Skip to content
All posts

· 5 min read · Muhammad Shah

AI-assisted grading: instant, personal feedback without sidelining the teacher

In Banoqabil LMS at Alkhidmat I delivered AI-assisted grading that evaluates student assignments and returns instant, personalised feedback. Students no longer wait days to learn what went wrong, and teachers spend their time on the cases that need a human.

Start from the rubric, not the model

A model asked "grade this" invents its own standard. Give it the same rubric a teacher uses: criteria, what each score level looks like, and the marks per criterion. The rubric is the contract; the model just applies it.

Ask for structure, not an essay

Request a fixed shape back: a score and a short justification for each criterion, an overall score, and two or three concrete next steps for the student. Structured output is easy to validate, store and show, and it stops the model from drifting into vague praise.

{
  "criteria": [{ "name": "Accuracy", "score": 3, "max": 4, "why": "..." }],
  "total": 14,
  "nextSteps": ["Show the formula before substituting values"]
}

Calibrate with real, already-graded work

  • Collect answers teachers have already marked, across strong, average and weak work.
  • Run them through the grader and compare scores criterion by criterion.
  • Adjust the rubric wording and examples until the gaps are small and consistent.
  • Keep the temperature low so the same answer gets the same grade.

Keep the teacher in charge

AI grades should be a draft the teacher can see, change and approve, especially for exams. Flag low-confidence or borderline results for review, and log overrides: they are the best signal for improving the rubric.

Write feedback a student can act on

Good feedback names the specific mistake, explains why it matters and says what to try next, in the student's language and level. "Show the formula before substituting" helps; "needs improvement" doesn't.

Done this way, AI grading speeds up the feedback loop without taking judgement away from teachers. If you're adding assessment features to a learning product, I'm happy to compare notes.