Join Us on Nov 6 for 24 Hours of Live Sessions at NODES 2025 | Register Today
Session Track: AI Engineering
Session Time:
Session description
LLMs struggle to reason over large private graphs. While recent advances in agent frameworks equip LLMs with external tooling, there is still no agent capable of selecting and executing a variety of graph algorithms for complex reasoning tasks. In this session, we will present a graph algorithms agent (MCP server) that can execute algorithms from the Graph Data Science (GDS) library on Neo4j. We will demonstrate how users can ask natural language questions that implicitly require graph algorithms to answer. We will demonstrate how users can collaborate with the agent to solve more complex graph tasks. We will also walk through the implementation of the GDS agent, discuss key challenges, and highlight future opportunities for the community. The server can be used off the shelf with any Neo4j database that includes GDS.
Senior Software Engineer, Neo4j
Brian has been working on graph algorithms, graph machine learning, and RAG on graphs with LLMs. His recent work includes building agents that reason over graphs by using graph algorithms as tools and leveraging methods from graph ML and algorithms to improve RAG on graphs.