Jesús Barrasa is AI Field CTO at Neo4j, where he develops practical approaches to combining large language models with knowledge graphs and explicit domain models. His work addresses a persistent weakness of enterprise AI: retrieving relevant text without understanding the entities, relationships, and organizational context that give it meaning.
Barrasa holds a Ph.D. in artificial intelligence and knowledge representation. Before joining Neo4j in 2015, he worked at Denodo and Ontology Systems; he subsequently led Neo4j telecommunications initiatives and its solutions-architecture team across Europe, the Middle East, and Africa.
He led development of neosemantics, a Neo4j plugin connecting property graphs with Semantic Web standards such as RDF and SHACL. Its capabilities include importing ontologies, mapping graph models, and validating data against explicit constraints, allowing companies to reuse established vocabularies alongside operational graph databases.
With Jim Webber, Barrasa coauthored Building Knowledge Graphs. He also coauthored Knowledge Graphs: Data in Context for Responsive Businesses with Webber and Amy E. Hodler, and cohosts Going Meta, maintaining code and examples for the series.
- Ontologies as domain playbooks. Barrasa uses implementation-independent models of entities, relationships, and hierarchies to give structured-data pipelines and document-extraction systems a shared vocabulary. Financial ontologies, for example, can encode corporate categories and governance relationships before ingestion begins.
- Schema-guided knowledge-graph construction. Structured records and unstructured documents are mapped into the same conceptual model. Document chunks remain linked to extracted entities, preserving connections between domain knowledge and its source material.
- GraphRAG beyond vector similarity. Vector search identifies an initial semantic match; graph traversal then recovers connected people, documents, events, and other context that similarity rankings alone can miss.
- Ontology-driven retrieval. Barrasa stores the ontology within the graph and uses it to determine which relationships a retriever follows. In his ontology-guided GraphRAG demonstration, changing contextual relationships alters retrieval behavior without rewriting the underlying Cypher query. He has extended these concerns into graph-backed agents, agent memory, text-to-Cypher, and evaluation.
