Pyxiscience offers an advanced educational platform based on artificial intelligence. Fixed handwritten copies, suggestion of exercises … There are many features.
What if AI made it possible to facilitate the learning of mathematics? This is the initial objective of Pyxiscience, a young French start -up launched in November 2024. Pyxiscience offers a complete educational platform for students from Bac +3 to BAC -3 (second, first, terminal). Before he even celebrated her first anniversary, the start-up announces a raising of 2 million euros from Newfund, Invess Ile-de-France and BPI France.
A platform based on artificial intelligence
Pyxiscience was designed to meet two major challenges of education: the heterogeneity of levels and the lack of time of teachers. “All teachers know it: to advance students, you have to assign homework in a frequent way and correct them in detail. But with often several hundred students and in a context of teachers shortage, it’s impossible,” explains Joachim Lebovits, CEO. The AI used by the platform therefore comes to assist teachers by offering them innovative tools: rapid creation of educational content, allocation of exercises and above all automated correction of the work.
Teachers can use the content already produced and validated by Pyxiscience or use their own support. The AI used by Pyxiscience will then suggest possible exercises. Technology allows you to generate an unlimited number of exercises that adapt in real time to the level of each student.
The automated correction of handwritten copies
But the most advanced feature remains the automatic handwritten copy correction. “Despite the progress of AI, no solution has so far allowed to recognize the handwriting of advanced mathematics and to follow a reasoning step by step to verify its validity”, according to Jacques Lévy Véhel, co-founder and director of R&D. Pyxiscience took up the challenge with a precision rate greater than 90% when the majority of OCRs on the market fail to exceed 70%. “Mathematical symbols, and handwritten copies in particular, are hell for recognition systems,” smiles Jacques Lévy Véhel.
To achieve such precision, pyxiscience technology works in two stages. The first, the most complex, recognition of handwriting with a multitude of LLM of vision, updated regularly. And the second, the contextual analysis of mathematical reasoning. “Our AI follows the student’s reasoning, identifies the errors or the consequences of these errors, and is capable of noting the copy,” says the co-founder. AI is not content to check a unique solution, but includes the diversity of mathematical approaches, a fundamental approach in higher education where a problem can have several resolution methods.
An economic model adapted to Europe and the United States
For the time being, Pyxiscience focuses on mathematics but is gradually opening up to physics. “The goal is to gradually cover all of the sciences,” says Joachim Lebovits. The platform mainly targets secondary and higher students, from the second to the first university cycles. The solution has already been adopted by more than 10,000 students, notably from Sorbonne Paris Nord University and the New York University in Paris.
Pyxiscience, which aims for a deployment in France but also in North America, has developed two distinct economic models. In France, it is the institutions that subscribe to licenses for their students and teachers, whether public universities or private schools. “In the United States, our model is different. Students start by paying individually until the platform reaches a critical size. Then, it is the university which takes over and redoing part of the costs to students,” explains Joachim Lebovits.
With its lifting, Pyxiscience intends to accelerate its growth. The objective is clear: go from 4 to 10 employees and prepare international expansion. “We want to invest in the North American market,” said Joachim Lebovits, with deployments scheduled for January 2026 in American universities, after consolidating its positioning in France.
