Archives nationales de France

The Archives nationales de France, working with Inria, launched the LECTAUREP project to automatically read Parisian notaries' repertoires — an enormous corpus running to over a million heterogeneous, heavily abbreviated handwritten pages. The project began with Transkribus but migrated to the open-source eScriptorium to secure open-source control and full ownership of its trained models. That meant exporting PAGE XML ground truth from Transkribus and painstakingly retraining models in the Kraken engine. The team produced specialised models for 19th-century French administrative hands and built an open pipeline outputting TEI-compliant XML, greatly improving searchable access to the notarial record. A flagship national-archive HTR migration.