Dialoguing with Hattie&TimperlAI: a Research Based AI-Feedback Coach for Math Teacher Training

This paper presents Hattie&TimperlAI (h&tai), a research-based AI agent that acts as a feedback coach for mathematics teacher education grounded in the Hattie and Timperley’s model. Building on a previous version (tp-h&tai), which generates examples of effective feedback for classroom use, we introduce tt-h&tai, a new configuration designed to help teachers refine their own feedback during training. tt-h&tai was implemented on three LLM platforms (ChatGPT, Gemini, and Claude) and evaluated through controlled dialogue simulations based on a standardized mathematics task. A qualitative content analysis examined how each implementation supported teachers’ reflection on feedback practices. Findings show that all platforms fostered the use of the Hattie and Timperleys framework for analyzing feedback, with Claude additionally supporting the construction and reformulation of feedback. The paper discusses the differences in relation to prompt configuration and proposes a replicable, research-grounded workflow for developing AI tools for teacher education.