Privacy-Preserving AI Grader

Deidentify LMS PDF submissions, grade them against a question repository, and reconnect results locally.

Identity boundary: the identity map and reidentification happen locally. Only anonymous filenames are sent to the API.

Prototype limitation: this version renames files only. It does not remove names inside PDFs or PDF metadata. Inspect submissions before using real student records.

1. Select grading materials

Submission ZIP
Choose or drop an LMS ZIPQuestion repository
Choose the reviewed TXT answer key

2. Configure identity detection

LMS filename styleAuto-detectCanvas: Last_First_12345_...Blackboard: assignment_username_attempt_...Identity before first dash/underscoreParent folder contains identity

Anonymous ID prefix

Analyze filesClear everything

3. Review identity mapping

Files sharing the same identity key receive the same anonymous ID. Correct a detected key before grading if needed.

0PDFs0students0repository questions

#	Original path	Detected identity	Anonymous ID	Anonymous filename	Status

Apply edited identitiesDownload anonymous ZIP

4. Grade with OpenAI

API-key caution: use this only as a local personal prototype. The key stays in page memory and is not saved, but browser-based key entry is not appropriate for a hosted or shared application.

OpenAI API key

Model

Additional grading instructions

Grade all submissionsStop

5. Download reidentified grades

0graded0exceptions

Student identity	Anonymous ID	Score	Possible	Feedback	Confidence	Exception

Download grade workbook

Close or clear the page to remove the API key and working data from browser memory.