Links
Overview
Eric Chen and I took inspiration from an in-person Vancouver skyline survey. We wanted to make the arguments around a policy question visible through a population of simulated citizens.
Process
A composer agent researches a prompt through Nia and generates citizens with distinct backgrounds and views. Small groups discuss the issue, then their stances and summaries update. We built the interface with Next.js, Tailwind, and Supabase, using CLōD for model access.
Reflection
Batching agent responses reduced the number of model calls while preserving individual perspectives. The generated citizens are a way to explore arguments, rather than measured public opinion.
Gallery
