Generative AI-Enabled Formative Assessment in a Higher Vocational Surveying Programming Course: Design and Initial Practice

Abstract

Programming courses in higher vocational geomatics education ask students to combine Python syntax with surveying computation fundamentals, data-processing workflows, and engineering-quality standards. In many cases, assessment of traditional assignments is both delayed and oriented toward final results, which makes it hard to pinpoint issues such as algorithmic mistakes, code-quality shortcomings, and each learner’s specific needs. This study proposes a generative AI–enabled formative assessment model for the Surveying Programming (Python) course. Using a platform that was developed in-house, the model integrates rule-based checks, test cases, domain-specific rubrics, large-language-model analysis, and teacher review to create a closed loop that supports task design, code submission, intelligent evaluation, personalized feedback, revision, learning analytics, and ongoing instructional improvement. The model was first piloted in the spring 2026 term in Engineering Surveying Class 2401 (n = 42). Data aggregated from five assignments were analyzed descriptively. The platform dashboard showed a 95% assignment completion rate, an overall average score of 84.6, and four students who required additional attention. Mean assignment scores rose from 78.1 on the first task to 88.9 on the fifth task. The most frequently recorded problems were file-reading exceptions (28%), non-standard variable naming (22%), and loop-boundary errors (18%). Overall, these results offer early evidence that the model is feasible for providing timely, domain-informed feedback and for turning programming-process data into actionable teaching information at the class level. Because this study involves only one class and relies on descriptive platform data without a control group, the findings should be viewed as implementation evidence rather than proof, in a causal sense, of improved learning effectiveness.