Excellence built an AI tutor that won't give students the answer
Three Stanford-trained builders think the same technology teachers fear - generative AI - can be turned into the thing classrooms have always been short on: a patient tutor. Excellence, out of Y Combinator's W25 batch, is their attempt.
Ask any high-school teacher what changed in the last two years and you will hear a version of the same sentence: the students are using AI, and it is not helping. Essays arrive fully formed. Homework is done, and understood by no one. The tool that was supposed to be a tutor became, in practice, an answer vending machine.
Excellence, a startup that went through Y Combinator's Winter 2025 batch, was built on the opposite premise. Its product is an AI tutor for maths and science - but the defining design choice is what it refuses to do. It does not hand over the solution. It asks the next question. A student stuck on a quadratic gets a nudge, then another, then a diagram, until the student is the one who arrives at the answer. The company's phrase for this, borrowed from its French-language site, is about reclaiming control over AI rather than surrendering to it.
01 / WHAT IT ISA tutor priced like software
The pitch starts with a gap that has sat unaddressed for decades. A good private tutor costs somewhere around $60 an hour and does not scale. Software costs almost nothing to copy and scales infinitely - but until recently, software could not actually teach. It could drill, quiz, and grade. It could not sit with a confused twelve-year-old and figure out why she was confused. Large language models, Excellence argues, finally close that distance. That is the whole bet: the technology crossed a line, and the tutoring gap is now closeable.
Excellence began by aiming squarely at the incumbent every American parent knows - the after-school worksheet programs like Kumon, and the private tutors that sit above them. The company pegs that market at more than $30 billion. Its web and iPad software adapts courses and exercises to a student's knowledge and learning style, delivering explanations in text, voice, and graphics, plus study plans, revision tools, and human mentorship when a student needs a person rather than a model.
02 / THE DIFFERENCEProgressive hints, not answers
Plenty of apps will solve a maths problem from a photo. That is the category Excellence is defining itself against. The mechanism it built instead is closer to how a good human tutor behaves: withhold the answer, expose the reasoning, correct the specific mistake. The platform adjusts exercises based on the errors a student actually makes, not a generic difficulty dial. A student who keeps dropping negative signs gets problems that surface that habit until it is gone.
Student attempts
The learner works the problem; the AI watches the reasoning, not just the final number.
Progressive hint
Instead of the answer, it asks a guiding question or shows a graphic.
Error-based adjust
Exercises reshape around the exact mistake the student keeps making.
Teacher sees it
A dashboard tracks mastery so an educator can step in when it matters.
That last step points at the second thing that sets Excellence apart. It was designed with teachers in the room, not as an afterthought. The current platform lets educators generate personalized homework, build exams, assemble interactive materials, and follow student progress in granular detail. That is partly a product philosophy and partly a distribution strategy - in edtech, the hard part has never been the model, it is getting a teacher to trust it.
Figures approximate, for illustration of the pricing gap Excellence is chasing.
03 / THE PEOPLEPhysics champions who chose the classroom
The founding team's resumes read like they were built for a trading floor, which makes the choice to build a tutor more interesting. Mattia Mauro, the CEO, is a former quant who led early-stage investing at Squarepoint Capital and won the International Physicists' Tournament in 2022. Sovann Linden, the CPO, started a mathematics degree at Oxford at 15, tutored for over a decade, and worked in finance and research. Alexis Bonnafont, the CTO, studied physics and computer science at Ecole Polytechnique before a research master's at Stanford.
Their European roots show up in the go-to-market. Alongside US parents, Excellence has drawn early institutional interest from French schools including Ecole Polytechnique and College Stanislas - the kind of slow, one-teacher-at-a-time adoption that is unglamorous and very hard for a competitor to copy.
04 / WHAT YOU GETThe product, in parts
- AI TutorGuided maths and science tutor that asks progressive questions and explains in text, voice, and graphics.
- Assignment GeneratorPersonalized homework and exercises with error-based adjustment and progressive hints.
- Exam CreationCustom test and digital-exam design aligned to a teacher's curriculum.
- Interactive MaterialsInteractive PDFs, quizzes, flashcards, plus study plans and memorization tools.
- Progress DashboardLearning analytics that let teachers track mastery student by student.
05 / THE MODELTwo customers at once
Excellence runs a hybrid model. On one side, direct-to-consumer subscriptions for parents and students - the segment that would otherwise pay for private tutoring or a Kumon franchise. On the other, a B2B2C motion selling AI tutoring tools to schools and institutions. The seed stage is modest by design: public data ties roughly $130,000 in early funding to the YC round, dated March 2025. This is an asymmetric bet - a tiny team against a 60-year-old incumbent - that only makes sense if you believe the underlying technology just changed.
06 / THE MARKETBetting the line has been crossed
Excellence lands in a crowded field - Khan Academy's Khanmigo, Photomath, Squirrel AI, Mathnasium, and every parent's default, a ChatGPT tab left open on the family laptop. Its wager is that the winning product in AI tutoring will not be the one that answers fastest, but the one teachers and parents trust to actually build understanding. Whether that belief pays off is unproven. The number the company keeps pointing at - one to two grade levels of improvement - is the one worth watching, because it is the only one that matters to a parent deciding whether the software worked.
For now, Excellence is a small team with an unusually clear opinion about what AI in education should refuse to do. In a category racing to answer everything, choosing to withhold the answer is at least a distinctive place to start.