English static mirror for SEO/GEO · AI-assisted translation · Read Chinese original

TeachAny: Open-Source AI Lesson-Generation System That Encodes Learning Science Into Rules

Forum topic · 小凯 · 2026-05-24

Summary

TeachAny is an open-source project (GitHub: weponusa/teachany) that turns learning-science theory into enforceable rules for AI-generated teaching materials, rather than relying on generic prompts. Instead of producing flat text walls, the system embeds ABT narrative structure, Bloom's taxonomy (exercises must span at least three cognitive levels), ConcepTest design targeting 30-70% correct rates, Sweller's cognitive load constraints, Mayer's multimedia principles, and a three-level scaffolding model. It rejects one-size-fits-all templates: nine subjects (math, physics, biology, geography, history, Chinese, English, chemistry, IT) each have dedicated teaching paths and interaction types. Original methodologies include a Five-Lens approach for hard concepts, a six-question pre-design framework, and a PBL decomposition engine that maps projects to standards. Courseware ships as zero-dependency, single-file HTML files that work offline, built with Canvas, SVG, and vanilla JS. The data layer includes 98 knowledge graphs across five curricula (Chinese standards, AP, Cambridge, IB, CCSS/NGSS) with 2,399 knowledge-point detail files. The project currently offers 137+ lessons and is dual-licensed under AGPL-3.0 with a separate commercial license. The author notes limitations: inconsistent community courseware quality, limited Chinese-curriculum coverage depth, model-dependence of AI skills, single-file architecture ceilings, and the absence of controlled experiments validating the pedagogical claims.

TeachAny: Open-Source AI Lesson Generation With Learning Science Built In

> GitHub: weponusa/teachany | 137+ lessons | 2,378 knowledge nodes | AGPL-3.0 + commercial dual license > > Positioning: every school, teacher, and parent can build a free, personalized Khan Academy for every child.

1. The Common Problem With AI-Generated Educational Content

Ask ChatGPT or Claude to create teaching material and you typically get a wall of text with a multiple-choice question bolted on the end — flat, lifeless, with no instructional design, no motivation structure, no error diagnosis, and no learning loop.

The root cause is not that LLMs aren't smart enough, but that prompt engineering lacks a pedagogical framework. TeachAny's solution is to write learning-science theory directly into the AI's workflow as enforced rules.

2. Six Learning-Science Theories, Systematically Embedded

2.1 ABT Narrative Structure (And-But-Therefore)

Every module opens with: And (what you already know) → But (the contradiction or knowledge gap) → Therefore (why this matters). Adapted from Randy Olson's *Houston, We Have a Narrative*.

2.2 Bloom's Taxonomy

Exercises must cover the six cognitive levels — remember, understand, apply, analyze, evaluate, create — with a mandatory minimum of three levels, versus the typical "remember → understand" ceiling of ordinary AI output.

2.3 ConcepTests (Mazur's Peer Instruction)

Concept questions are designed for a 30–70% correct rate — the "sweet spot" Eric Mazur validated at Harvard. The assessment system checks that question difficulty falls in this range.

2.4 Cognitive Load Theory (Sweller)

Three hard constraints: each info card ≤ 75 words; each module carries exactly one core question; every new concept is immediately paired with an example.

2.5 Mayer's Multimedia Learning Principles

Seven principles (contiguity, signaling, segmenting, pre-training, etc.) are enforced on every layout decision rather than left as suggestions.

2.6 Scaffolding Strategy

A three-level support system — full template (Level 1) → hints only (Level 2) → fully independent work (Level 3) — based on Wood, Bruner & Ross's classic 1976 tutoring model, implemented as an adaptive mechanism inside the courseware.

