From lived need to responsible scale.
T-Twice is not another answer engine. It is a mathematics reasoning environment designed for students who must learn to produce reasoning, teachers who need visibility without surveillance, and institutions that want AI tutoring without weakening cognitive autonomy.
It started from a learning failure, not a market slide.
The most important evidence behind T-Twice is the experience it was built from: the difference between understanding a solution and being able to produce one.
Candis’s starting point
As a mathematics student, Candis used generative AI to understand exercises and lectures. The explanations felt clear, the solutions looked convincing, and the feeling was one of progress. But exam situations revealed a different reality: reading a solution was not the same as building the reasoning required to reproduce it.
The turning point came when a former mathematics student refused to give direct answers. He made her write her reasoning down, even when incomplete, and helped her identify the exact point where the reasoning broke. That slower, sometimes frustrating process was the first time mathematics learning became durable.
Later, as a private tutor, including with students facing dys-related frictions, she observed the same pattern: many students do not need one more complete solution. They need someone to keep them inside the reasoning process long enough for learning to happen.
One interaction gives polished solutions. The other creates a protected space for attempts, errors, correction and autonomy. T-Twice is built for the second.
Why learning through error is not a weakness.
T-Twice is aligned with a strong learning-science idea: mistakes are not noise to hide, but signals to process.
Productive failure
Kapur’s work on productive failure in mathematics shows that students can learn more deeply when they first attempt to solve problems, generate imperfect solutions, and then consolidate the correct concepts afterwards.1 The point is not failure itself; it is guided processing of failure.
Elaborating on errors
Research on learning from errors indicates that error elaboration and comparison are crucial when problem solving comes before instruction.2 This directly supports T-Twice’s design: identify the type of error, then ask one targeted question.
Attention to mistakes
Moser and colleagues found that a growth mindset was associated with stronger attention to mistakes and better post-error accuracy.3 T-Twice therefore avoids humiliating correction and frames errors as usable data.
Retrieval and active effort
Roediger and Karpicke showed that retrieval practice improves long-term retention more than passive restudy.4 T-Twice asks students to retrieve, attempt, justify and repair — not just read.
T-Twice does not celebrate being wrong. It turns wrong or incomplete reasoning into a structured learning event.
The demand already exists. The direction is the issue.
Students, families and institutions already seek AI-supported learning. The strategic question is whether this demand is captured by answer engines or by tools that preserve reasoning.
Grand View Research estimates the global education technology market at USD 187.01B in 2025, projected to USD 437.54B by 2033.5
The French online tutoring services market generated USD 438.2M in 2024 and is projected to reach USD 1.04B by 2030.6
Mordor Intelligence estimates the AI in Education market at USD 6.90B in 2025, projected to reach USD 41.01B by 2030.7
How T-Twice differs from existing AI tutoring tools.
The competition validates the need. T-Twice’s value is to occupy a precise gap: mathematical reasoning, course anchoring, teacher calibration and cognitive friction reduction.
Chatbots
Powerful and flexible, but generally optimised for helpful completion. Students can receive a full solution before attempting the reasoning.
Photomath / Gauth
Photomath emphasises scanning problems and step-by-step explanations, while Gauth presents itself as an AI homework helper with answers and expert help.89
Khanmigo
Khanmigo validates the “no direct answer” direction by challenging students to think critically without giving direct answers.10
Reasoning companion
T-Twice starts from the student’s attempt, diagnoses mathematical reasoning errors, adapts to the teacher’s course and keeps the teacher in the loop.
Course anchoring
Definitions, notation and expected proof style come from the professor’s actual course, not from a generic explanation.
Reasoning diagnosis
The system distinguishes missing hypothesis, theorem misuse, logical jump, definition confusion, quantifier error and writing friction.
Teacher visibility
The teacher sees patterns of misunderstanding rather than only final answers, with consent and pedagogical context.
Interface as cognition
Keyboard, voice and natural-language-to-symbol input are treated as learning infrastructure, not cosmetic features.
Autonomy by design
The goal is not time-on-platform. The goal is for students to internalise the questions T-Twice asks.
Math-first depth
T-Twice is dedicated to mathematical proof, formalisation and rigorous problem solving, especially where generic tutors become too broad.
Inclusion is not an extension. It is the architecture.
T-Twice is designed for students whose mathematical ability may be hidden by reading load, writing effort, notation friction, attention difficulty, visual accessibility needs, or lack of access to private tutoring.
Dys profiles
Dyslexia, dysorthography, dyscalculia and dyspraxia are treated as friction patterns to reduce, not labels to display. The mathematical standard remains high; the access path changes.
Voice and multimodal input
Students can express reasoning orally, through natural language, symbols or guided input. This protects students whose thinking is stronger than their written production.
Accessible learning
UNICEF’s accessible digital learning work stresses affordable, personalised and accessible formats for diverse needs.11 T-Twice follows that direction for mathematical reasoning.
Teacher-supported inclusion
The dashboard shows behavioural and reasoning patterns without diagnostic labelling, helping teachers adjust support while reducing bias and surveillance.
Dedicated to mathematically demanding learning contexts.
T-Twice is not built for every subject. It is designed for diplomas and training paths where mathematical reasoning is central.
Education levels
Institution types
Use cases
- Students preparing exams in mathematics, probability, analysis, algebra, statistics, actuarial science, engineering or quantitative fields.
- Teachers who want to see where a class is blocked before an exam, without turning AI into a grading or surveillance tool.
- Institutions that want a controlled alternative to generic AI use in assignments.
- Students with dys, attention or accessibility frictions who need the barrier reduced without lowering the intellectual demand.
A European product with an African access mission.
The vision is not only to sell a software licence. It is to make responsible mathematical AI tutoring deployable in different educational infrastructures.
France pilots
Consolidate university, lycée, tutoring and training-school pilots. Validate learning outcomes, teacher workflows and accessibility needs.
European scaling
Deploy across French-speaking and European institutions with GDPR-first hosting, professor calibration and multilingual mathematical notation support.
Africa partnerships
Work through universities, ministries, NGOs, foundations and education-focused organisations rather than direct consumer-only expansion.
Offline-first mode
Develop low-bandwidth and offline modes for constrained environments: cached course content, local exercises, delayed synchronisation and mobile-first access.
Why sustainability matters ethically.
A responsible AI tutor needs maintenance, evaluation, compliance, accessibility work and teacher support. Sustainability is not a commercial add-on; it is what prevents the prototype from becoming unsafe or abandoned.
What must be funded over time
- Independent evaluation and controlled pilots.
- Accessibility and dys/neurodivergent adaptations.
- European hosting, model monitoring and data protection.
- Teacher onboarding and support.
- Offline and low-bandwidth modes for partner deployments.
- Pedagogical quality review by mathematics teachers and researchers.
The ethical point
If only answer engines survive economically, the market rewards solution consumption. T-Twice tries to make the opposite model viable: an AI tool that succeeds when the student becomes more autonomous, not more dependent.
UNESCO’s digital education work emphasises human agency, critical thinking and ethics in the age of generative AI.14 T-Twice turns that principle into a mathematics-specific product design.
The deployment hypothesis.
This document does not claim that T-Twice has already proven large-scale impact. It argues that the need is real, current alternatives leave a specific gap, the POC is testable, and the next responsible step is a larger, independently evaluated deployment with inclusion and accessibility built in from the start.