GraphRAG in Motion: Real-Time Knowledge Graphs for Agentic AI on AWS

Session track: AI Engineering

Session time: 01:00 am - 01:30 am UTC November 13

Session description:

Most knowledge graphs are static — rebuilt nightly from batch pipelines.Agentic AI demands graphs that reflect the world in real time. Harshit Kohli, Senior Technical Account Manager at AWS, will show you how to architect a continuously synchronized Neo4j knowledge graph powered by real-time streaming data from Amazon MSK and Apache Flink. You will learn three production-tested patterns: how to stream entity updates and relationship changes from operational systems into Neo4j using CDC pipelines with sub-minute latency; how to design Cypher queries that replace expensive multi-hop API calls inside agentic workflows built on Amazon Bedrock; and how to solve the hardest streaming graph problems — entity resolution, relationship deduplication, and schema evolution — at production scale. The session will walk through a live end-to-end pipeline where a Flink application extracts entities from a streaming event feed, resolves them against existing graph nodes, and writes new relationships to Neo4j in real time. An Amazon Bedrock agent will then query this live graph to assemble relationship-rich context — reducing what would require 8–12 sequential API calls to a single low-latency Cypher query. You will leave with working Cypher patterns, a Flink-to-Neo4j connector configuration, and a streaming graph data model template you can apply immediately to your own agentic AI workloads.