data
Splits paper full text into overlapping semantic chunks for embedding, retrieval, and citation-anchored answers.
Paper Chunker is the upstream of every retrieval-augmented surface. It splits the full text of an exhaustively-extracted paper into overlapping semantic chunks, each carrying a section header, a page anchor, and the embedding model version. Signal Canvas and the Research Kernel retrieve from the chunker output so every cited claim resolves to a chunk in the original paper.
Source: curated glossary catalog. Freshness: git_versioned_curated_catalog.
Term API · apps/web/data/glossary/terms.ts