How to Build the Next Wave of Graph-Based Agentic AI

Session Track: Graph Memory & Agents

Session Time: 07:30 am - 08:00 am April 15

Session description

We are entering the “Graph-Based Agentic Era”—a shift where AI moves beyond simple chat to autonomous reasoning, retrieval, and action. But for agents to be truly reliable, they need more than just a Large Language Model; they need a structured brain.

In this session, we will explore how to overcome the critical bottlenecks of current GenAI agents: hallucinations, lack of memory, and “black box” reasoning. We will dive into the architecture of the next wave of AI, where Knowledge Graphs provide the essential scaffolding for agents to reason with precision.

Drawing on benchmarks from Lettria’s new Perseus model, we will demonstrate why the standard “text-to-graph” approach is failing and how a fine-tuned, ontology-driven strategy changes the game. You will learn:

1. The “Text2Graph” Challenge: Why generic LLMs struggle to build reliable graphs and how schema-constrained extraction achieves 99% output validity.

2. Speed vs. Accuracy: How to reduce latency from seconds to milliseconds (key for real-time agents) while outperforming GPT-4 and Claude in entity extraction F1 scores.

3. Building the Stack: A practical look at combining Document Parsing, Ontology Enforcement, and GraphRAG to create agents that are transparent, verifiable, and enterprise-ready.

Join us to learn how to stop building chatbots and start building trustworthy, graph-powered agents.