Overview
TeachAny (GitHub: weponusa/teachany) is an open-source effort to encode learning science as executable rules for AI courseware generation. The repository ships 137+ courseware modules and 2,378+ knowledge nodes, licensed under AGPL-3.0 with a separate commercial license.
The project's stated positioning is: every school, every teacher, and every parent should be able to build a custom Khan Academy for every child at zero cost.
The Problem with AI-Generated Teaching Content
Generic ChatGPT or Claude outputs typically produce flat text walls ending in a multiple-choice question, with no motivational design, no error diagnosis, and no learning loop. TeachAny's diagnosis is that the issue is not model capability but prompt design lacking a pedagogical framework.
Six Learning-Science Theories as Hard Constraints
TeachAny converts theory into enforceable rules, not advisory suggestions:
1. ABT Narrative Structure (And-But-Therefore)
Every module opens with what the learner already knows, the contradiction, and why it matters. Sourced from Randy Olson's *Houston, We Have a Narrative*.2. Bloom's Taxonomy
Practice items must cover all six cognitive levels (Remember through Create), with at least three levels required per module. Plain AI output typically stops at Remember and Understand.3. ConcepTest (Mazur Peer Instruction)
Question difficulty targets a 30–70% success rate, the "sweet spot" validated by Eric Mazur at Harvard. Higher means too easy; lower causes frustration. The evaluation system flags items outside this band.4. Cognitive Load Theory (Sweller)
Three hard limits: each information card ≤75 characters; each module carries only one core question; new concepts appear together with examples. Built directly on John Sweller's cognitive load theory.5. Mayer's Multimedia Learning Principles
Seven principles are enforced on every layout decision, including spatial contiguity, signaling, segmenting, and pre-training. Mayer reports 50–100% learning gains when these are followed.6. Scaffolding
Three levels: full template with fill-ins, hints only, or independent work. Students progress from Level 1 upward. This implements the Wood, Bruner, and Ross (1976) tutoring model as an internal adaptive mechanism.Nine Subject-Specific Frameworks
TeachAny rejects one-size-fits-all templates. Each discipline has its own pedagogical path and assessment style:
- Math: visual intuition → algebraic reasoning → generalization; graphing, dragging, step-by-step derivation
- Physics: observe → model → quantitative analysis; parameter sliders, experiment prediction
- Biology: structure → process → function; labeling, sequencing, flowcharts
- Geography: spatial distribution → causes → regional comparison; map interaction, causal chains
- History: timeline → causation → evidence → multiple perspectives; ordering, source comparison, standpoint analysis
- Chinese language: close reading → rhetorical technique → thematic transfer; annotation, rewriting, imitation
- English: input → scaffolded output → situational application; dialogue cards, role-play
- Chemistry: macroscopic → microscopic → symbolic representation; experiment flow, equation balancing
- IT: task-driven → step-by-step demo → debugging; click-through, flowcharts, code execution
- 👁️ See It – observe phenomena, examples, data
- 🔧 Break It – decompose structure, steps, components
- 💡 Explain It – clarify causes, mechanisms, rules
- ⚖️ Compare It – contrast with similar/opposite/incorrect cases
- 🎯 Transfer It – apply to new contexts
- 5 curriculum standards: China K-12, AP, Cambridge IGCSE/A-Level, IB MYP/DP, US CCSS + NGSS
- 48 subjects × 5 systems = 98 knowledge graphs (JSON)
- 2,399 detailed knowledge-point files with standard alignment, difficulty grading, and linked resources
- Geographic/historical map assets via Leaflet and dynasty overlays
- AGPL-3.0: free for personal learning, public schools, non-profit education, open-source forks
- Commercial license: required for SaaS, paid courses, and corporate training
- Quality consistency: 12 official modules vs. 124 community modules; no peer review mechanism yet
- China-curriculum depth: 2,399 nodes are a subset; regional curricula (e.g., Zhejiang "expansion courses") may not be covered
- Model dependency: recommended with CodeBuddy; works with Claude, Cursor, Windsurf. Claude 3.5 Sonnet follows constraints well; GPT-3.5 may ignore scaffolding rules. Model floor determines courseware floor.
- Single-file HTML ceiling: zero-dependency is also a limitation. PhET-level simulations, 3D visualization, real-time collaboration, and learning analytics exceed the architecture's reach.
- Empirical validation: the six learning-science theories are implemented as constraints, but no controlled studies in the repository confirm that ABT openers outperform traditional ones, or that 30–70% ConcepTest success rates transfer across cultural contexts. These are hypotheses encoded as practice, not validated outcomes.
- GitHub: https://github.com/weponusa/teachany
- Online gallery: https://weponusa.github.io/teachany/
- Gitee mirror: https://gitee.com/weponusa/teachany
- Olson, R. (2015). *Houston, We Have a Narrative*. University of Chicago Press.
- Sweller, J. (1988). Cognitive load during problem solving. *Cognitive Science*, 12(2), 257-285.
- Mayer, R.E. (2009). *Multimedia Learning* (2nd ed.). Cambridge University Press.
- Mazur, E. (1997). *Peer Instruction: A User's Manual*. Prentice Hall.
- Wood, D., Bruner, J.S., & Ross, G. (1976). The role of tutoring in problem solving. *Journal of Child Psychology and Psychiatry*, 17(2), 89-100.
Original Methodologies
Five-Lens Method
For any difficult concept, the system forces selection of 2–3 lenses:Six-Question Pre-Design Framework
Before any code is written, the AI must answer: Who is the student? What are prerequisites? What should the student be able to *do*? What is the real-world context? Where do students typically get stuck? How is mastery verified?PBL Decomposition Engine
Input a project description (e.g., "design a smart greenhouse", "build a weather app"); the engine identifies required knowledge points, maps them to standards (China K-12, AP, Cambridge, IB, CCSS), and generates a learning path spanning multiple subjects. The core claim is that PBL is the main vehicle for teaching, not a supplement.Technical Implementation: Zero-Dependency Single-File HTML
Every module is a single HTML file. No build step, no npm dependencies, no framework. Teachers can open files directly in a browser, run them offline, on any OS, including older tablets. Files are editable via "View Source".
Interactions are implemented with native Canvas API, SVG, and small amounts of vanilla JavaScript. Examples include Canvas photosynthesis animations, draggable equation balancing with TTS, and a virtual circuit lab with real-time V-I plotting.
Knowledge Layer and Curriculum Alignment
The data/ directory contains:
For input like "Grade 8 math — quadratic functions", the system can locate the China-standard knowledge point, retrieve prerequisite chains, and surface common error patterns.
Licensing: AGPL-3.0 + Commercial
The author has registered the TeachAny™ trademark and also claims prior-use rights on the Chinese mark 教我学™. Forks must be renamed. Commercial inquiries: weponusa@gmail.com (subject prefix [TeachAny Commercial] or [TeachAny Trademark]).
Limitations
Significance
TeachAny's distinctive contribution is not "using AI for courseware" (many tools do that) but converting learning science from advisory reading into enforced production rules. ABT, Bloom, cognitive load, and scaffolding have been in pedagogy textbooks for decades; teachers know they exist but rarely know how to enact them per lesson. TeachAny encodes them as machine-readable constraints so every generated module passes a teaching-design check.
For Chinese teachers specifically, the curriculum alignment layer is a separate value: inputting "Grade 9 physics — Ohm's Law" returns not just a module but the standards position, prerequisites, common errors, and AP/Cambridge equivalents.