Abstract

EE-Eval is an automated multi-agent pipeline for evaluating AI-generated interactive learning materials. Existing evaluation benchmarks focus on static properties such as code correctness or visual quality, but fail to assess interaction quality, or whether an interactive system actually supports meaningful learning.

EE-Eval addresses this gap by modeling interactive learning materials as a new data structure based on finite state machines. Through a full-stack pipeline, it enables systematic testing between transition states to assess whether the interaction flow aligns with pedagogical goals.

This project is a continuation of research from Interactive Neural Networks.