NODES AI 2026 - Neurosymbolic Architecture in Production with Neo4j, GDS, Node2Vec & LLM Grounding

In this session at NODES AI, João Cunha explores the implementation of a neurosymbolic architecture in Kipon’s Neo4j-based performance management system. He details how complex domains involving people, skills, tasks, and projects are modeled as a graph, enabling efficient multi-hop queries and rapid data retrieval. The talk covers the use of algorithms such as Jaccard Similarity and PageRank through AuraDS to determine profile similarities and community detection, along with Kipon’s novel hybrid embedding pipeline that combines Node2Vec for structural embeddings and LLM for semantic understanding. Additionally, João discusses the role of a proprietary RDF-inspired ontology in grounding LLM inference, enhancing factual consistency, and supporting explainability within the graph schema. He will also share insights on data modeling, engineering choices, and best practices for building robust neurosymbolic AI systems in a production environment.