Jeroen Nelen
Jeroen is a Machine Learning Engineer at KickstartAI, a Dutch non-profit focused on accelerating AI adoption. Holding a Master’s degree in Computer Science focused on LLMs and their explainability, he now builds AI solutions for partners such as KLM and bol. In a recent collaboration with think tank DenkWerk, he developed this simulation as an additional empirical proof point for the latest report, Democracy in Revision.
Session
How do you test systemic democratic reforms when you can’t run live experiments on a
national parliament?
In collaboration with Dutch think-tank DenkWerk for their report “Democracy in Revision”, we
built a multi-agent digital twin of the Dutch House of Representatives. By configuring 15
party agents, a chairperson, and a minister, we simulated parliamentary debates to evaluate
structural reforms, such as introducing electoral thresholds.
In this talk, we break down the engineering and data science setup behind this multi-agent
architecture to simulate a debate. We’ll cover how we translated parliamentary procedure
into a LangGraph state machine, grounded agents using party-filtered RAG, and set up an
experimental framework to A/B test systemic reforms across 238 simulated debates and five
legislative bills.