Through my work at Learnful over the past decade or so, I’ve helped build systems that make it easier for educators to create, adapt, and share Open Educational Resources (OER).
GenOER (genoer.org) is one of my favourite projects so far: a fully autonomous OER generator where every resource is released into the public domain. GenOER is not just a tool — it’s a testbed for exploring one of the most pressing questions in open education today: what happens when the “author” of an open resource is an algorithm?
I hope GenOER sparks discussion, critique, and collaboration across the OER community. This is an experiment, but one that asks real questions about scale, pedagogy, licensing, and the future of authorship.
Best,
Yasin Dahi
LinkedIn
How It Works
GenOER runs on a simple but powerful model: two autonomous agents that simulate aspects of educational design.
Educator Agent – synthesizes a need for a quiz by defining the topic, education level, and cognitive goal (e.g., “Grade 10 Biology – Apply – Photosynthesis”). This role reflects the pedagogical judgment educators bring to identifying what learners need.
Designer Agent – refines those parameters, applies pedagogical principles, and generates the quiz itself, including the questions and tailored feedback for each answer.
This two-agent structure mirrors the way humans design assessments: one identifies the why and what of learning, the other focuses on the how. In the future, this model could scale to many Educator and Designer agents, each with distinct expertise, styles, or pedagogical goals.
Role of Humans
Although GenOER is fully autonomous, educators remain central to shaping, evaluating, and applying its outputs. The platform is designed not to replace educators, but to explore how they can interact with AI-authored resources in meaningful ways.
Educators and users can:
Remix – take an existing quiz and modify it through prompts, creating variations or adapting it to new contexts.
Review – evaluate AI-generated quizzes for accuracy, clarity, and pedagogical quality, providing feedback that can improve future outputs.
Curate – browse the searchable library of quizzes, selecting resources that best fit their courses, students, or learning goals.
Generate – submit custom prompts to the Designer Agent, influencing quiz creation while experimenting with different styles, topics, and levels.
Adapt – export quizzes into multiple formats (text, LMS packages, H5P, etc.) and integrate them directly into their teaching practice.
By giving humans these roles, GenOER makes it possible to test not only the scalability of AI-authored OER, but also the ways educators can guide, adapt, and validate synthetic resources in practice.
Licensing Synthentic OER
All resources generated by GenOER are released into the public domain (CC0). This ensures that anyone can freely use, adapt, remix, and redistribute them without restriction.
At the same time, the legal status of AI-generated works is still unclear. Different jurisdictions treat authorship and copyright for machine-generated content in inconsistent ways, and there is no solid consensus on what “licensing” should look like in this context.
For now, CC0 provides a clear signal of intent: everything produced by GenOER is meant to be maximally open. But ultimately, this is a conversation that needs to involve educators, researchers, legal experts, and the OER community at large. The role of licensing in an AI-driven future is less about certainty, and more about ongoing dialogue, review, and shared governance.
Scale & Cost
GenOER shows that AI can generate open resources at scale for very low cost — and as models become more efficient, the economics of autonomous OER production will only improve.
Current Scale (Single Educator + Designer Agent)
| Generation Rate | Quizzes per Hour | Quizzes per Day | Quizzes per Month | Quizzes per Year |
|---|---|---|---|---|
| 1 quiz / 10 min | 6 | 144 | ~4,320 | ~52,560 |
| 1 quiz / 1 min | 60 | 1440 | ~43,200 | ~518,400 |
Cost
| Item | Detail | Approx. Cost |
|---|---|---|
| Quiz generation | ~10,000 tokens per quiz (ideation → creation) | ~$0.01 USD per quiz |
| Annual quiz generation (52,560 quizzes) | ~520 million tokens | ~$520 USD |
| Server hosting | Lightweight VM (optional, can also run locally) | ~$20 USD / month |
Future Scale (Multiple Domain-Specific Educator Agents)
As GenOER expands, multiple domain-specific Educator agents could run in parallel (e.g., math, biology, history, languages). Each agent could simulate a unique teaching perspective, pedagogy, or style of assessment.
| Number of Educator Agents | Quizzes per Year | Total Questions per Year | Approx. Annual Cost |
|---|---|---|---|
| 1 (current) | ~52,560 | ~262,800 | ~$520 |
| 5 | ~262,800 | ~1.3 million | ~$2,600 |
| 10 | ~525,600 | ~2.6 million | ~$5,200 |
| 50 | ~2.6 million | ~13 million | ~$26,000 |
Open Source Code
GenOER isn’t just producing open resources — the platform itself is built on open principles. All of the code used to create and run GenOER is released as open source, built on top of Drupal and extended through custom modules and services.
SynthOER – the core autonomous OER generation framework.
QuizGen – quiz generation engine that handles prompts, question creation, and feedback.
QuizRen – quiz rendering and export tools, making quizzes usable across multiple formats and platforms.
By making the code open and reusable, GenOER ensures that institutions, educators, and developers can experiment, adapt, and extend the platform — further aligning with the values of the OER movement.
Roadmap
GenOER is starting with autonomous quiz generation, but the platform will evolve in ways that expand both functionality and educator involvement. Planned developments include:
Remixing Tools – intuitive interfaces for educators to adapt and remix generated quizzes directly on the platform.
H5P Integration – seamless export of quizzes into H5P for interactive use and embedding across LMS platforms.
Guided Generation – a feedback loop that puts the educator “in the middle of design,” steering how quizzes are generated and refined.
User-Built Educator Agents – enabling users to design their own Educator agents that passively generate OER aligned with their domain, style, or teaching philosophy.
Beyond Quizzes – expanding generation beyond assessments to include lesson plans, activities, and other open educational resources.
This roadmap reflects the project’s goal: not only to scale the production of open resources, but also to create new ways for educators to shape, adapt, and evaluate Synthetic OER.
Community Involvement
GenOER is designed as an open experiment, and its value grows with the community around it. We are actively seeking collaborators who want to shape, critique, and extend this work. Opportunities include:
Researchers – studying the outputs, economics, and pedagogy of AI-authored OER.
Prompt Whisperers – experimenting with prompts to test the boundaries of autonomous generation.
Curators – selecting, organizing, and adapting quizzes into meaningful collections.
Moderators – helping ensure quality, trust, and alignment with open education values.
Educators & Practitioners – integrating GenOER outputs into real courses and providing feedback.
If you are interested in exploring Synthetic OER — whether from a technical, pedagogical, or community perspective — we’d love to hear from you: hello@learnful.io