A Neurosymbolic Integration Planning Agent on Neo4j MCP

Session track: Modern Applications

Session time: 07:30 pm - 08:00 pm UTC November 12

Session description:

Turning planning inputs — business goals, evaluation findings, integration patterns, and historical estimates — into a credible technology plan is slow, inconsistent work that teams increasingly hand to an LLM. But a raw LLM produces a plausible plan, not a verifiable one: it invents dependencies, misses prerequisites, and cannot explain its estimates. In this session, Dmitry Izumskiy will show you a neurosymbolic Planner Agent that makes an LLM planner trustworthy by using Neo4j as its knowledge and enforcement layer. Through the Cypher MCP server, the agent reads project documents and structures them into a knowledge graph of assumptions, architecture decisions, work items, and the dependencies that connect them — grounded in historical estimates already in the graph. You will learn how schema introspection and read Cypher keep the agent grounded in real patterns and prior projects; how graph validation checks reject inconsistent dependencies; and how a human reviews and approves each plan before write Cypher persists it. Because the plan is a graph rather than prose, it stays genuinely reviewable — and once approved, it drives timeline and resource planning directly. Dmitry Izumskiy will share the ontology, the Cypher queries, and the validation-and-approval guardrails: a reusable knowledge-management pattern for GenAI applications that yields plans which are validated, not merely plausible.