Case study · Luxury watches · Astral Watches
14 luxury brands, one catalog
In short
Astral Watches, an authorized retailer, needed one catalog from 14 luxury brand sites that have no API and sit behind Akamai and Cloudflare, in a field where one wrong technical term costs trust. alfa built one adapter per brand on a shared core, a locked technical glossary, a gate before every release and an AI scan of the live storefront. The catalog is live: 4,641 product pages across 194 collections.
Counts as of 28.10.2026.
The problem
A luxury watch page carries precise specs: caliber, movement, water resistance, case and bezel material. The 14 brand sites had no API, heavy Akamai and Cloudflare protection, and haute horlogerie language that generic translation gets wrong. A wrong term on a luxury watch page costs trust immediately.
What alfa built
- Protected-site intake
- One adapter per brand
- Locked technical glossary
- Gate before every release
- AI storefront scanning
Intake and one adapter per brand
Intake that reads each brand site through its protection, and one adapter per brand on a shared core. The core absorbed every new brand as an adapter, not a rewrite, including a dedicated ultra-luxury sub-store.
A locked technical glossary
Every watchmaking term is fixed once and used the same way on every page: mother-of-pearl, hairspring, power reserve, skeleton movement, with Côtes de Genève kept in French by rule.
A gate before every release
Before a product is published, the gate confirms that every spec field is mapped and in its required format. A product that fails does not go live.
AI agents on the live storefront
Data that is right in the master can still fail to render on the store, for example through a theme bug. AI agents open the live product pages as a shopper would, read the rendered spec values and compare them with the master. One sweep found watches hidden from the store by a single metafield error, which no database query had shown.
The result
An official authorized-retailer catalog, live at astralwj.co.il: 4,641 product pages across 194 collections, checked against the master and against the live storefront.
The intake runs on the build pipeline, and every release passes the validation gate.

