Course Materials
An extraction primitive: turn the messy artifacts of a real course — PDFs, slide decks, casebook excerpts, syllabi — into clean structured text that downstream tools can use.
Course Materials is the extraction primitive that sits underneath the rest of the Teaching Tools cluster. The job is narrow and load-bearing: take the messy artifacts of a real course — PDFs, slide decks, casebook excerpts, syllabi, lecture transcripts — and turn them into clean, structured text that other tools can consume reliably.
It’s not a casebook-creation project. It’s the layer that makes casebook creation, virtual-TA retrieval, exam generation, and class-simulation tooling possible in the first place.
What we’re doing
- Maintain the
materials-mdextraction toolchain — a deterministic CLI that converts PDFs, DOCX, PPTX, and HTML to clean markdown. - Handle the cases that matter for legal materials: scanned casebook pages, court opinions in messy formats, slide decks with speaker notes, structured tables of cases.
- Ship the output in a shape downstream Lab tools can use directly — Heron’s retrieval, exam-grader’s reference texts, class-simulations’ problem stems, casebook-builder’s source corpus.
Why this matters
Every other Teaching Tool either works or fails on the quality of its input. Hallucinations in a virtual TA, miscalibrated exam grading, drift in a class simulation — most of these trace back to dirty source text. Getting the primitive right once means the rest of the cluster gets cleaner inputs by default.
Status
Active. The CLI is in production use across Polk’s own teaching workflow and across other Lab skills. Next work: harden the casebook-extraction path against scanned-page edge cases and document the output schema so other Lab projects can integrate against a stable contract.