PropSim

• 

  Simulate public opinion before the campaign is run.

• 

  Agent beliefs update in real time as the simulation progresses.

• 

  After simulating real events leading up to Proposition 33, the agent swarm accurately represented population sentiment.

Inspiration

A single political event can shift millions of opinions. PropSim lets you see that shift before it happens.

What it does

• PropSim is a real-time multi-agent simulation for predicting population response to political events
• Builds a swarm of AI agents matching target demographics and simulates how events spread, debates form, and sentiment evolves
• Captures network effects like virality, polarization, and influence

How we built it

• Modeled agents as demographic personas with beliefs and incentives
• Connected agents through a sociological network graph and simulated interactions, persuasion, and sentiment updates over time
• Backtested using real-world data from California’s Prop 33

Challenges we ran into

• Making the system behave like a network, not isolated LLM calls
• Designing realistic agent interactions and influence dynamics
• Generating meaningful emergent behavior

Accomplishments that we're proud of

• Successfully simulated the collapse of Prop 33
• Captured real-world narrative shifts from a single shock event
• Demonstrated power of networked agents vs standalone LLMs

What we learned

• Network structure matters more than individual agent intelligence
• Social influence drives opinion more than isolated reasoning
• Emergent behavior is key to realistic simulations

What's next for PropSim

• Expand to future product launches, branding, and policy testing
• Improve realism of social graphs and agent diversity

Built With

• html/css/js
• python