This paper presents a prompting engineering
implemented within a web-based application prototype that leverages GPT-4o to
automate the generation of Indonesian high school mathematics questions. In
Indonesia, teachers are required to prepare a wide range of question types, including
daily exercises, midterm tests, and final examinations, that align with
curriculum standards while varying in difficulty and format (e.g.,
multiple-choice, essays, and open-ended problems). This manual process is
time-consuming, inconsistent, and especially difficult to sustain in
under-resourced or remote areas. The proposed system assists teachers in
efficiently producing exam materials for daily practice, midterms, and finals,
supporting multiple formats while maintaining consistent difficulty levels in
accordance with the national curriculum. Implementation in high schools,
involving several mathematics teachers and 104 students, demonstrates that the
application substantially accelerates question creation, ensures strong
curriculum alignment, achieves high accuracy, reduces errors, and produces
questions that are clear and easily understood by students. Performance
evaluation further shows that the system achieves an average response time of
3.65 seconds per question under a simulated concurrent load of 100 requests, confirming
its suitability for near real-time educational use.
|