Fig. 01 · YouTube cutKorean CSAT · radius overlap
AI-native · Goal-oriented · STEM tutoring
Near-1:1 STEM tutoring,
at software cost.
ConvyCanvas maps the knowledge space, then turns it into conversational tutoring, adaptive curriculum, test prep, and interactive video lessons.
Manybooks mined
10k+problems and answers
ThousandsSAT problems and answers
100svideos and animated lectures
Knowledge engine
Books and exams become adaptive curriculum.
Rawcurated
Inputs
Books, exams, figures, solutions, and problem corpora from K-12, SAT, and college STEM.
mine
Codeengine
Graph
Semantic, spatial, prerequisite, answer, figure, and solution edges map the learner's knowledge space.
generate
Artifactsgenerated
Tutor
One graph powers hints, checks, conversational explanations, video lessons, and goal-specific paths.
Exhibit · one book mined
Stewart Precalculus 8e, fully traversed.
18,928items
32edge types
7,741exercises linked
2,207answers
1,189figures
1,029pages
Across sources: 1.74M document lines · 12,780 problems/examples · ~2M generated content lines · 29 shared skills. Stewart is the worked example.
Currently shipping
Manybooks mined
6calculus books
10k+problems and answers
ThousandsSAT problems and answers
100svideos and animated lectures
Interactive output
Lessons become movable, watchable, and testable.
Fig. 02 · Book mapStewart Precalculus 8e
An entire textbook, as a graph.
18,928items
12/12chapters
359,336edges
32 edge types
71,128 semantic
288,208 spatial
1,189 figures
1,029 pages
Fig. 03 · 3D explainerBesant · Conic Sections (1884)
Static 2D becomes touch-friendly live 3D.
Fig. 04 · Proof tourNewton · Principia · Prop. XI
Proof. Play. Tour. Plot.
Fig. 05 · YouTube cutAnalytics explainer
Ships as video, too.
Fig. 06 · YouTube cutEuclid · construction build
Ancient proofs, modern frame.
Video channels
Video is distribution. The tutor is the product.
Video channels prove demand and expose learner questions. Those signals route back into adaptive practice for SAT, school support, and college STEM.
ReferencePremath
What demand looks like today.
Ours · generatedGeoMethod
Premath-style, better production.
Ours · free-rangeChang's Canvas
Creative problem solving, clear aha moments.
Distribution becomes tutoring data: comments, wrong turns, and repeated questions tell the system what learners need next.
GoalSAT score, exam, topic, proof, or missing prerequisite.
ConversationQuestions, numeric checks, voice prompts, and mistakes reveal understanding.
AdaptationThe graph selects next problems, hints, and low-cost tutoring paths.
Why now
Every AI advance makes the tutor better and cheaper.
01
Map the knowledge space.
AI extracts problems, diagrams, answers, prerequisites, and relationships from dense books and exams.
02
Direct the lesson.
An AI director handles flow, narration, scripts, and motion while users set goal, pacing, voice, and style.
03
Adapt through conversation.
Questions, numeric checks, and mistakes select the next hint, exercise, explanation, or test-prep path.
04
Compound with AI progress.
Better models improve mining, explanations, translation, code generation, and tutoring quality on the same graph.
Goal-driven STEM tutoring: interactive, conversational, adaptive, and low-cost at scale.
ConvyCanvas · Touch and voice · Mined curriculum · Adaptive learning