Gradescope Raises $2.6M to Apply Artificial Intelligence to Grading Exams


Gradescope Raises $2.6M to Apply Artificial Intelligence to Grading Exams

By Blake Montgomery     Apr 18, 2016

Gradescope Raises $2.6M to Apply Artificial Intelligence to Grading Exams

Gradescope, which has graded millions of exam questions, has made the grade itself.

The company has raised a $2.6 million round of funding from Freestyle Capital, Bloomberg Beta, Reach Capital and the House Fund. Existing investor K9 Ventures also participated. Dave Samuel from Freestyle will be joining Manu Kumar from K9 Ventures on Gradescope’s board. The company, started as a side project at the University of California Berkeley in 2012, makes a software that helps science and engineering professors and teaching assistants grade exam questions on handwritten tests. Gradescope plans to use the funding to expand its team and develop new product features.

The tool works by allowing users to create rubrics for each exam question, allowing them to flag common mistakes with a click and standardizing the grading process. Users upload exams into the platform, administer grades and return the online grade report to students. As of now, the platform can grade questions with a single answer, though the rubrics do allow for partial credit.

The company has, so far, focused on higher education, but the problem it aims to solve is a universal one: grading demands an inordinate amount of time. According to the company, their users find that grading with Gradescope takes about half the time that grading on paper does. Cofounders Arjun Singh, Sergey Karayev, Ibrahim Awwal and Pieter Abbeel do not, however, see their tool as a replacement for professors; that’s why they named Gradescope’s function “assisted grading.”

“We want professors to see the answers and give the perfect feedback every time,” Singh said. “We want students to get the same quality feedback as they would if the professor were looking at their exam closely, but we want them to get it faster.”

The company claims that its program has graded 8 million questions at UC Berkeley, Stanford, MIT, the University of Michigan, Georgia Tech and others. All told, roughly 100 universities use Gradescope to grade 55,000 students. Singh said that a handful of those use Gradescope most heavily and that the company is looking to expand among the more casual users.

The company has passed its first test of raising funding for its product, but a bigger exams looms: creating a tool that can grade complex answers.

“It will be a few months until the artificial intelligence feature comes out,” Singh said. “We want a feature that will offer the convenience of a Scantron without watering down the questions. It will be a feature that uses machine learning to grade the short answer questions on exams.”

The company currently operates on a freemium model. Grading simple exam questions is free. The artificial intelligence feature will be available for a price.

The artificial intelligence to grade more complex questions is not only the company’s future; it is its raison d’etre. Singh, a former computer science teaching assistant at Berkeley, said he was initially uninterested in taking the company beyond the UC Berkeley computer science community, where it started.

He and his cofounders were teaching assistants for UC Berkeley’s biggest class: Introduction to Computer Science. There were 400 students in the class in 2012 (today there are 1400; 950 of their exams were graded with Gradescope). Grading complicated handwritten exams took Singh and other teaching assistants 200 man hours between them all. That labor led Singh and his cofounders to wonder why, since they were computer scientists, they had not created a program to automate the more onerous parts of grading. Thus, Gradescope was born.

This is not a tale of immediate startup explosion, though. Over the next two years, Singh used the tool once, letting others do with it what they would. By spring of 2014, the early version of Gradescope had saturated the computer science department, and the servers were crashing. The founders faced the choice of committing fully to Gradescope or shutting it down. Singh wasn’t enthused; he didn’t see how he could apply PhD work in machine learning and artificial intelligence to a grading tool.

“After talking with my cofounders and advisors, though,” he said, “I saw that there was valuable data being thrown away on paper. When it’s just a grade, both students and instructors lose data could significantly impact instruction and studying habits.”

That data may be Gradescope’s future, the linchpin that allows it to digest more open-ended questions.

Learn more about EdSurge operations, ethics and policies here. Learn more about EdSurge supporters here.

More from EdSurge

Get our email newsletterSign me up
Keep up to date with our email newsletterSign me up