SAGE: Smart Automated Grader for Essay

SAGE is a personal project I built to help lecturers at Politeknik Negeri Lhokseumawe grade students' essay exams more easily, consistently, and in line with their rubrics.

Essays are assessed by a fine-tuned Qwen3 (8B) Large Language Model. The model was fine-tuned with LoRA on a custom dataset of essay questions, reference answers, and lecturer-defined grading rubrics to produce more consistent, rubric-aligned evaluations.

Every answer receives a score along with an AI note explaining the reasoning, and the results are compiled into a PDF exam report.

The backend is built with Django REST Framework and the frontend with Vue.js. AI inference runs locally through Ollama, while Celery and Redis handle essay evaluation asynchronously.

The entire system is containerized with Docker Compose and uses PostgreSQL for persistent data storage.

The system monitors student activity throughout the exam. The exam sheet is locked in fullscreen mode, switching tabs or browser windows is detected, and copy and paste are disabled on the essay sheet.

Unauthorized actions, such as leaving the exam page or trying to copy and paste exam content, are automatically detected and recorded as violations. The student's account is restricted immediately to preserve exam integrity, and lecturers can review it in the violation log.