Digital wardrobe
Camera-first capture, clothing catalog, favourites, wear history, laundry state and manual e-commerce import.
Case study · Product engineering · 2026
ZUVAMA.
A digital wardrobe that turns clothes, context and personal preferences into easier style decisions — using AI only where it creates real value.
Private betaVisit Zuvama.it
RoleProduct · UX · Engineering
PlatformWeb app · PWA · iOS path
FocusWardrobe · AI · Try-on
StatusPrivate beta
01 · The problem
The wardrobe is full.
The decision is still hard.
Most people already own enough clothes. The harder problem is remembering what they have, combining it better and choosing quickly for the context they are in.
Zuvama is designed to reduce that friction: bring the wardrobe into a useful digital memory and turn it into concrete suggestions without replacing the user's own taste.
02 · The product
Three layers.
One experience.
The experience is designed to work as a useful product first and as an AI product second.
Outfits & context
Outfit creation and planning, calendar, mood and contextual hints connected to weather and city when the user chooses to share them.
Applied AI
Azzurra as a persistent stylist and an asynchronous virtual fitting room to preview outfits while preserving identity and garment characteristics.
03 · AI with a job to do
AI enters
only where it helps.
I did not want a chatbot placed on top of a wardrobe. I separated two different problems: deciding what to recommend and visualising how an outfit could look.
01
Azzurra · Stylist
Advice that remembers context.
A persistent text stylist built around the user's wardrobe and profile, with per-user AI budget control and usage accounting.
02
Virtual Try-On
Try before deciding.
Asynchronous generation with multiple references: the person's photo plus outfit garments. Moderation, retention and source-photo cleanup are part of the flow, not later add-ons.
04 · Engineering
A beta built
like a real product.
The technical direction favours an architecture that stays readable, observable and simple enough to evolve without premature infrastructure.
Core stack
Python / FlaskPostgreSQLRedis + RQOpenAIPrivate object storagePWA + CapacitorSentryPrometheus
Asynchronous AI flow
- 01Request
- 02Budget check
- 03Moderation
- 04Queue
- 05Generation
- 06Private storage
- 07Telemetry
- 08Cleanup
05 · Trust by design
Privacy and security
are part of the UX.
When a product handles personal photos and generative AI, trust cannot depend on a footnote.
- 01
Private storage and controlled access to results
- 02
Moderation before AI generation
- 03
Per-user budget and AI usage audit
- 04
Explicit retention for generated results
- 05
Source-photo deletion after virtual try-on
- 06
Technical telemetry without turning the product into surveillance
06 · My role
From problem
to working beta.
Zuvama is where I bring together the way I like to work: product thinking, engineering, UX, AI and operational discipline.
I drove product definition and priorities, architecture direction, responsive UX flows, AI feature integration and the hardening work required to turn a prototype into a beta that real people can use.