AI System for Comparing Exam Document Revisions

The problem

When question papers, answer keys and scoring information are held in different sources, staff must match fields manually before preparing a revision difference list.

The assessment specialist must compare every suggestion in the revision difference list with the source record; incomplete or conflicting evidence must not replace the final decision.

If a formatting change that appears to be a content difference is not set aside for separate review, the wrong answer key could be used; the record must not proceed without a human decision.

What we set out to improve

• Time taken for the draft revision difference list, with its supporting evidence, to reach the assessment specialist for review
• Outcomes of records sent for human review because a formatting change appears to be a content difference
• Number of suggestions corrected after identifying a risk that the wrong answer key could be used

How the system works

1. 01 Retrieve the question paper, answer key and scoring information from permitted sources
2. 02 Verify field and document versions
3. 03 Prepare the revision difference list with supporting evidence
4. 04 Check the exception: a formatting change appears to be a content difference
5. 05 Have the assessment specialist accept, correct or reject the suggestion
6. 06 Decision gate: exam content, score and grade
7. 07 Record the decision and its source separately

Methods we used

• A source-identified data schema for the question paper, answer key and scoring information
• A structure that separates deterministic rules from AI suggestions
• A screen showing the revision difference list alongside the supporting evidence
• A human review queue for formatting changes that appear to be content differences
• A record identifier that prevents duplicate processing

Where people stay involved

Data and security

Only the necessary fields from the question paper, answer key and scoring information should be processed. Role-based access should be applied, and unnecessary identity, health, financial and internal institutional data should not be sent to the model. Source access, the AI suggestion and the assessment specialist's decision should be recorded separately.

Who this suits

Frequently asked questions