Apprenant — a complete educational platform
The problem
7 million adults in France are functionally illiterate. They can speak French, but cannot read or write.
Digital platforms exist, but they offer fixed catalog content. None lets the teacher create custom exercises from workshop texts — and none combines the MNLE approach, RGAA accessibility, and sovereign hosting.
Teachers — often volunteers, aged 55-70 — work with paper materials. They need a simple, respectful, ready-to-use tool.
The solution
Apprenant is a complete web platform for teaching adults to read and write — targeting functional illiteracy and French as a foreign language. The teacher creates training sessions from real-life texts (driving code, administrative forms, cooking recipes), and the platform generates fully customizable interactive exercises.
The learner works on mobile, at their own pace. The interface is clean, respectful — formal address, adult feedback, no childish gamification. A voice AI (Mistral Small, hosted in France) helps non-readers navigate.
The numbers
35
Exercise types
Functional complexity — not a CRUD
4,100+
Automated tests
Industrial quality, not a prototype
1
Developer
Product architect productivity
€52/mo
Infrastructure
Frugal architecture
Vue 3 + NestJS + Prisma + tRPC + PostgreSQL
Tech stack
76 Playwright E2E tests
Verified user journeys
Clever Cloud (France)
Data hosted in France, GDPR compliant
Mistral Small
Sovereign AI to assist teachers and learners
How AI helped
I designed an automated workflow — BMAD — that orchestrates Claude Code agents. Every feature goes through a complete pipeline:
- 1Detailed specification (acceptance criteria, a11y constraints, edge cases)
- 2Strict TDD development (test first, implement second)
- 3Automated triple review: code, accessibility, security
- 4Fix all findings
- 5Local gate: lint + typecheck + 4,000+ unit tests + 76 E2E tests
- 6Deployment
A rigorous framework where AI is a guided tool, not an autonomous colleague.
What AI didn't know
AI writes code. But a product isn't just code — it's decisions.
- An engine, not a libraryExisting platforms offer catalog content. I made the opposite choice: the teacher enters their workshop text, the platform generates the exercises. Zero pre-made content. This is the foundation of MNLE.
- 4 combined pedagogical frameworksMNLE (Freinet/De Keyzer), action-oriented approach (CEFR), competency-based approach (CléA/RCCSP), functional approach (ASL).
- Free for nonprofits, alwaysThe business model relies on companies (via OPCO training funds), not on volunteers. Nonprofit organizations will never pay — that's a founding principle, not a launch phase.
- 100% visual and audio feedbackUsers can't read. Every interaction — instructions, feedback, navigation — works without text. Formal address, no confetti, no stars. Respect, not school-style gamification.
- Data collection from day 1Even though AI analytics won't arrive until 2028, the architecture silently collects error patterns, response times, and progress. Funder data (ESF+, OPCO, Qualiopi) is native.
- Sovereign hosting, GDPR by designVulnerable users = sensitive data. Clever Cloud (France), Mistral Small (sovereign AI), 2 separate Prisma schemas (identity / pedagogy), formalized DPIA. Not a compliance add-on — a day-1 architecture decision.
- Complement, not competitorApprenant integrates into the existing ecosystem. Teachers keep their paper workshops, their methods, their habits. The tool complements — it doesn't replace.
Designed with
- Francine Dessis, pedagogical advisor
- Elisabeth Pelloquin, practitioner-researcher in basic learning pedagogy
With guidance from:
- Bernadette Gueritte-Hess, speech therapist and psychomotor therapist
Your project has the same ambition?
A complete product, well-built, that respects its users. Let's talk.
Discover the approach