3. Nine Subject-Specific Frameworks

TeachAny refuses one-size-fits-all templates. Each of the nine subjects has its own teaching path, interaction types, and assessment style:

| Subject | Teaching Path | Interactions | Assessment | |---|---|---|---| | Math | Intuition → algebraic reasoning → generalization | Plotting, drag-and-drop, stepwise derivation | Standard problems + explanations | | Physics | Observe → model → quantify | Parameter sliders, experiment prediction | Predict + calculate + explain | | Biology | Structure → process → function | Labeling, ordering, flowcharts | Diagrams + process explanation | | Geography | Spatial distribution → causes → regional comparison | Map interactions, causal chains | Material analysis + comparison | | History | Timeline → causality → evidence → multiple perspectives | Ordering, source comparison, stance analysis | Source-based questions + essays | | Chinese | Close reading → technique → thematic transfer | Annotation, rewriting, imitation | Annotation + writing + rubrics | | English | Input → scaffolded output → contextual use | Dialogue cards, cloze, role-play | Integrated skills assessment | | Chemistry | Macro phenomena → micro explanation → symbolic representation | Lab procedures, equation balancing | Lab explanation + structured response | | IT | Task-driven → stepwise demos → debugging | Click-throughs, flowcharts, code execution | Task completion + process checks |

4. Original Methodologies

4.1 Five-Lens Method

For any difficult concept, 2–3 lenses are mandatory:
  • See It — observe phenomena, examples, data
  • Break It — decompose structure and components
  • Explain It — clarify mechanisms and rules
  • Compare It — contrast with similar/opposite/wrong cases
  • Transfer It — apply to a new context to verify understanding
  • 4.2 Six-Question Pre-Design Framework

    Before generating courseware, the AI must answer: Who is the student? What prior knowledge exists? What should students be able to *do*? What real-world context motivates this? Where do students typically get stuck? How is mastery verified?

    4.3 PBL Project Decomposition Engine

    Given a project description (e.g., "design a smart greenhouse"), it identifies knowledge points, maps them to standards (Chinese curriculum / AP / Cambridge / IB / CCSS), and generates a learning path. Core claim: PBL should be the primary vehicle of instruction, not a supplement.

    5. Technical Implementation: Zero-Dependency, Single-File HTML

    All courseware is single-file HTML — no build steps, no npm, no frameworks. Files open directly in a browser, work offline, run on any platform, and are easy to edit by viewing source. Interactions use native Canvas, SVG, and vanilla JS (e.g., Canvas animations and TTS in the photosynthesis lesson; a virtual circuit lab with live V-I plots in the Ohm's law lesson).

    6. Standards Alignment

    The data/ directory contains:

  • 5 curricula: Chinese standards (primary/middle/high), AP, Cambridge IGCSE/A-Level, IB MYP/DP, US CCSS + NGSS
  • 98 knowledge graphs (48 subjects × 5 systems, JSON)
  • 2,399 knowledge-point detail files with standards mappings, difficulty levels, and linked resources
  • Geography/history map assets (Leaflet, dynasty overlay maps)
  • 7. Business Model: AGPL-3.0 + Commercial License

    Free for personal learning, public schools, non-profit education, and open-source forks (which must be renamed). SaaS, paid courses, and corporate training require a commercial license (weponusa@gmail.com, subject prefix [TeachAny Commercial]). The author claims trademark rights to TeachAny™.

    8. Limitations

    1. Quality consistency: 12 official lessons vs. 124 community lessons, with no peer review mechanism for pedagogical rigor. 2. Chinese-curriculum depth: 2,399 knowledge points remain a subset of China's full K-12 content; local curricula coverage is unclear. 3. Model dependence: Instruction-following varies widely across LLMs — the weakest model sets the floor for courseware quality. 4. Single-file ceiling: PhET-level simulations, 3D, real-time collaboration, and learning analytics are hard in this architecture. 5. Unvalidated claims: The pedagogical principles are encoded but no controlled experiment data demonstrates improved learning outcomes.

    9. Conclusion

    TeachAny's value is not "using AI to make slides" — many tools do that. Its distinct contribution is turning learning science from reference material into enforced production standards, so every auto-generated lesson passes instructional-design checks, while standards alignment tells teachers exactly where a topic sits in Chinese, AP, and Cambridge curricula.

    References

  • 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.

Tags

#teachany#ai-education#learning-science#instructional-design#k12#open-source#pbl#courseware

This page is an English static mirror generated for search and AI citation. It may be a full translation or structured summary of the Chinese original. Canonical interactive discussion lives on the Chinese page: https://zhichai.net/topic/177620713