Overview
TeachAny (GitHub: weponusa/teachany) is an open-source initiative with a bold positioning: let every school, teacher, and parent build a personalized Khan Academy for every child at zero cost. Its stats: 137+ courseware items, 2,378 knowledge nodes, AGPL-3.0 + commercial dual license.
The Problem with AI-Generated Educational Content
Materials generated by ChatGPT or Claude typically look like flat bullet-point summaries followed by a multiple-choice quiz — no motivation design, no error diagnosis, no learning loop. The root cause is not that LLMs are insufficiently capable, but that prompt engineering lacks an educational framework. TeachAny's solution is to write pedagogy directly into the AI workflow as enforced rules.
Six Learning Science Theories as Hard Rules
- ABT narrative structure (And-But-Therefore): each module opens by connecting prior knowledge (And), surfacing a cognitive gap (But), and motivating the lesson (Therefore). Adapted from Randy Olson's *Houston, We Have a Narrative*.
- Bloom's taxonomy: exercises must cover all six cognitive levels (remember, understand, apply, analyze, evaluate, create), with a minimum of three levels enforced.
- ConcepTest (Mazur peer instruction): concept questions target a 30-70% correct rate — the validated "sweet spot" from Eric Mazur's Harvard physics teaching.
- Cognitive load theory (Sweller): hard constraints — each info card ≤ 75 characters, one core question per module, examples immediately follow new concepts.
- Mayer's multimedia learning principles: proximity, signaling, segmenting, pre-training and more are mandatory layout rules, not suggestions.
- Scaffolding (Wood, Bruner & Ross, 1976): a three-level support system — full templates, hints only, then fully independent work — progressively withdrawn.
- Five-Lens Method: for any hard concept, choose 2-3 lenses — See It, Break It, Explain It, Compare It, Transfer It.
- Six-Question pre-design framework: before generating anything, the AI must answer: Who is the student? What prior knowledge? What should they be able to *do*? What real-world context? Where do students typically get stuck? How is mastery verified?
- PBL project decomposition engine: input a project (e.g., "design a smart greenhouse"), output a full knowledge network mapped to curricula (China standards / AP / Cambridge / IB / CCSS) with an ordered learning path.
- GitHub: https://github.com/weponusa/teachany
- Online gallery: https://weponusa.github.io/teachany/
- China mirror (Gitee): 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.
Subject-Specific Frameworks (No One-Size-Fits-All)
Nine disciplines each get a dedicated teaching path, interaction types, and assessment styles:
| Subject | Teaching path | Interactions | |---|---|---| | Math | Intuition → algebraic reasoning → generalization | Plotting, drag-and-drop, stepwise derivation | | Physics | Observe → model → quantify | Parameter sliders, experiment prediction | | Biology | Structure → process → function | Labeling, ordering, flowcharts | | Geography | Spatial distribution → causes → regional comparison | Map interactions, causal chains | | History | Timeline → causality → evidence → perspectives | Source comparison, stance analysis | | Chinese | Close reading → technique → theme transfer | Annotation, rewriting, rubrics | | English | Input → scaffolded output → application | Dialogue cards, role-play | | Chemistry | Macro phenomena → micro explanation → symbolic | Equation balancing, lab flows | | IT | Task-driven → demo → debug | Click ops, code execution |
Original Methodologies
Technical Implementation: Zero-Dependency Single-File HTML
All courseware is single-file HTML — no build step, no npm, no framework. This means no programming skill required for teachers, offline usability, cross-platform compatibility (even old tablets), and easy editing via view-source. Interactions use native Canvas API, SVG, and vanilla JS (e.g., a virtual circuits lab with real-time V-I plotting in the Ohm's Law courseware).
Curriculum Alignment
The data/ directory contains five curricula (China K-12, AP, Cambridge IGCSE/A-Level, IB MYP/DP, US CCSS + NGSS), 98 knowledge graphs (48 subjects × 5 systems), and 2,399 knowledge-point detail files with prerequisite chains, difficulty ratings, and common error databases.
Business Model
Dual licensing: AGPL-3.0 for personal learning, public schools, non-profits, and open-source forks; commercial licenses required for SaaS, paid courses, and corporate training (contact: weponusa@gmail.com with subject prefix [TeachAny Commercial]). Forks must rename; trademarks TeachAny™ / 教我学™ are claimed.
Limitations
1. Consistency: only 12 of 137 courseware items are official; community contributions vary and lack peer review. 2. China curriculum coverage: 2,399 knowledge points remain a subset of full K-12 content; local curricula may be missing. 3. Model dependency: instruction-following varies widely across LLMs; the model's floor determines courseware quality. 4. Single-file ceiling: PhET-level simulations, 3D, real-time collaboration, and learning analytics strain the zero-dependency architecture. 5. Unverified efficacy: the learning-science claims lack controlled experimental data — they are claims to be validated, not validated conclusions.
Conclusion
TeachAny's distinctive value is turning learning science from "reference material" into "enforced production rules," and pairing it with curriculum-standard alignment so a teacher can type "Grade 9 physics: Ohm's Law" and receive pedagogically validated, standards-mapped courseware.