Founding Engineer, Mentiora

Mentiora is an AI-powered GCSE revision platform. As founding engineer I led its full-stack development, and the part I'm proudest of isn't that it "used AI." It's that it turned assessment signals into a persistent learning loop that kept working as it scaled to roughly 10,000 monthly learners at peak.

The problem

Most study tools throw content at students. The harder, more useful problem is knowing what a specific student should do next: which topics are weak, which are overdue for review, and how to turn a wrong answer into a better recommendation tomorrow.

My role

Founding engineer. I designed the core architecture and built across the stack: authentication, database schema, payments and subscriptions, usage-based access control, diagnostics, daily study plans, mastery tracking, AI feedback, and question-generation workflows.

The learning loop

1. 01Student authenticates and picks subjects, exam boards, topics, and target grades.
2. 02Diagnostic questions produce an initial mastery estimate.
3. 03A daily plan prioritizes weak or overdue topics.
4. 04Exam-style practice sessions run; objective questions are graded locally.
5. 05Written responses get AI-assisted grading and feedback.
6. 06Mastery updates, and the next day's recommendations change accordingly.

The design principle throughout: keep the product logic deterministic and let the model do the parts only a model can. Which topic to schedule, when a student has demonstrated mastery, and what they're allowed to access are decided by code. The model generates practice and explains written answers.

NoteThe interesting engineering problems+

• Modeling mastery in a way that's stable enough to schedule against but responsive to new evidence.
• Avoiding repetitive question generation: students notice immediately when the "new" question is the old one reworded.
• Structuring AI outputs so generated content slots into deterministic product logic instead of leaking uncertainty into it.
• Consistency across many exam subjects, each with its own conventions.
• Usage gating and subscriptions that hold up under real, paying traffic.

Architecture

Next.js App Router + TypeScript, Supabase for auth and database, OpenAI APIs for generation and grading, Stripe for payments, Resend for email, PostHog for product analytics, and a Turborepo/pnpm setup. Radix/shadcn-style components and Recharts on the frontend.

What I take from it

This is where I learned production ownership: not just shipping features, but being responsible for the architecture, the payments, the reliability, and the experience of thousands of real learners. It's also the direct ancestor of WayPoint, an AP-study platform where I'm applying these lessons to a product I own more deliberately (currently building).

Status

Completed role (Dec 2025 – Mar 2026). Founding Engineer.