Subsidiemaatje
AI prototype for the City of Amsterdam
- AI prototype helping Amsterdammers find, choose, and understand the city's ~100 subsidy schemes, built at the Amsterdam AI Lab in Q4 2025 in opdracht van directie SIS (Subsidies, Inkoop, Juridisch Bureau Sociaal).
- Preceded by a three-month discovery phase: stakeholder interviews, a working RAG-based question-answering tool over the existing subsidy texts, and a Figma prototype tested with users, all used to scope what the build phase should become.
- Designed and built the Next.js frontend on the Amsterdam Design System: AI-element UI language (purple highlight, transparency notices) and the parallel “personal explanation / official source text” pattern that always keeps the legal regulation one tab away.
- Wrote a custom amsterdam.nl scraper after the site stopped being scrapable. It's now whitelisted by the city's web team and feeds a RAG layer with up-to-date regulation text for all 100+ schemes.
- Co-built with an AI specialist: the Flask backend, the multi-step LLM pipelines, and the synthetic-recall eval (5 fictive cases per regulation, measured across 99 schemes).
- Tested with 10 Amsterdammers; most found subsidies faster than via Google or amsterdam.nl and felt more confident in their choice. Code open-sourced; classified as a low-risk AI system by the city's Digital Strategy directorate.















