Abstract:

Ongoing technological change is reshaping educational and professional demands. At the same time, the growing diversity of learner profiles is increasing the need for formative assessment approaches that can detect learning gaps early and inform targeted pedagogical action. In hands-on engineering courses, this challenge is no longer only how to collect answers, but how to verify whether students can justify technical claims with evidence. This challenge is sharper in AI-rich settings, where Generative AI can help learners produce fluent explanations that appear convincing even when understanding remains incomplete. This paper presents a work-in-progress assessment instrument for PSI3541 (Distributed Embedded Systems) that addresses this problem through artifact-grounded GenAI verification follow-ups for block-level formative micro-assessment.

At the end of each learning block, students submit compact technical artifacts together with a short explanatory narrative describing what they measured, how they measured it, the results obtained, and the conclusion they reached. Rather than accepting this narrative at face value, the proposed workflow interprets the submission against explicit block objectives, expected evidence, misconception patterns, and rubric criteria; formulates a diagnostic hypothesis; asks a brief verification follow-up only when relevant uncertainty remains; and terminates under an explicit stop rule with a teacher-facing progression recommendation.

The paper makes three practical contributions. First, it presents a reusable micro-assessment workflow in which learning objectives, expected evidence, rubric criteria, and verification probes are explicitly connected. Second, it operationalizes a feasible questioning policy centered on evidence-seeking follow-ups rather than open tutoring. Third, it provides a concrete Week 1 instantiation in PSI3541 focused on local period/jitter and WAN RTT/telemetry, showing how authentic student artifacts can support low-overhead but defensible formative assessment.

As a design-stage study, the preliminary evaluation was conducted under controlled conditions on a curated synthetic response set rather than on classroom deployment data. In a first controlled run over twelve synthetic cases, the workflow closed all cases within the predefined maximum number of turns, identified the intended primary gap and matched the expected external classification and progression recommendation in eleven cases, and avoided false mastery in all five high-risk fluent-but-unsupported cases. These findings do not demonstrate classroom effectiveness, but they provide preliminary evidence that the proposed instrument is diagnostically plausible, operationally interpretable, and suitable for subsequent expert review, consistency auditing, and controlled pilot deployment.

Keywords:

Formative assessment, Generative AI, engineering education, micro-assessment, verification follow-ups, embedded systems, rubric-based assessment, learning evidence.