Building an AI Customer Support Agent with Mastra & Neo4j

Session track: AI Engineering

Session time: 06:30 am - 07:00 am UTC November 12

Session description:

Enterprise support AI has a reasoning problem, not a data problem. An agent might know a customer filed three tickets this month. But sometimes it doesn’t trace them back to the same defective component or notice that a similar case was resolved six months ago through an undocumented escalation path. The gap between data and connected reasoning is where most support agents hallucinate. As artificial intelligence shifts from passive chatbots to autonomous agents making business decisions, they risk hallucination and "context rot" and may also lack explicit guardrails. A Context Graph, built on Knowledge Graphs, is a structured memory layer for AI Agents. It can prevent gaps by providing a "semantic backbone." Instead of generic data or vector searches, an agent queries the context graph. It can then evaluate precedents, follow decision trees, and apply specific business rules (e.g., “we always give healthcare companies a 10% discount”). It also helps agents navigate gray areas safely. In this session, Abhishek will combine Mastra, an Agentic AI framework, with Neo4j AuraDB. You will see how an agent can access a live context graph via the Neo4j MCP server, giving LLMs relationship-aware context across customers, tickets, products, and resolutions. Explore how a traditional knowledge graph evolves into a full context graph. Whether you're new to knowledge graphs or already shipping agents, you'll leave with concrete patterns for building production-ready AI that reasons over connected data. Learn to model a support domain as a context graph in Neo4j AuraDB, wire it to an agent via MCP, and use Mastra to bring it to life.