kakomon-university-rebuild

Automating university past exam PDF ingestion requires discovering official PDFs, downloading and validating them, splitting bundled documents by subject, and ingesting them into a DB and R2 repository. This workflow handles each phase end-to-end while preventing data collisions.

Splitting bundled exam PDFs into individual subjects uses Gemini classification to identify subject-level granularity, followed by manual review to ensure accuracy. This process prevents data collisions by managing subject variants during the ingestion phase.

You can perform a dry run before ingesting past exam PDFs to execute non-destructive safety checks. This validates the downloaded files and subject splitting logic without writing any data to the database or R2 storage.

Subject_variant management prevents data collisions during PDF ingestion by assigning unique variants to split subjects and generating distinct R2 paths. This ensures multiple exams for the same subject do not overwrite each other in the repository.

Processing each university separately is required to avoid cross-contamination of exam data. The workflow executes a complete rebuild for one university at a time, managing subject variants and R2 paths independently for each institution.

Ingesting university exam PDFs requires satisfying frontmatter metadata requirements during the phase-based workflow. This ensures each split document is properly classified and registered before being committed to the database and R2 storage.