Tatami

The problem

For a high-level athlete in high school, working more is not an option. When training ends at 9:30 PM and there's a math test tomorrow, generic exercise sheets and private tutors don't fit the agenda. What's needed is a tool that understands exactly where the gaps are and doesn't waste a single second.

The solution

Tatami is an adaptive math coach that does for mathematics what a good sensei does on the mat: diagnose, target, adapt. It reads handwritten work through computer vision, identifies recurring errors through LLM analysis, and generates targeted exercises at exactly the right difficulty level.

The name says it all — on the tatami, a coach observes every movement and adjusts the drill in real time. Tatami does the same for math.

How it works

AI-Powered Diagnosis

Computer vision reads handwritten work. A computer algebra system verifies answers symbolically. An LLM analyses the reasoning behind errors — not just what went wrong, but why.

Targeted Exercises

No one-size-fits-all. Each exercise targets a specific gap at the right difficulty level, like a drill designed for a specific weakness.

Spaced Repetition

The SM-2 algorithm ensures nothing that's been learned is forgotten. Each concept comes back at the right moment — not too soon, not too late.

Visible Progression

Inspired by the belt and stripe system in Brazilian jiu-jitsu. Progress is concrete and visible — not abstract percentages, but earned milestones.

The parallel

On the mat	On Tatami
The coach observes and corrects in real time	The AI diagnoses recurring errors
Drills target specific weaknesses	Exercises target detected gaps
Difficulty adjusts to the student's level	Adaptive difficulty, no one-size-fits-all
Belts and stripes: visible progression	Stripe-based progression paths
You fall, you understand why, you try again	You make mistakes, you understand why, you move forward

The story

Tatami was born on a Thursday evening in March 2026. A father and son share the same BJJ club — same sessions, same passion. The son is a triple French champion who also needs to succeed in math. The father is a Rust developer diving into AI.

That evening, the idea was obvious. The next morning, the first line of code was written.

Read the full story on tatami.fpinsight.com

Technology stack

Layer	Technology
Language	Rust
Frontend	Leptos (server-side rendered)
Vision	Computer vision for handwriting recognition
Math Engine	Computer Algebra System (CAS)
AI	LLM for diagnosis and guided reasoning
Learning	SM-2 spaced repetition algorithm