Is Math Giving You Problems? It May Not Be the Numbers

Students who struggle with math often think they’re just not good with numbers. But what if the problem is that they are actually stumbling over the words?

Recent master’s graduates Abdirahman Guleed and Kedaar Sridhar found a way to help struggling math students, especially students who are multilingual or from diverse backgrounds. While in the Ed School’s Learning Design, Innovation, and Technology Program this past year, they created an AI-based education platform called M7E. The platform uses AI to evaluate how clear the language is in K–12 math curriculum, applying seven linguistic criteria developed from 330 peer-reviewed studies. Basically, it flags certain words or phrases that have multiple meanings or could be unfamiliar to all students and then suggests revised ways to present the material. 

For example, in this problem, M7E identifies complex embedded clauses and a “high cognitive load” required to decipher the language:

“Lena has 3 bags, each containing twice as many marbles as her friend Jamal, who gave away 5 marbles to Priya before buying more.” 

M7E recommends this revision: 

"Jamal had some marbles. He gave away 5 marbles to Priya, then bought more. Lena has 3 bags, and each bag has twice as many marbles as Jamal has now. How many marbles could Lena have?"

In another example related to temperature, the word “plunged” is used — a figurative term that might be unfamiliar to students whose first language is not English. Simply changing “plunged” to “decreased” makes the example more concrete and precise, allowing students to focus on the math, not just the language. 

Guleed and Sridhar, current HGSE Education Entrepreneurship Fellows, say the idea for the platform started in a J-term class with Lecturer David Dockterman on developing educational products. During the class, students got to meet a textbook publisher who explained the gap between the math curricula their editorial teams created and the evolving demographics of students being served. As a result, math lessons and tests were often written, unknowingly, in ways that confused students or caused them to spend too much time trying to figure out what the questions were even asking. Teachers often didn’t know it was the wording — not the math itself — causing confusion. 

During J-term, Guleed and Sridhar started interviewing teachers, about 40 in total, as well as researchers from across the United States, to better understand the connection between language and math struggles. They also reached back to their own early difficulties with math. Guleed grew up in Norway to Somalian parents and had to turn to tutors for help. “Fitting into the society, understanding the language, and excelling at school required a lot,” he says, “and it required a big support system,” which he got from tutors.

Sridhar had a similar experience growing up in the Middle East after his parents emigrated to Oman from India. “I struggled with math initially because I went to an international school where it was mostly American and British curriculum. I was getting questions about lacrosse or snowboarding. Living in a very hot place, these were things that weren’t relevant to me personally and took away from my ability to pursue math in general.” The confusion was contextual, he says, and could have been avoided. 

“We started to understand that the reason why students are struggling is not because of their mathematical ability,” says Guleed, “but because of the way math is presented.” According to the latest National Assessment of Educational Progress (NAEP) scores released in September 2025, 45% of 12th-graders scored below the “basic” level of proficiency in math, five percentage points higher compared to both 2019 and 2005.

Initially, the publisher who came to the J-term class was reluctant to consider using AI directly with students, citing ethical and other concerns, so Guleed and Sridhar instead came up with an AI platform that works behind the scenes, “not as an automated tutor or chatbot, but as a professional support mechanism empowering human decision-makers across the ecosystem.” The goal of the app, they say, is not to take away complexity in math problems, but to make muddied wording visible and allow teachers to decide when to make the language simpler and when to “teach through complexity.”

They acknowledge that humans can certainly read through math curriculum and flag certain complex words and phrases and can do it well. But as Guleed points out, given the amount of work involved teachers take on every day and the variation in 50 different state standards, “AI is a tool that does things at a fast, fast pace. That is the thing that AI is good at.”

Currently, M7E is being piloted in five schools and Guleed and Kedaar are gearing up to partner with more schools and districts, while continuing to gather feedback from both teachers and curriculum designers. They also continue to read through more research, which is what initially inspired their platform’s name, M7E. 

“We basically collated more than 300 sources of research and distilled it into these seven criteria for linguistic comprehension and math,” says Kedaar, referring to the 7. “When we were presenting it to the publishers in class, Professor Dockterman said, ‘Oh, this is the magnificent seven. You should call it the magnificent seven criteria,’ which became magnificent seven education, or M7E for short.”