Integration · 2025 – present
Search, recommendations & personalisation
Product discovery for a multilingual catalogue: from Klevu and Elasticsearch to Wizzy and Google Vertex AI Search, with Vertex AI recommendations and Insider personalisation.
Search is where an e-commerce catalogue is won or lost. At Elryan I led the work on Klevu and Elasticsearch for relevance and response time, and integrated Constructor and Insider for search and on-site personalisation.
2026: a new search stack
- Search. The storefront and the app moved from Klevu to Wizzy, and the web storefront then moved to Google Vertex AI Search.
- Recommendations. Vertex AI recommendation blocks are a CMS component, so editors place them on any page of the site and the app. Customer events are indexed to feed the models.
- Frequently bought together. Bundles on the product page, chosen by hand or suggested by AI.
- Index. Magento’s own OpenSearch replaced the older search bridge, and reindexing moved to a dedicated server.
The harder problem is language: the catalogue and the customers mix Kurdish (Sorani and Kurmanji), Arabic and English, often in one query. That constraint shaped the open search and recommendation platform I designed.