Exploratory Proposal · LMS Content Pipelines · SUNY Polytechnic Institute
Large language models are increasingly capable of reviewing and revising structured documents. One potential application in higher education is using AI to analyze course packages exported from learning management systems.
D2L and many other LMS platforms support the IMS Common Cartridge (CC) standard, which packages course content into a portable format containing HTML pages, XML manifests, assessments, and supporting files.
The proposed idea is to build a workflow that allows course cartridges to be exported, interpreted as structured content, reviewed or revised with AI tools, and then rebuilt for re-import into the LMS.
The intention is not to automate course design, but to create a human-in-the-loop review pipeline that could support instructional design teams and faculty.
The proposed workflow follows a structured pipeline from export through re-import:
| Stage | Description |
|---|---|
| 1. Course Export | Course exported from D2L as a Common Cartridge (.imscc) file. |
| 2. Cartridge Extraction | The cartridge is unpacked into its underlying files: manifest XML, HTML pages, assessment XML, and linked resources. |
| 3. Structured Interpretation | Content is parsed into a structured representation of modules, pages, assignments, and resources. |
| 4. AI Review | AI systems analyze course content and suggest revisions or improvements. |
| 5. Reconstruction | Revised content is written back into the course structure. |
| 6. Cartridge Rebuild | The package is rebuilt into a valid .imscc file. |
| 7. LMS Re-Import | The updated course package is imported back into D2L for faculty review. |
If implemented carefully, this workflow could support several instructional design tasks:
Identify outdated references, broken links, or unclear explanations in older courses.
Flag missing alt text, unclear headings, or readability issues.
Generate draft discussion questions, summaries, or formative prompts for existing content.
Review for formatting consistency across modules and pages.
Generate module summaries or study guides from existing materials.
Convert course exports into human-readable documentation for program review or accreditation.
Important technical and policy limitations must be acknowledged before attempting a system like this.
Common Cartridge files depend on precise XML relationships between resources. Incorrect edits can easily break cartridge imports.
Quizzes and tests rely on QTI XML formats that vary between LMS platforms. Automated editing of assessments could introduce errors.
Although Common Cartridge is a recognized standard, LMS platforms frequently add proprietary extensions that complicate automated manipulation.
LLMs generate plausible language but may introduce inaccuracies. All AI-generated revisions require human review before adoption.
Course materials may contain licensed or copyrighted content. Any AI workflow must align with institutional policies regarding FERPA, copyright, and AI governance.
If there is interest in exploring this concept, a small prototype could test the workflow before any broader commitment:
Build a parser that safely extracts content from a D2L cartridge.
Run AI analysis on HTML course pages only.
Rebuild a valid cartridge without modifying structural metadata.
Test import into D2L and review results with instructional designers.
This staged approach allows feasibility testing before considering broader adoption.
Before pursuing development, it would help to know whether similar work is already underway across SUNY:
If work is already happening in this space, coordination could help avoid duplication and accelerate development.