IOFFICINA VESTIUMMMXXVI · 2026
Modeora
Tab. IAn AI content workbench for fashion brands: say one sentence, and out come the styling, the images and the copy. The model can only act through registered business tools.
Brief
Making marketing content for fashion e-commerce means gathering material from everywhere: product data in a spreadsheet, the brand’s voice in someone’s head, reference images on a cloud drive, copy in a chat history. Run it all through a single chat window and the context gets lost, nothing generated is kept, and an expensive image run is one wrong click away.
Process
I put the model inside a workbench made of business objects. Products, brand profiles, assets and results are things you can manage, not lines in a conversation.
The model cannot touch the file system or the database. Everything it can do is registered as a tool: read an object, change it, make an image, write copy, export. Each tool has its own schema and permissions. Bulk deletes, exports and costly generations first produce a plan, wait for a person to confirm it, and leave an audit record once they run.
Every generation is a task that can be queued, paused, and resumed after confirmation. Progress reaches the front end over SSE; if the stream drops, the front end catches up from the event log by cursor.
Outcome
The whole chain, from products, brands and assets to generation and export, works end to end, with all 33 back-end and 17 front-end acceptance checks passing. The site is live; for now it is a product prototype.