25 APR 2026
David debugged: Lecture notes generated from Google Slides links with no actual slide content
David discovered that a user's Spanish course lecture notes were created from Canvas page stubs that linked to Google Slides — meaning the agent had nothing real to insert — and investigated why the pipeline failed to handle external links.
Kerra (David's AI study assistant product) generates lecture notes workspaces by ingesting a user's course materials from Canvas. A Notre Dame user's Spanish course produced workspace documents with only thin summaries — no actual slide content — because all 24 course topics were Canvas pages containing only external Google Slides links, not actual files.
David asked:
further, for his spanish class, he clicked on one of the lecture notes generations and went into the worskpace, and none of the documents had any of the lecture slides inserted. they were just summaries. what is going on??? dont we force all the pages of the lecture slides to be isnerted??
The investigation found that Kerra's pipeline classified those topics as page-backed runs, which explicitly banned insert_pdf_pages, because no shared_file_id existed — the actual slide bytes were behind Google Slides links never fetched. David's follow-up question sharpened the diagnosis:
why didnt it download the google slides? and for that course, how much of the content is posted as google slides links or just links to external sites in general vs how much of it is actual canvas topic files?
The deeper audit revealed that 97 file topics in the Organic Chemistry course were similarly unlinked — all 97 had lms_content_id values but zero rows in shared_canvas_files. David probed whether this was a real problem or an invented one:
what is the shared_canvas_files table used for? when and where is it populated? and should it have been populated with those Organic rows? or are we inventing a problem where there isnt one?