Neo4j for Enterprise AI: Architecture & Integration

from neo4j import GraphDatabase
import os

NEO4J_URI = os.environ.get("NEO4J_URI", "bolt://localhost:7687")
NEO4J_AUTH = (os.environ.get("NEO4J_USER", "neo4j"), os.environ.get("NEO4J_PASS", "password"))

driver = GraphDatabase.driver(NEO4J_URI, auth=NEO4J_AUTH)

def execute_graphrag_hybrid_query(query_vector: list[float], top_k: int = 5):
  """Hybrid GraphRAG query combining HNSW vector search with 2-hop Cypher traversal."""
  cypher_query = """
  CALL db.index.vector.queryNodes('document_embeddings', $top_k, $query_vector)
  YIELD node AS doc, score
  MATCH (doc)-[:MENTIONS]->(entity:Entity)-[rel:RELATED_TO]->(target:Entity)
  RETURN 
      doc.title AS document_title,
      score AS similarity_score,
      entity.name AS entity_name,
      type(rel) AS relationship_type,
      target.name AS target_entity
  LIMIT 20
  """
  with driver.session() as session:
      result = session.run(cypher_query, query_vector=query_vector, top_k=top_k)
      subgraph_facts = [record.data() for record in result]
      return subgraph_facts

if __name__ == "__main__":
  dummy_vector = [0.021] * 1536
  facts = execute_graphrag_hybrid_query(dummy_vector, top_k=3)
  print(f"Retrieved {len(facts)} GraphRAG entity relationships from Neo4j.")