Skip to content

@Ldataengineer walks through OntoSmith, an open-source knowledge-graph builder where a cheap model proposes facts, plain code checks the quotes, and TypeSafe's Jev, as a separate judge, decides what gets written.

Guides

Gul Saeed Khattak @Ldataengineer

EN

  • 6.2k views
  • 122 likes
  • 211 bookmarks
  • 19 reposts
Poster from Gul Saeed Khattak (@Ldataengineer)

About this case

Guide. Gul Saeed Khattak (@Ldataengineer) says most LLM graph pipelines let one model read, find facts and decide they are true. OntoSmith uses a proposer, Claude Haiku 5.5 by default, which attaches the source sentence; a code gate that checks the quote appears word for word and that the fact fits the schema; and TypeSafe's Jev as the judge of what is written. The schema comes first, and a human approves it before anything is built. Neo4j stores the result, and every fact keeps its evidence sentence. He notes v0.1.9, first released on 24 Oct, MIT licensed, that the ontology is a flat list of types rather than OWL, and that the published numbers only hold with Jev as the judge. He says the README points people who want a fast demo toward the GraphRAG family. Links are in a reply, not in this post. Video attached.

Original posthttps://x.com/Ldataengineer/status/2119417274314396148

Related cases