Each year, Malmö City receives hundreds of documents from associations applying for grants. Manually reviewing them takes over 400 hours annually.

 In this talk, we present a prototype developed by Malmö Civic Lab that uses Meta’s vision model together with the Bespoke Minicheck model to automate the review process. The system can read, interpret, and verify document content. We built the prototype using open-source tools and employed Vibe coding throughout the process.

We’ll walk through how the system was built, how well it performs, and the key challenges we encountered—particularly those related to OCR and the importance of clean, structured input data. This session demonstrates how AI can help streamline public sector workflows in practice.