Graphwright Publications¶
Three books covering the full pipeline from raw text to a language model that can reason over what that text contained.
The Typed Graph: Naming, Knowing, and Trusting Machine Knowledge¶
The trustworthiness book. How canonical identity, a typed schema, and structural provenance together make machine knowledge defensible. Covers the epistemic commons (MeSH, HGNC, RxNorm, UniProt), the identity server's domain-agnostic core and plugin contract, the entity lifecycle, and the typed graph's central argument: that a finite predicate ontology with declared domain and range makes certain classes of error inexpressible rather than merely discouraged.
Knowledge Graphs from Unstructured Text¶
The extraction book. How to build a knowledge graph from raw documents using LLMs: schema design, the ingestion pipeline, identity resolution, provenance, and diagnostics. Includes the medlit biomedical reference implementation.
BFS-QL: A Graph Query Protocol for Language Models¶
The interface book. The knowledge is in the graph; the LLM can't get to it — this is a book about the missing interface. Argues that SPARQL and Cypher are the wrong abstraction for LLM-driven graph exploration, and that breadth-first traversal with topology-first, metadata-on-demand design is the right one. Covers the five-tool MCP protocol, the working-set framing for context-window efficiency, backends for SPARQL, Postgres/pgvector, and Neo4j, and cross-graph composition via shared canonical identity.
Read in order, the three books cover: ensuring knowledge is trustworthy → getting knowledge in → getting it out.