Systems Engineering
AI Course Authoring System
Draft and transform learning materials into Canvas-ready courses with WCAG 2.1 AA and ASU Design standards embedded.
EdPlus at ASU
- Python
- Claude
- Canvas
Measured
3 to 4 wks to 2 to 3 days
Course build time when curriculum is supplied.
Measured
8 of 8
Pilot cohort users rated it 4 or 5 out of 5 for satisfaction and for speed and quality of work.
Measured
28 skills
Covering needs analysis, building, assessment, accessibility remediation, staging, and alignment reporting.
The Friction
A course build ran three to four weeks. Standards were applied by hand at the end, which is the point at which fixing them is most expensive, and the checks that mattered most, accessibility and alignment, were the easiest to defer. Designers were also still editing in the live course because there was nowhere safer to work.
What I Did First
This is the production successor to the Canvas Course Authoring Workbench. The interface questions had already been answered by interviewing four designers about how they actually build; what was unresolved was everything around the editor. Standards had to be enforced while someone was authoring, when a fix is cheap, and nothing could reach a live course without a review step.
What I Built
A Claude Code and Codex plugin: 28 skills, 112 Python scripts, 17 standards documents, and 8 page templates. The WCAG 2.1 A and AA static subset runs on every content surface, pages, quizzes, assignments and discussions, as additive findings that never block a push. A three-tier course level system detects Introductory and Advanced from the course code and calibrates Bloom's ranges, scaffolding, and rubric descriptors to match. All HTML body changes stage before push, with preview and rollback. Separate skills validate CLO, MLO, assessment and material alignment, and score whether an assessment strategy can still certify learning when students have AI.
What Came of It
Course build reduced from three to four weeks to two to three days when curriculum is supplied. In a pilot cohort survey, all 8 respondents rated it 4 or 5 out of 5 for overall satisfaction and for improving the speed and quality of their work. It is distributed as a versioned plugin with a contribution process scoped to ASU colleagues, and contributions have started arriving.
Scope and Limits
Still in beta. The pilot cohort is small and its time figures are self-reported. Confidence in the correctness of generated output scored lower than every other dimension, which is where the next work goes. Assessment assurance scoring produces a judgment. It certifies nothing.
More systems, and the learning experiences behind them.
or write directly to brent.michael670@gmail.com