Live - MVP
LinkedCulture
Unified Cultural Search
LinkedCulture Semantic Search
A unified semantic search index of open-access records across museum collections including the Art Institute of Chicago (58,443), Rijksmuseum (47,156), Cleveland Museum of Art (41,279), The Getty (59,979), Harvard Art Museums (6,509), and the Met (4,277). Tests whether vector embeddings can surface meaningful relationships across collections when users search by concept, mood, symbol, material, or meaning, not just catalog terms. Built on Ollama embeddings (nomic-embed-text) and Qdrant. No LLM in the retrieval loop. The ingestion and embedding pipeline is custom-built to be rerun against updated collections and extended to additional institutions.
LinkedCulture Topics
Extends the same index into unsupervised discovery. Each cluster is labeled using the object descriptions and promoted to a draft topic for editing. The result is a set of thematic groupings that emerge purely from the geometry of the embedding space: objects that land near each other semantically, regardless of institution, date, or catalog category. Topics are a way of reading what the model already knows about the collection. To date, the system has identified over 1,939 clusters across 92 published topics, each under review before publishing. LinkedCulture Topics are then folded back into the semantic search results to enrich discovery further.
LinkedCulture Multilingual
Recherche sémantique multilingue dans les collections du Getty. Interface entièrement en français pour les utilisateurs francophones découvrant les collections par concept et signification. Utilise un index d'intégrations multilingues pour rechercher au-delà des limites linguistiques et institutionnelles.
LinkedCulture Hybrid Search (Beta)
Combines keyword precision with semantic discovery. Tune the balance between exact matches and exploratory search using an interactive slider. Hybrid fusion uses Reciprocal Rank Fusion to weight OpenSearch BM25 results alongside Qdrant vector similarity, returning faceted results across institution, object type, medium, creator, and image availability.
Who It Serves
Museums, archives, libraries, researchers, digital humanities teams, and cultural heritage organizations exploring semantic discovery across collections.
Pilot Fit
A useful pilot would index one collection or a small cross-institutional set, then evaluate search quality against real research, education, or public discovery tasks.
Shared Process
From fragmented inputs to usable outputs.
Ingest open-access metadata
Generate Ollama embeddings
Index in Qdrant
Search by concept, material, or meaning