DOREMUS

What it does

• Catalogue: record a work, a concert programme, a performance (foreseen or realised), a recording or a publication on simple forms. Each form becomes RDF, is checked on every save and keeps its history.
• Discover: search by casting (“a work for oboe and string trio”), follow a work’s history and the gap between programmed and performed, browse chronologies, maps and networks of composers and performers, and get recommendations that explain themselves.
• Ask in plain language: a Claude agent answers by writing and running its own SPARQL queries on the graph.
• Claude also drafts programme notes and reads printed concert programmes.

How the data is made available

• Source. The open data of data.doremus.org, built from the catalogues of the Bibliothèque nationale de France, the Philharmonie de Paris and Radio France: more than 550,000 works with their creators, casting and titles, and a selection of concerts and recordings, about 24 million triples.
• Alignment. People are linked to ISNI and Wikidata; composers recorded several times under different spellings are merged with owl:sameAs, which can be undone.
• Model. The DOREMUS ontology, an extension of FRBRoo and CIDOC-CRM for music, with its controlled vocabularies (instruments, genres, keys). Catalogue records are checked by SHACL shapes and carry a provenance record.
• Publication. The imported data stays read-only in its own named graph, so the catalogue’s records never mix with the institutions’ data. The interface is in French, English and German.

Technical stack

Knowledge graph

Apache Jena Fuseki, SPARQL, named graphs

Ontologies

DOREMUS (FRBRoo, CIDOC-CRM), SKOS, PROV-O, SHACL

AI

Claude API: an agent writing SPARQL (Opus 5.5), programme notes, reading of programmes

Search

Meilisearch; PostgreSQL for the application’s own data

Application

Python, FastAPI, pyshacl

Hosting

Docker Compose, Coolify on an OVH server in France