Mirasuki v2
A learning platform for cram schools. I built the question-selection logic of "Rank-Up Dojo," which picks the next question based on each answer
About
Version 2 of Mirasuki, a learning platform for cram schools. It is a rebuild of the v1 service that is still in operation, and I split the implementation with one other engineer.
I built the question-selection logic for "Rank-Up Dojo," a practice mode. It connects fine-grained learning topics within each unit through prerequisite relationships ("to understand this, you need to understand that") and chooses the next question after every answer.
When a student answers incorrectly, the logic passes a "possible gap" signal to the topic's prerequisites and asks questions that trace back to the most suspicious one. If it finds a gap, it reinforces that topic and then leads the student back up to where they started. With consecutive correct answers, the difficulty rises within the unit from C to B to A to S, one question at a time.
The logic is written as pure functions that take the current time and random seeds as inputs, so the same inputs, such as the sequence of answers, the time, and the seed, always produce the same questions. This lets us reproduce and verify its behavior while tuning.
Built with
- Next.js
- React
- TypeScript
- Prisma
- PostgreSQL
- Supabase
- Cloudflare Stream
- Cloudflare R2
- Vitest