Cur8or — AI-Powered Product Curation Engine

Challenge

Building and maintaining product collections for a large retail catalogue is slow, manual work. Merchandisers have to hand-pick items, then revisit every collection as products go out of stock, prices change or new lines arrive. At scale — thousands of products across multiple brands and locales — keeping curated collections fresh and on-brand becomes a full-time job, and the results still drift out of date. The client needed a way to describe what a collection *should be* and have the system keep it accurate automatically.

Solution

We built Cur8or for Shareight — a curation platform that lets curators write a plain-language brief, for example "flowy summer dresses in bright colours," and have AI assemble a matching collection from the live product catalogue. Collections are matched on meaning rather than keywords alone, so curators get relevant results without hand-picking. Business constraints such as brand minimums, quotas and product placement are applied automatically, and collections refresh on their own to stay current as the catalogue changes. Curators work in a clean, modern editor with live previews, backed by a service that handles permissions and a full audit trail.

Project Results

Cur8or replaces manual product picking with brief-driven curation that maintains itself, so collections stay accurate without constant attention. Curators can stand up a new collection in minutes and trust it to refresh on its own as stock and pricing move. The platform runs in production on Google Cloud across multiple locales (en-GB, de-DE), powering live storefronts.

Features and Benefits

• Brief-Driven Curation: Curators describe a collection in plain English and AI fills it from the live catalogue.
• Meaning-Based Search: AI matches products by meaning, surfacing relevant items that keyword search would miss.
• Business Rules: Brand minimums, quotas and placement rules keep every collection on-brand and commercially sound.
• Automatic Refresh: Scheduled re-runs keep collections current as stock, pricing and new products change.
• Audit & Access Control: Role-based permissions and full change history make curation safe and accountable.

Technologies Used

• Go: Core API for collections, users and audit, with secure authentication.
• Python: AI-powered search and collection-refresh service.
• Elasticsearch: Search index powering fast, relevant product discovery.
• Google Gemini & Vertex AI: AI models behind meaning-based product matching.
• Next.js & Tailwind: Curator-facing collection editor with drag-and-drop and live previews.
• PostgreSQL: Source of truth for collections, audit history and access control.

Development Process

We began by modelling how curators actually think about collections — as intentions rather than fixed lists — and designed the brief-to-collection flow around that. The platform was built as a set of well-separated services covering the API, AI-powered search, and the curator interface, so each part could scale independently. The editor was built for non-technical users, with live previews and drag-and-drop. Throughout, we focused on multi-locale support and on running cleanly in the cloud for scalable, repeatable deployments.