Case Study: Pento models aesthetic taste with Qdrant

Pento improves taste-based recommendations by keeping 50% of clusters with Qdrant

Pento, an art-focused platform, faced the challenge of building a recommender system that could understand the nuanced and deeply personal aesthetic taste of its users, moving beyond traditional systems that merely suggest popular items. To connect people through shared artistic preferences, Pento turned to vendor Qdrant and leveraged its powerful recommendation API and multivector support to model these complex user profiles.

The solution implemented by Qdrant involved mapping user-art interactions into a semantic vector space, clustering them to identify aesthetic preferences, and creating a dynamic multivector representation for each user. This allowed Pento to use Qdrant's recommendation API to find artists whose work strongly aligned with a user's positive taste clusters while avoiding their negative ones. The result was a production-ready, taste-based discovery engine that provides precise and meaningful recommendations, effectively addressing the cold start problem and demonstrating a system that can be universally applied across various domains.

Pento