Delivery Hero's Quick Commerce service provides customers with an easy and efficient way to order grocery items for delivery. However, with thousands of stores and millions of products available on the marketplace, the company faces the risk of revenue loss and customer churn if users struggle to find products that meet their needs and preferences. This talk will explore how Delivery Hero developed a product semantic similarity recommender using transformer-based product embeddings and vector search to identify similar products across various points in the customer's purchasing journey. We will also discuss the challenges encountered, the solutions implemented, and the next steps for this initiative.