Ed.ai, a French company building artificial intelligence to grade handwritten math homework, has struck a deal with Edlink, the education interoperability platform announced in late July. The integration opens a path to learning management systems and student information databases across thousands of U.S. K–12 districts.
Whether that translates to actual classroom adoption remains to be seen. The company claims more than 5,000 math teachers use its platform, though that figure is uncorroborated by independent sources. Ed.ai runs its American operations through a Boston-based entity, Ed AI Technologies, Inc., and routes student data through U.S.-hosted servers, according to privacy disclosures on its site.
A Toehold in American Classrooms
So far, the company's disclosed footprint is modest. Two school districts appear on Ed.ai's website: Caldwell County Schools in North Carolina and Schuylerville Central School District in New York. The company hasn't said when these partnerships began or how many students are involved. CEO Jonathan Banon has mentioned expansion in Colorado and Hawaii in social media posts over the summer, based on updates visible on his LinkedIn profile.
Ed.ai's grading engine is designed to decipher handwritten work in middle and high school math, awarding partial credit for multi-step reasoning and aligning scores to Common Core and Texas state standards. The platform claims 99 percent accuracy on handwriting recognition and promises to grade assignments 50 percent faster than a teacher working by hand, according to marketing materials aimed at district administrators.
Under the hood, the system farms out grading tasks to a suite of large language models: Azure OpenAI, Azure Claude, Azure Mistral, and Google Gemini. The company says it picks models based on the type of problem being graded, a detail buried in its AI transparency documentation. Ed.ai insists it doesn't train foundation models on student work and de-identifies data before processing.
The Team Behind the Tool
Banon previously co-founded Lelivrescolaire.fr, a digital textbook company in France. His co-founders include Cédric Bignon, a CentraleSupélec graduate now serving as CTO, and Rémi Mazières, who holds the title of chief learning officer. The trio raised €1.7 million in a seed round announced in April 2024, bringing in backers like Ring Capital (through its Generations fund), AFI Ventures, CentraleSupélec Venture, Kima Ventures, Evolem, and Super Capital. Third-party data suggests the company employs between 11 and 50 people, though headcount estimates for startups at this stage are notoriously slippery.
Crowded Territory
Ed.ai is wading into a competitive field. At least five other venture-backed startups are pursuing AI grading for handwritten math. Pensive raised $6.8 million recently for higher education use cases. YC-backed Frizzle and EdLight, which reports around $4 million in total seed funding, are chasing similar K–12 workflows. Fermi and PaperGrade market comparable products, and MathGPT.ai announced its own handwritten grading tool for colleges in a July release.
Academic researchers have started poking at the underlying question: Can AI really grade math reliably? Studies circulating in preprint suggest multimodal large language models can approach human-level agreement on rubric items in controlled settings, though handwriting styles still trip up transcription accuracy in unpredictable ways.
For its part, Ed.ai claims that 82 percent of teachers who used the platform during the most recent school year plan to stick with it next year, though this retention figure lacks external validation. The Colorado Council of Teachers of Mathematics lists Ed.ai on its partners page, a modest but noteworthy endorsement in a market where teacher buy-in can make or break edtech products.
The real test will be whether districts beyond the handful named so far decide the technology is worth the investment, and whether parents and educators trust an algorithm to evaluate work that's still, for many students, painstakingly written out by hand.
