Loading...
Loading...
A production-grade VCE Mathematics exam prep platform
Live
Product
2,138+
Questions
6
Topics
Stripe
Payments
As a VCE student, I spent equal time on every topic — only to realize after the exam that most marks came from a few core areas.
Students don't know what actually matters on the exam.
Can't identify where the thinking broke down.
Expert judgment stays locked in individuals.
The real challenge isn't content — it's ranking. Which question helps this student most right now?
What Students See
/topics | /q/{id} | /me
What Matters
topic_weights + tier + backtest
What Happened
questions | topics | solution_steps
Composite ranking: frequency (25%) + marks (30%) + difficulty (20%) + dependency (15%) + trend (10%). Recent years weighted higher to capture curriculum shifts.
Backtested against 20 years of VCE data — R² = 0.87
Each question annotated into 9 structured tables via Claude API: skill type, concept hierarchy, difficulty (logic chains, calc load, novelty), and risk categories.
A real product handling payments needs real security — not an afterthought.
JWT sessions + bcrypt password hashing
CSRF protection on all mutations
XSS sanitization + HTTP security headers
Rate limiting + brute-force protection
100-level progression system with Explorer ranks (Space Cadet to Galaxy Legend)
Consecutive study day tracking with freeze system (1 free/week) and loss notifications
Concept mastery badges, streak milestones, accuracy achievements
Topic mastery heatmap, accuracy charts, study time tracking, progress visualization
Free
Practice by year, 3 solutions/day
Scanned
+ Practice by concept, unlimited brief solutions
Premium
+ Skill variants, detailed solutions, mistake notebook

Students set their target study score — the system adapts everything to their goal.

2,138 real VCE exam questions organized across 6 core topics with concept hierarchy.

Behind every question: structured annotations that tell the system what skill is tested, how hard it is, and where students go wrong.
I chose a statistical model over black-box ML for topic weights. Teachers need to see why a concept ranks high — a model people trust beats one that's 5% more accurate but nobody understands.
Adding Stripe with 3-tier access forced me to answer: what's worth paying for? That question reshaped the entire feature set and taught me more about product strategy than any textbook.
Raw LLM output varies every run. I designed structured annotation schemas that turned unreliable AI extraction into a repeatable pipeline — a product design problem, not just an engineering one.
20 years of exam data could support endless features. The real skill was deciding what not to build — keeping the product focused on what actually helps students score higher.
"The complexity lives in the system. The simplicity lives in the experience."