document-to-facts

What it does

Runs extraction against a schema that's fixed before the first prompt is written, turning a source document into typed entity and relationship nodes instead of a prose summary. The schema decides in advance what shape an acceptable answer takes, so the extraction step has something concrete to check its own output against (this is the write-path move covered in Step 6 of the course).

Inputs

• A source document, or a chunk of one small enough for a single extraction pass.
• A schema describing which entity types and relationship types the graph accepts.

Outputs

• One or more typed nodes representing entities found in the document.
• Typed edges connecting those nodes, each tagged with the document it was drawn from.

Failure mode if skipped

Without a schema to extract against, the default is a free-form summary — readable to a human in the moment, but nothing later can query it, cross check it against another document, or verify a downstream claim against it.

Link to starter kit

Kit: starters/document-to-facts/README.md