The problem
The Dienst Publiek en Communicatie (DPC), part of the Ministry of General Affairs, publishes around a quarter of all central government publications on open.overheid.nl: thousands of documents a month, with the inflow growing every year. Every document is still classified and described by hand. Backlogs grow while the inflow keeps rising.
The solution
A feasibility study with a working prototype. The prototype reads a document, proposes a classification and an introduction text, and shows a faithful preview of the publication page as it would appear on open.overheid.nl. The editor checks and decides. We compare several models, including European and open-weights ones, on a golden dataset of real publications, judged by the team and by an LLM as judge. An ethics session with the team and stakeholders is part of the project. Everything runs on shielded EU infrastructure, with public data only.
The result
A substantiated answer to what AI support delivers for DPC in quality, turnaround time, infrastructure and cost, and what it means for the editor, whose role shifts from processor to director.
