auto-prime

To convert unprocessed client documents into RAG priming files, the system reads PDFs, HTMLs, and Word drafts, extracting structured knowledge with provenance. It then generates per-document priming files containing frontmatter, TL;DR, key entities, constraints, and evidence.

RAG priming and evidence tagging support PDFs, HTMLs, and Word drafts. These unprocessed documents are automatically read and classified by type, such as textbooks, curricula, inserts, and deliverables, to ensure accurate knowledge extraction.

Knowledge extraction ensures no fabrication by tagging evidence directly from the source material. It captures provenance through standardized frontmatter, including source, type, domain, processed date, and confidence level for each extracted entity and constraint.

The best way to structure extracted knowledge for calibration digests is to save per-document priming files under references/priming-rag/ and update calibration-digest.md. This approach reports processed counts and new calibration levels for validation.

Yes, you can use it for textbook and curriculum document processing. It automatically classifies these document types and generates corresponding priming files with standardized frontmatter, ensuring structured knowledge extraction without fabrication.

Your document processing pipeline needs standardized frontmatter, including source, type, domain, processed_date, and confidence, to maintain provenance and ensure reliable RAG priming. This metadata enables accurate validation and prevents knowledge fabrication.