SONG JUNHYUK MATH LAB · AI Mathematics Content Lab
We connect a past-exam and textbook question bank, AI classification, verified solutions, automated textbook publishing and student management into one AI pipeline for Korean mathematics education — tested first in our own classes, then offered to academies and instructors.
6,697
Problems analyzed
31 commercial textbooks · 6,401 pages
695
Past exam sets
National CSAT, regional & academy exams
115
Taxonomy nodes
Unit → type → sub-type
574
Solution guides
Standard methods & wording per type
Pipeline
Collect
Past exams, textbooks, school tests
Extract
Vision AI → digital problems + LaTeX
Classify
Type, importance, difficulty
Solve
Two independent verifications
Publish
Typeset books & exams (HWPX/PDF)
Teach
Marking, error review, reports
A single photo or PDF page becomes digital problems, including math expressions, stored in one question bank.
In operationEvery problem is classified on the same scale, so any new workbook fits into one searchable system.
In operationStep-by-step solutions in the lab's consistent house style, self-checked by AI and confirmed by an instructor.
In operationSelect problems and the main book, homework and answer key are typeset to HWPX and PDF.
In operationAssignments, marking, error logs and grades in one place, fed back into lesson design.
In operationHow we use AI
We build our product and backend with Claude Code on a Claude Max plan — the Next.js/Supabase classroom platform, the content pipeline and the publishing tools. AI also drives extraction, classification and solution drafting in that pipeline.
Integrate the Claude API directly into the pipeline — vision extraction, taxonomy classification, solution drafting and independent re-verification — and evaluate math accuracy with structured outputs and batch processing.
Every explanation and textbook is checked by an instructor before it reaches students. Low-confidence classifications and mismatched solutions are routed to human review automatically.
Company
Also from the lab
A prototype-stage product project: a toothbrush-mounted motion sensor and companion music app for children aged 4–8. It is not yet on sale and does not offer brushing diagnosis.
singbrush.mathstudyapp.comWork with the lab
Mathematics content · AI learning solutions · research partnerships