Self-Learning Agentic Workflows: Using Neo4j to Build a Reusable Network Validation Layer

Session track: Modern Applications

Session time: 04:30 am - 05:00 am UTC November 12

Session description:

Why let your AI agents reinvent the wheel every time a system configuration needs validation? While LLM agents are brilliant at dynamic problem-solving, they lack an innate, structured memory to reuse previously validated execution steps across varying product versions. ​Enter the Neo4j-backed Validation Layer. In this session, we will demonstrate how to move away from open-ended agentic research and toward structured graph traversal. We will explore how to model system components, product versions, validation commands (CLI/UI/API), and agentic sub-tasks into a highly optimized Knowledge Graph. This graph serves as an operational harness: bifurcating complex issues, dynamically spinning up specialized sub-agents on demand, and pinning their execution to verified historical data. ​Key Takeaways: ​1. How to replace costly LLM search steps with efficient Neo4j path finding for system validation. ​2. Designing a schema where nodes represent environment states and edges represent validated transitions or agent skills. ​3. Strategies for scaling a multi-agent system’s memory so it becomes smarter and faster with every command it executes.