Add misinformation spread model: LLM-powered agents vs rule-based com…

[Harsh-617]

Adds a new example model demonstrating mesa-llm features through a misinformation spread simulation.

What this model does

12 agents with unique personas (teacher, doctor, journalist, shopkeeper, etc.) are placed on a grid and react to a rumor about water contamination. LLM-powered agents use ReAct reasoning to decide whether to spread or challenge the rumor based on their personality and conversations with neighbors.

A rule-based comparison model with the same agents and setup uses simple mathematical rules instead of LLMs, allowing direct comparison of the two approaches.

Features demonstrated

• mesa-llm LLMAgent with ReAct reasoning
• Custom tools (@tool decorator): check_neighbors, spread_rumor, challenge_rumor, update_belief
• Inter-agent communication via send_message
• Memory system for tracking conversations
• DataCollector for tracking belief dynamics
• Rule-based vs LLM agent comparison

Files

• misinformation_spread/agents.py — CitizenAgent (LLM) and RuleBasedAgent
• misinformation_spread/model.py — MisinformationModel and RuleBasedModel
• misinformation_spread/tools.py — Custom agent tools
• run.py — Runs LLM simulation (requires Ollama)
• run_comparison.py — Runs rule-based simulation (no LLM needed)
• readme.md — Documentation and findings