Excellence Learning Is Building an AI Tutor That Won't Just Give You the Answer

The Macro: AI in Education Is Happening Whether Schools Like It or Not

Every teacher I have talked to in the past year has the same problem. Students use ChatGPT to do their homework. Not to learn. To get the answer, paste it in, and move on. The tool that was supposed to democratize knowledge is being used as the world’s most sophisticated copy machine.

This is not a technology problem. It is a design problem. ChatGPT was not built for education. It was built to be helpful, and in education, being immediately helpful is often the opposite of being useful. A good tutor does not give you the answer. A good tutor asks you questions until you figure it out yourself. That is pedagogically sound and also deeply annoying, which is why it works.

The AI tutoring space is growing fast. Khan Academy launched Khanmigo with support from OpenAI. Photomath (now owned by Google) does instant math solutions from photos. Socratic (also Google) answers homework questions. Mathway, Symbolab, Brainly. The list is long. But most of these tools are answer engines with educational branding. They solve the problem for you and call it learning.

The market opportunity for something genuinely different is real. Parents spend roughly $12 billion a year on tutoring in the US alone. Kumon has 25,000 centers worldwide. Sylvan Learning charges $40 to $100 per hour. If an AI tutor can deliver even half the pedagogical value at a fraction of the cost, the demand is enormous.

The Micro: Oxford Math Meets Stanford Engineering

Excellence Learning was founded by Sovann Linden, Alexis Bonnafont, and Mattia Mauro. Sovann has a math masters from Oxford and an engineering masters from Stanford, plus over a decade of tutoring experience. Alexis studied physics and computer science at Ecole Polytechnique before doing energy engineering research at Stanford. Mattia is an ex-quant and former physics competition champion. They are a four-person team out of Stanford, backed by Y Combinator (Winter 2025 batch).

The credentials matter here more than they would for most startups. If you are building a sales tool, I do not care where you went to school. If you are building an AI math tutor, I care a lot that your founder has ten years of actual tutoring experience and an Oxford math degree. Sovann has sat across from struggling students thousands of times. That informs product design in ways that a pure engineering team cannot replicate.

The product takes a specific pedagogical stance. When a student gets stuck, Excellence does not show the solution. It gives progressive hints. It asks guiding questions. It makes the student work through the logic. The platform emphasizes that handwritten work is essential for reasoning, which is a position backed by cognitive science research and almost never reflected in edtech products.

For teachers, Excellence generates homework, creates exams, builds interactive course materials like flashcards and quizzes, and tracks student progress. The B2B angle is smart. Selling directly to parents is expensive and slow. Selling to schools gives you distribution and recurring contracts.

The product is live and serving schools, with a demo available on the website. They are running a French-language version as well, targeting the European market where Kumon-style programs are less entrenched.

The Verdict

I think Excellence Learning has the right philosophy. The “guide, don’t answer” approach is how good tutoring actually works. Most AI education tools are built by engineers who think learning is about information transfer. Excellence is built by people who have done the actual teaching and understand that learning is about struggle.

The competitive concern is Khan Academy’s Khanmigo, which has OpenAI’s backing and Khan Academy’s massive existing user base. That is a real threat. But Khanmigo is a chatbot bolted onto existing Khan Academy content. It does not have the adaptive exercise generation or the strong stance against giving answers that Excellence has.

The bigger risk is the B2B sales cycle. Schools move slowly. Budget approvals take months. Pilot programs take a semester. This is a team of four trying to sell into institutions that take six months to buy a new textbook. Patience and runway will matter as much as product quality.

Thirty days, I want to see how many schools are running pilots. Sixty days, the question is whether students actually engage with the progressive hint system or find it frustrating enough to switch back to ChatGPT on the side. Ninety days, I want measurable learning outcomes. Test scores, completion rates, time-on-task. If the pedagogical approach works, the numbers will show it. If it does not, no amount of good intentions will save the product.