Poligent
Poligent, a policy simulation platform, is a multi-agent system that takes a government policy document and streams parallel reasoning from four demographically-grounded UK population archetypes using Claude Opus 4.7 wit
Core trick
Runs four demographic archetype agents in parallel with extended thinking, each computing household-budget impact from its own income profile.
Description
Poligent, a policy simulation platform, is a multi-agent system that takes a government policy document and streams parallel reasoning from four demographically-grounded UK population archetypes using Claude Opus 4.7 with extended thinking.
The platform ingests policy documents (starting with the 2010 UK VAT rise), uses a supervisor agent to produce personalised briefings anchored in each archetype's specific income and spend profile, then runs four archetype agents in parallel, each reasoning in first person about what the policy means to their actual household budget. A reporter agent synthesises the four reactions into a structured policy brief with distributional impact analysis and concrete recommendations. Results are validated against IFS published distributional findings.
Extended thinking is central to output quality: the archetypes don't produce generic policy commentary, they do the arithmetic against their own persona profile and cite specific numbers. The React dashboard at ai-policy-simulation.vercel.app streams thinking tokens, UK neural-voice narration (edge-tts, one voice per archetype), and the final policy brief in real time, backed by a FastAPI service on Azure App Service. All runs persist as JSONL for replay without API calls.
README excerpt
# Poligent: Policy Simulation Platform A multi-agent AI system that generates structured stakeholder reasoning for proposed fiscal policies in minutes — validated against published distributional evidence. Live demo: **https://ai-policy-simulation.vercel.app** --- ## Why this matters UK fiscal and social policy decisions are made with poor visibility into distributional effects across population segments. Standard practice is spreadsheet microsimulation combined with qualitative focus groups — typically weeks of turnaround between a policy proposal and any distributional analysis. This platform compresses that cycle to minutes. It ingests a policy document and streams parallel first-person reasoning from four demographically-grounded UK population archetypes, grounded in ONS Family Resources Survey income and expenditure data. The output is a structured policy brief with distributional impact analysis, concrete recommendations, and automatic validation against IFS published findings. --- ## What it does **Scenario:** The 2010 UK VAT rise (17.5% → 20%), validated against IFS 2011 distributional analysis. **Pipeline:** 1. A **supervisor agent** reads the policy document and all four persona profiles, producing personalised briefings anchored in each archetype's specific income, spend, and household structure. 2. Four **archetype agents** run in parallel, each reasoning in first person about what the policy means to their actual household budget. Extended thinking is