The problem
Online learning largely failed because it copied a classroom format that was already failing in person. The shift we wanted to make was simple: every student gets a tutor sitting next to them, every lesson adapts to where the student actually is, and the gamification layer is honest — rewarding mastery, not minutes-on-task.
What we built
BrightPath is a full learning environment: a course library spanning maths, literacy, science, history, and electives; an AI tutor (Pax) embedded in every lesson; an XP-and-streak rewards system; and a dashboard for parents that shows what was learned versus what was just clicked through. The frontend is a responsive Next.js app; the backend is a Node API on Postgres; the AI tutor is routed through OpenRouter so we can pick the model with the right cost-quality trade-off per task.
The AI angle
Pax has two jobs. The first is reactive — answer the student’s question without giving the answer away when the goal is comprehension. The second is proactive — notice when a student is stuck, surface a hint scaffolded to their level, and adapt the next problem. We use Gemini Flash for the cheap real-time work and Claude for the longer, harder tutoring conversations. Every interaction is logged against the curriculum node, which gives us an evals corpus to keep Pax pedagogically sound.
How it’s used
- Homeschool families use BrightPath as the spine of their week.
- Hybrid schools use the platform between in-person sessions to keep students moving.
- Tutoring centres use the AI tutor as a force multiplier for their human tutors.
What it taught us
That kids will tell you immediately if your AI tutor is bad. The feedback loop is brutally fast and you cannot hide behind metrics. The single most important investment was the prompt that controls how Pax handles “just tell me the answer” — balancing patience, humour, and the line between scaffolding and stonewalling.