CYBERLEARN
CyberLearn is the project everything else orbits. It is a Turborepo monorepo on Next.js 15, with Supabase and Prisma underneath — a French-first platform for learning cybersecurity that is built as a product, not a demo. The interesting engineering was never the lesson content. It was the data model, the authorisation boundary, and the systems that decide what a learner sees and when.
Authorisation is enforced in PostgreSQL through Row-Level Security, not in the application layer. The reasoning is the same one the rest of my work keeps arriving at: the application is not the only thing that can reach the database, so the application cannot be the only thing that enforces access. RLS hardening pushes the trust boundary down to the row, where it survives a compromised API route, a misused service key, or a query written by someone who forgot the tenant check. Getting the policies right — and proving they fail closed — was the part that took longest and mattered most.
On top of that sits a gamification layer: streaks, quests, leagues, cosmetics, and an end-of-period "wrapped". The design constraint was to reward consistency without turning learning into a slot machine — motivation systems that respect the learner's time rather than farming it. The content side runs a video pipeline built on Remotion and ElevenLabs, turning written scripts into narrated lessons, because content production is the real bottleneck for any learning platform and automating it is what makes a catalogue maintainable by one person.
The catalogue itself was reorganised into structured DevOps and DevSecOps tracks — sequenced paths rather than a flat list of topics. A learning platform is a schema problem disguised as a content problem: the decisions about progression, prerequisites, and how completion unlocks the next thing constrain everything the product can become later. Most of those decisions had to be made once and be correct, because migrating progression logic under live users is a cost that prototype environments never reveal.