March 2026 — DS4SD/docling-core focused on stabilizing the document tagging pipeline by aligning picture classification enums with the DocumentFigureClassifier-v2.0 outputs, preventing serialization errors, and preparing for continued model evolution. Key changes include adding missing tokens/labels and updating extraction logic, resulting in improved reliability and downstream data integrity. This work enhances compatibility with the latest model and reduces risk during DocTags serialization across documents.

February 2026: Delivered a robust enhancement to PDF export in DS4SD/docling by including failed pages to preserve correct page numbering and structure, improving document processing reliability and downstream consumption. Implemented new pipeline support (_add_failed_pages_to_document), added comprehensive tests, and tightened resource management and code quality.

January 2026 monthly summary: Delivered two high-impact improvements across HKUDS/LightRAG and DS4SD/docling. Implemented Korean localization with language settings, UI localization, and ko.json cleanup; and improved preprocessing error diagnostics by including page numbers and refining exception handling. These changes extend reach to Korean users, reduce debugging time, and improve reliability of preprocessing workflows. Key results include faster localization readiness, improved user experience, and stronger maintainability.