FURI | Spring 2026

Expanding LLM-Assisted Translation of Natural Language to ML-Augmented SQL Queries with Applications in Database Education

This research evaluates large language models (LLMs) as automated graders for SQL assignments in university database courses. Using 139 student submissions, models including Gemini, GPT-4.1-mini, GPT-o4-mini, and Claude are compared against human graders. Through prompt engineering techniques such as rubric integration, few-shot examples, and strictness calibration, model accuracy improved significantly. An ensemble system combining multiple model outputs is being developed to further improve grading consistency and bring automated scores closer to human-level performance.