Graph Exchange May 2025: AI Agents and Knowledge Graphs
Graph Exchange is an innovative mini-conference on Graph-powered AI Agents, Knowledge Graphs, GraphRAG, and all ways where graphs underpin the modern GenAI stack. Brought to you by Neo4j, the category-defining graph database, we’ll review key use cases for graphs and bring together the leading graph experts.
Join us at the iconic San Francisco hub for half a day on May 21!
Doors open at 2:30 pm.
We start with technical talks on the core concepts:
Graphs and MCP Servers
Knowledge Graphs in industry, from construction to GraphRAG
Agentic ASI with Graphs
Global data infrastructre for AI
The talks will be followed by a panel setting topics for community discussion and work. After breakout sessions we’ll reconvene for the final panel, reviewing the best practices for a complete end to end OSS AI Knowledge Stack.
Agenda:
3:00 pm - 3:20 pm: Alexy Khrabrov, Neo4j
Grand Wecome
Open Agentic AI Knowledge Stack introduction
3:20 pm - 3:40 pm Jure Leskovec, Stanford & Kumo
Foundation Models for Graph Learning
4:00 pm - 4:20 pm Rishi Puri, NVIDIA
GNNs & LLMs in PyG
This talk will cover how Graph Neural Networks can be used to enhance LLMs using PyG to improve accuracy for RAG like tasks across any kind of data domain. This will include examples on real world data. We will also cover how LLMs can be used to enhance GNNs for graph machine learning tasks.
Rishi Puri graduated from UC Berkeley and is a lead engineer for the Deep Learning FrameWork PyG at NVIDIA. He is also a core contributor to the open source PyG framework and community. His main focus is researching how to combine state of the art graph and language modeling techniques. He enjoys teaching about this work at Stanford, conferences, webinars, and through the PyG Slack and LinkedIn communities.
4:20 pm - 4:40 pm: Willem Pinaar, Cleric
The Graph is Not the Territory
At Cleric we use service graphs to diagnose production incidents. However, production systems evolve as services redeploy, configurations change, and dependencies shift—making perfect system representation impossible. We've been building knowledge graphs that acknowledge this limitation while still providing value during incident investigation. These graphs combine infrastructure data, monitoring information, and operational knowledge to help guide causal reasoning.
I'll discuss the technical challenges in graph maintenance, our approaches to handling graph drift, and how we ground graph based hypotheses through direct system interaction. We'll examine cases where imperfect graphs accelerated root cause analysis and others where they misled us.
Willem is the Co-Founder and CTO of Cleric, an AI Site Reliability Engineer that autonomously investigates and resolves production issues. He also created the Feast Feature Store, an open source project widely adopted for ML feature management. Prior to Cleric, Willem was a Principal Engineer at Tecton and led the ML Platform at Gojek
4:40 pm - 5:00 pm: Nikita Ivanov, Humatron
From Agentic RPA to Self-Leaning Autonomous Agents
Discussion around the ongoing active transition from the initial agentic architecture with hardcoded agentic workflows (RPA) to reasoning LLMs, MCP, recursive self-learning and autonomous working.
Nikita Ivanov is a hands-on technology executive with over 25 years of experience. Product leader with extensive experience in developing new HPC data management and GenAI/LLM/Agentic products, raising capital, and nurturing PLG throughout all stages of the business. Founder/CEO of GridGain Systems, DataLingvo and Humatron AI. Advisor to multiple AI startups, frequent international speaker and Forbes Technology Council member, active open source contributor, and Java/Scala founding community member.
5:00 pm - 5:20 pm: Francisco Ferreira, DataLinks, Tabular Graphs
Lets chat about how AI agents can use automated graphs as their user experience layer to collaborate, where this lives in their agentic workflows, and what types of Agents are surfacing in today's enterprises. Let's also explore how to make Knowledge Graphs and Ontologies become the shared memory layer that allows for humans and AIs to work together!
Francisco Ferreira founded DataLinks after nine years at Palantir, solved complex data challenges across various industries. He's now focused on bringing the world's data together by automatically building graphs that serve as a foundational data layer for enterprises in the era of AI agents.
5:30 pm - 7 pm: Breakout Sessions
We'll add talks and sessions on the topics below.
MCP/A2A
AI Memory
GraphRAG
7:15 pm - 8:15 pm: Panel and plenary discussion
Leaders of breakout sessions join the panel and summarize their findings for the best ways to build AI applications.
8:30 pm: closing
Don't forget to add the event to your calendar! See you soon!