Five-agent DPT simulated classroom on a static-friction problem

Design · All designs
Evidence · no claims cited

Description

A simulated class of five AI students, each embodying a distinct point of failure in the reasoning process: Davide (Hazards A+B), Elena (Hazard C), Tommaso (Hazard D), Chiara (in transition) and Andrea (a near-expert peer). It runs on an LLM via a Gradio application on Hugging Face Spaces, with personas defined entirely in the system prompt, session reset and one-click export of the anonymised conversation log. Named agents respond, with roughly 30% probability a second agent intervenes, and an open question draws two to four responses.

Design Implications

Context

Requirements

• Requires a valid API key from a supported LLM provider, supplied by the user; no specialised infrastructure or model training is needed.

Constraints

• The hazards are reified into stable characters — a deliberate pedagogical simplification the framework itself warns against.
• Participants noted the environment's limits as a proxy for real classrooms: "real students do not ask all those questions".

Target Learners

• prospective physics teachers in MSc programs

Learning Goals

• responding in real time to student reasoning with differentiated questioning strategies

Claims

Related Designs

• Practice-based clinical mastery through progressively challenging simulated encounters
• Dual-process reasoning hazards framework with hazard-matched questioning strategies

Examples

-

Key Sources

• Eugenio Tufino. (2026). A multi-agent AI classroom based on dual-process reasoning hazards: a pilot with prospective physics teachers. https://huggingface.co/spaces

Page metadata

Field	Value
type	design
id	dpt-five-agent-simulated-classroom
title	Five-agent DPT simulated classroom on a static-friction problem
description	A simulated class of five AI students, each embodying a distinct point of failure in the reasoning process: Davide (Hazards A+B), Elena (Hazard C), Tommaso (Hazard D), Chiara (in transition) and Andrea (a near-expert...
status	review
generated	by process:settle-candidates on 2026-10-11
trust	unverified — no one has confirmed this page's content
sources	1 — eugenio-tufino-2026

Frontmatter is the page's machine-readable record — see the Schema & Guide.