Dual Database Architecture
The graph answers connection questions, but the same fleet emits hundreds of thousands of sensor readings, better crunched as columns than traversed as nodes. Pair the graph with a columnar analytics store and route each question to the store that answers it best.
Each store handles the workload it is best at, and a supervisor on AWS Bedrock AgentCore routes each question to the right one.
|
Columnar analytics store |
Neo4j Aura |
| Stores |
Tables and files |
Nodes and relationships |
| Best at |
Hundreds of thousands of sensor readings; SQL aggregations and trends |
Topology, component hierarchies, maintenance; multi-hop traversal without JOINs |
| Answers |
"How much?" and "How often?" |
"How is this connected?" and "What is affected?" |
| AI capability |
SQL analytics, foundation models |
Vector indexes, GraphRAG, MCP |
AWS Glue with the Neo4j Spark connector moves data between the stores. Numeric questions go to the analytics agent, relationship questions to the Neo4j agent, so the two stores stay connected, not siloed.