Higher ed, meet co-teacher mode

Home to brilliant scientists, the University of Maryland is a nationally recognized public research university with a strong reputation in STEM. Its Department of Mathematics ranks #21 in the country for its graduate program, with its applied mathematics program in the top 15. The perfect equation to foster innovation in teaching math.

Sean Gruber, a lecturer and researcher in mathematics, is interested not only in mathematics itself but in how it’s taught. His lectures have earned him a Dean’s Award. And few things say more about a teacher than students on Reddit calling him “the best math professor” on campus.

His research on course redesign reflects an interest in rethinking teaching practices. So it's no wonder he was interested in a new AI grading tool that could free up time for other parts of the job.

Sean works with a team of teaching assistants across several courses, and Precalculus is by far the largest.

Every week, that adds up to hundreds of handwritten pages of quizzes and exams. An ocean of papers. When Sean found Ed.ai, he knew what he wanted to hand off:

And his first date with us was a big one: 330 pages all at once.

The math lecturer needed a tool that could handle an enormous batch of scanned exams and fit the way his team shared the grading. Volume was clearly going to be the real test, and we took it on with a free pilot.

Sean and his five teaching assistants began a free pilot covering 252 Precalculus students across eight lecture sections, and 13 students in History of Mathematics.

The first point of friction came at import. At UMD, quizzes and exams are scanned in batches, so each batch has to be split into individual papers and matched to the right student before anyone can grade. At the start of the pilot, our team had to do it... by hand. We gradually built a way to automate the task, so no one would have to carry a stack from one TA to the next.

But our biggest challenge, and perhaps where the most interesting feature was developed, was making the tool fit how the team already worked. Collaboratively. Like most university course teams, Sean’s TAs split the work by question: each TA takes one problem and grades it on every paper. To make that possible on the platform, we built co-teacher access.

Finally, this was also Ed.ai’s first semester in higher education, so a lot of effort went into adapting the platform to Sean’s standards.

After plenty of back-and-forth, rubrics became much more flexible. Sean can give the tool guidelines as specific as he likes, with or without an answer key, and it writes a first draft of the rubric. If one question needs different criteria, the team can edit the rubric and rerun grading without restarting the whole assessment.

Interestingly, Sean was considerate of his students and explained how the tool does a first pass, while the instructional staff review every paper before releasing the results. He also recorded a demo for his classes showing them how to read their feedback.

In the first three weeks of the pilot, Sean and his team graded 772 papers for 265 students. That's more than 2,000 pages.

Import caught up. Since mid-October, splitting the batches has been automatic.

The team could then grade the same assessment simultaneously, each TA working on their assigned questions. With AI grading taking a median of six minutes per paper, the assessment was ready for the TAs to start reviewing about an hour after batch import.

When a TA decided a student's method deserved partial credit, they could revise the rubric and rerun grading across all 231 papers at once. The team could adapt its criteria without having to start over.

The next step for the team is Canvas integration. But the most interesting part may be what Sean plans to do next.

The time saved on repetitive grading can go toward helping students work through what they haven't understood.