AI-powered recommendations for a digital art platform
Turning real user data into intelligent, scalable systems that solve concrete business problems.
Sparsity cooperate to design and build a recommendation engine capable of suggesting personalized artwork (paintings, illustrations, and photography) to its users. The goal: help people discover pieces they would love, while giving emerging artists real visibility on the platform.
Recommendation is a deceptively hard problem. Get it wrong, and users see the same artist, the same collection, and the same style over and over again. Get it right, and the platform becomes a genuine discovery engine. Our team tackled this from two complementary angles.
Collaborative filtering, which learns from how users actually interact with content (likes, playlists, time spent viewing a piece) to recommend media enjoyed by people with similar tastes. Starting from simple distance-based methods, we progressively moved to matrix factorization, a technique that uncovers hidden patterns in user behavior and scales efficiently to large catalogs. The improvement was measurable: our evaluation metric (AUC) rose from 0.76 with early models to 0.93 with the refined approach, while recommendations became noticeably more diverse across artists and collections.
Content-based filtering, which looks at the artwork itself (tags, descriptions, and artist biographies) turned into numerical representations using AI-driven embeddings. This approach is essential for solving the "cold start" problem: brand-new artists and freshly uploaded pieces have no interaction history yet, but can still be recommended based on their content alone, helping new creators get discovered from day one. Combining both approaches into a single, efficient recommendation library gives the platform a system that adapts to established users' tastes while remaining fair to new and emerging artists, a balance that's critical for any healthy creative marketplace.