fitted in person close to home. The public store ships in German and English; the back-office adds Russian for the teams that run production. Formamind built all of it - storefront, six staff portals, the order engine, and the AI - from the ground up.
The challenge
Selling without a showroom
A made-to-measure wedding dress cannot be dropped in a cart and shipped, and there is no shop to walk into. Each order opens a long chain - measurements, design, production at one of several manufacturers, shipping, fittings, alterations - handled by people who never share a room and who must each see only their own part of it. The brief was a store that feels calm and personal to a bride, sitting on top of an operations system precise enough to coordinate that whole chain while hiding from each party what isn't theirs to see.
Designing for quality
Two design languages
To carry couture, the storefront follows a strict editorial discipline: Playfair Display serifs over a champagne-and-pearl palette, hairline half-pixel borders, near-square one-pixel corners, wide-tracked uppercase labels, and restrained motion - underlines that draw in from the left, a loading mark built from two orbiting squares instead of a spinner. The AI features speak a deliberately different language: soft 18-to-28-pixel and fully rounded corners, dark violet glass gradients under a backdrop blur, a sparkle mark, springy easing and typewriter text. The split is intentional. The human craft reads as timeless and exact; the machine help reads as modern and alive - so a bride always feels which is which.
The architecture
Six roles, one pipeline
The platform runs six roles, each with its own portal: customer, admin, support, manufacturer, couturier, and tailor partner. A bride's order moves through a long, explicit lifecycle - awaiting measurements, payment, manufacturer assignment, production, shipping, arrival at her chosen tailor, fittings, adjustments, completion - and every stage stamps its own timestamp, so each portal shows a readable, ordered history rather than a vague status. Boundaries are enforced both at the edge and in every server layout: the manufacturer who sews a gown never sees its retail price, the tailor sees only the fittings that are hers, the couturier owns bespoke creations. Onboarding a new manufacturer or tailor is a first-class flow - the team creates the partner, a passwordless account is provisioned, and they are live; a manufacturer can even be set up before they have an email, so dresses can be assigned ahead of their first login.
The platform
Commerce engine, reshaped
Most of this is not how a shop works out of the box, so we used Medusa's flexibility to model a path a normal store cannot. A bespoke order workflow sits one-to-one on top of each native Medusa order: the commerce primitives - items, totals, tax regions - stay battle-tested, while the entire couture lifecycle lives in its own layer above them. Dresses and accessories are both native products distinguished only by metadata, each carrying the fields that drive the business - the manufacturer it routes to, plus-size surcharges, and the pre-assembled text the AI search reads - so a paid order auto-routes to the right maker with no manual assignment. Eighteen custom modules cover measurements, appointments, alterations, payouts, documents, support and promotions: a full operations backend built on commerce foundations rather than bolted beside them.
The AI
AI that sells the dress
When a bride cannot step into a showroom, the search has to understand her. Every dress is embedded into a Qdrant vector index from its real, admin-written attributes; a question in her own words is matched semantically, and a compact model - constrained to recommend only from the genuine candidates it is shown - explains the picks and links straight to those gowns. It reads intent, not keywords, and can never invent a gown that doesn't exist; if the AI is ever switched off, search falls back to a keyword matcher instead of breaking.
The virtual try-on then answers the question every bride asks: how will this look on me? From one selfie it generates a full-length, editorial image of her in the chosen gown - and accuracy was the hard part. The image model is given three references in a fixed order: her selfie as the only source of identity, the catalogue dress as the only source of the garment (the model wearing it in the photo is ignored), and a fixed salon backdrop. Both the person and the dress are first distilled into structured descriptions injected into the prompt, down to the lace type and her skin tone, and the prompt forbids beautifying, slimming or ageing her. It runs as a single pass on purpose: an earlier two-pass version regenerated the face twice and lost her likeness.
Responsible AI
Affordable and compliant
Generative AI is only worth shipping if it is affordable and lawful, so both were engineered in. Costs are held down with cheap default models, caching that describes each dress only once and memoises query embeddings, lazy one-time generation, a one-shot preview that is never re-billed, and a credit system that gates the expensive image step to signed-in customers; every call is written to a running cost ledger in dollars. On compliance: before any photo is sent, the bride ticks an explicit consent box; every AI image is labelled as an estimation, per the EU AI Act's transparency rule; her selfie and its analysis are never written to disk, and the raw photo bytes are never logged; and each AI action is recorded with a trace id for AI-Act traceability.
The engineering underneath
Installable, real-time, paid
The store installs as a Progressive Web App and, more usefully, carries web-push: the moment a bride places or pays for an order the admin team is notified, a booked consultation pings the couturier directly, and a booked fitting pings the specific tailor - each notification routed to the right role on the right device, with dead subscriptions pruned automatically. Payments run on Stripe Checkout across the four things a bride can pay for - her dress (VAT handled by Stripe Tax), a fitting, a fixed-fee online consultation, and AI try-on credits - and the flow is wired so a live payment and a developer test produce the exact same fulfilment, which kept the whole order pipeline honest throughout testing.
In conclusion
Craft, online
The result is a store that feels unhurried and personal to a bride, and runs like an operations desk for the manufacturers, couturier and tailors behind her - one source of truth from first enquiry to final fitting, with AI doing the parts a showroom used to. It is the kind of build we like most: equal parts beautiful and load-bearing.
If you are looking for a Tech Partner to help your business reach the next level, contact us below, and let’s get started!
Get in touch
Get in touch with us, describe your request and we will guide you through the process of turning your ideas into reality.
>_ HERE FOR YOU
email: contact@formamind.com
tel.: +48 512 478 042
By sending a message you agree to our privacy policy.