Generative AI for Project-Based Assessment in Energy and Sustainability Education

Teaching energy transition and sustainability to interdisciplinary students requires learners to integrate technical concepts of energy systems with energy economics and link sustainability with real-world decision-making. This is challenging in short courses where building domain knowledge may be more important than mastery in techno-economic analysis. We report a design-based evaluation of a custom, dialogic GPT deployed as an interlocutor in a project-based assessment on digital energy systems. The AI assistant was configured to elicit objectives and data needs, probe domain-specific topics, and assist students in economic calculations and policy reasoning. We studied two graduate cohorts with different modalities: an in-class exam (all questions visible at once) and a remote, stepwise exam (task revealed incrementally). Methods combined rubric-based scoring of conversational and technical performance, analysis of protocol fidelity, and an anonymized post-exam survey. Results show that when the dialogue followed the intended sequence (with a score of 0.88/1), the AI assistant supported coherent, stepwise reasoning (0.78/1) and defensible, context-sensitive calculations (0.86/1). However, mandatory checkpoints (e.g., domain-specific questions on renewable energy variability, demand flexibility, and energy prices) and visual cues were triggered inconsistently in depth, and question verbosity varied, which created fairness and consistency concerns. We contribute a transferable assessment pattern for AI in higher education: instruction-based coverage checkpoints, single-question turns, timed multimodal prompts, and transparent rubrics, along with practical implementation guidance (e.g., LMS embedding to stabilize access and streamline grading). The work illustrates how dialogic large language models (LLMs) can mediate for authentic, domain-specific assessment while identifying controller-like safeguards needed to ensure equity, reliability, and disciplinary rigor.