2026 ASEE Annual Conference & Exposition

Large Language Models (LLMs) are increasingly used in engineering and computing education, most often through single-agent conversational interfaces that encourage rapid convergence. Such approaches offer limited support for deliberation in ill-structured problems that involve competing constraints and perspectives. This study investigates the classroom use of a multi-perspective chatbot interaction design in which students engage sequentially with multiple role-specialized AI chatbots. The activity involved three independent chatbots representing technical, institutional, and community perspectives. Each chatbot was configured with a distinct professional perspective and communication style, including limited variation in tone and assertiveness. Students interacted with the chatbots sequentially during an in-class exercise focused on two realistic infrastructure decision scenarios. Interaction order and persona tone were varied across teams to examine how these design choices influenced learner–AI dialogue. Data sources included pre-activity problem framing, chatbot interaction transcripts, in-class artifacts, post-activity questionnaires, and written reflections from twenty-four students working in twelve teams. Analysis focused on interactional patterns rather than decision outcomes or learning gains. Results indicate that engaging multiple role-specialized chatbots supported diverse deliberation trajectories, including convergence, confirmation through deliberation, and comparison across perspectives. Interaction order influenced when constraints entered reasoning, while persona tone shaped how explicitly students articulated trade-offs and justifications. Students assumed responsibility for synthesizing perspectives across chatbots, highlighting the role of learner-mediated coordination. The study contributes design-relevant insights into how role definition, interaction sequencing, and persona characteristics shape deliberative dialogue and learner–AI interaction in engineering and computing education. Findings provide practical guidance for instructors seeking to use multi-chatbot configurations to support perspective-taking and justification without complex system integration.