AI News

Researchers introduce the Epydemix Agent Framework, an additive layer over the open-source Epydemix Python library for stochastic compartmental epidemic modeling. It is designed to make large-language-model agents more reliable when interacting with scientific software. The framework lets an agent discover available models and parameters, validate a declarative scenario specification before execution, run tested library code, and inspect the resulting outputs. The workflow can take a natural-language scenario through quantitative results, figures, and interpretation without requiring the agent to write custom code. Each step reads input files and stores its results in a separate output bundle, making the process auditable and reproducible. A case study applies the workflow to comparing vaccination strategies for a novel respiratory virus. In an evaluation covering 50 agent sessions and five modeling tasks, researchers compared framework use with direct use of the Python interface. The framework reduced the number of turns, output tokens, and cost on most tasks, although it could trade additional resources for reproducibility at the level of individual points.