← back
The Knowledge Graph Mullet: Trimming GraphRAG Complexity - William Lyon
Takeaway
Hybridizing property-graph ergonomics with RDF-triple storage (Dgraph) lets GraphRAG combine vector, geospatial, and image entry points into a single knowledge graph.
Summary
- William Lyon proposes the 'knowledge graph mullet': property graph in the front (nodes/labels/properties, Cypher-like query), RDF triples in the back (scalable storage, semantic web).
- Demos Dgraph (open-sourced 2017 by ex-Google Spanner-graph team) which uses property-graph model with RDF triple storage and posting-list optimization indexed by predicate.
- DQL is GraphQL-inspired: root-criteria starting point + nested selection set that doubles as graph traversal; returns JSON shaped like the query.
- NYT news graph example: articles, authors, topics, orgs, geo, images, paragraph-as-chunk with vector embeddings as node properties — vector search becomes one of many sub-graph entry points alongside geospatial and image-similarity entry points.
graphragdgraphknowledge-graphs
Original description
There are typically two approaches to working with graphs: property graphs and RDF. These systems are often thought of as different knowledge graph paradigms optimized for different workflows. This talk examines how combining property graph interfaces with RDF triple storage creates an optimal foundation for GraphRAG systems. We'll show how to build and use knowledge graphs using the Dgraph graph database and how knowledge graphs are the foundation of building AI Agents. Resources: * Dgraph docs: https://docs.hypermode.com/dgraph/overview * Hypermode: https://hypermode.com * hyper-news GitHub repo: https://github.com/johnymontana/hyper-news * Hypermode Agents early access: https://hyp.foo/agents