Independent research: GraphRAG makes AI agents 80% more truthful | Read the report

NODES 26 — November 12, 2026

GDS Agent for Advanced Graph Algorithmic Reasoning

Session track: Data Intelligence

Session time:

Session description:

Frontier LLMs and agents excel at question answering, reasoning and coding tasks. When equipped with the Graph Data Science (GDS) library via tools, they can execute graph algorithms. However, this alone does not make the agent an autonomous graph data scientist. In this session, we will present how a GDS agent is able to solve graph reasoning tasks that would have required weeks of effort, beyond simple Q&A. This gives users an actual graph data scientist to collaborate with. We will walk through how it's built and highlight benchmark results, discuss key challenges and future opportunities for the community. The agent is open-sourced can be used off the shelf with any Neo4j database.

Speaker

photo of Brian Shi

Brian Shi

Lead Software Engineer, Neo4j

Brian works on graph algorithms, graph machine learning and the intersection with agents. His recent work includes building agents that reason over graphs by using graph algorithms & ML, for simple Q&A, long-running tasks and open-ended problems.