Research Summary: Driving Epidemic Models with AI Agents: the Epydemix Agent Framework

The Epydemix Agent Framework is introduced as an additive layer over the existing Epydemix Python library for stochastic compartmental epidemic modeling. This framework enhances the library by enabling AI agents, particularly those based on large language models, to manage the entire epidemic modeling process from natural language input to quantitative results. Key functionalities include discovery of models and parameters, preventive validation of scenario specifications, execution via tested library code, and inspectability of outcomes, aiming to improve the reliability of AI-driven scientific software interactions.

Why it matters

This development is crucial for integrating advanced AI capabilities into complex scientific and operational modeling, potentially accelerating analysis and decision-making by making sophisticated tools more accessible. By enabling AI agents to handle the entire modeling process reliably, it lowers the barrier to entry for complex simulations and enhances the efficiency of scientific inquiry and operational planning.

Key insights

• Artificial Intelligence (AI) agents, specifically those leveraging large language models, offer natural language interfaces for scientific software.
• The reliability of AI agents in interacting with scientific software is not automatically guaranteed.
• The Epydemix Agent Framework extends the open-source Epydemix library for epidemic modeling.
• The framework provides four core capabilities for AI agent interaction: discovery, validation, execution, and inspectability.
• It allows AI agents to manage the full modeling lifecycle, from scenario description to result generation, using natural language inputs.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2609.28692