Diagrammatic dataset on AI-generated formative feedback for XML-based UML models

This dataset describes a diagrammatic, XML-based corpus of student-generated UML models and corresponding AI-generated formative feedback created within a Software Engineering course at Constantine the Philosopher University in Nitra. The released dataset corresponds to one course implementation in the academic year 2024/2025, with data collected and exported in the summer semester of 2025. The release combines 112 anonymized student-level records (student_data25.csv), 448 Slovak-language formative feedback records (feedback_data25.csv) and 700 raw XML reports produced in Enterprise Architect (v16). The XML files encode structural model information such as packages, elements, connectors, stereotypes, attributes, and operations, which were used as the primary machine-readable input for automated evaluation. AI-generated feedback was produced in Slovak language through an OpenAI GPT-4-Turbo (gpt-4-0125-preview) API endpoint available during spring 2025 with temperature 0.3, max_tokens = 700. This linguistic setting provides valuable insights into the performance and adaptability of large language models in non-English educational environments, particularly in technically oriented disciplines. Each record in the dataset links XML-encoded UML models, Slovak-language formative feedback, numeric scores, letter grades, self-reported perceived helpfulness ratings, and corresponding human evaluation. The repository therefore supports research on automated formative assessment, prompt engineering for structured models, human-AI feedback comparison, and multilingual feedback analysis in education. By publishing the dataset structure, prompt templates, evaluation rubric logic, and supporting metadata, the release provides a reproducible basis for future benchmarking, exploratory analysis, and methodological extension in software engineering education.