Show HN: Semantic Search over the National Gallery of Art (nga.demo.mixedbread.com)

A new open-source tool (repo: Mixedbread on GitHub) provides natural-language semantic search over the National Gallery of Art’s public collection, letting users query more than 50,000 images with everyday phrases like “still life paintings,” “woodcuts of landscapes,” “portraits of women,” or “ancient coins.” The interface translates free‑text queries into semantic matches across image metadata and visual content, surfacing relevant artworks beyond exact keyword matches and making discovery far more intuitive than traditional catalog search. For the AI/ML community this is a handy, real-world instance of multimodal retrieval: it demonstrates practical use of image-text embeddings and vector similarity search on a curated cultural-heritage corpus. That makes the project useful as a testbed for evaluating retrieval models (e.g., CLIP-style encoders), experimenting with fine‑tuning for art-domain semantics, building downstream applications (visual question answering, recommendation, dataset curation), and probing dataset bias or metadata gaps in museum collections. The repo and dataset can accelerate research into scalable vector search, relevance ranking, and human-in-the-loop curation for art and cultural‑heritage AI systems.