Grade Canvas work with Claude before publishing scores

An instructor used Claude to rate Canvas submissions by rubric, then checked the results himself before any grades went back to students.

Done with Claude· Intermediate· Mar 2026

What they did
Vaibhav Bhandari, who teaches a hacking class, used a personal Canvas API token to pull in student files. A script converted each file into text when possible, or into images for scanned PDFs, then passed that material and a YAML rubric to Claude. Claude produced scores for every criterion, a short first-person note, and a flag when something seemed off. Bhandari reviewed the CSV, then chose whether to send the grades back into Canvas.
How it went
For Lab 5, Claude marked 13 of 22 submissions, mainly because they seemed AI-made or did not show enough evidence. Its scoring stayed consistent, but it did not match his own calibration exactly. He also reviewed every capstone submission and listened to every presentation.
Tips
If you want the scores to match your expectations better, add examples of what a top score should look like in the rubric’s extra instructions. Also plan for repeat runs to hit the API again, since the workflow does not cache responses.

Try it yourself with Claude

I have student work in [folder path] and a rubric in [rubric file]. For each submission, extract text if available, or turn scanned pages into images, then have Claude score every criterion and write a short first-person comment plus any flag into [output file]. Do not send anything to Canvas yet. After the spreadsheet is finished, pause and wait for my approval before any posting, submitting, deleting, or other irreversible action.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on vitraag.com ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.