Unleashing the Power of Gen AI to Manage Live Systems
Applying generative AI to live systems, for example, to monitor a telecommunications or transportation network, requires both timely information and an understanding of the system producing it. Today’s retrieval-augmented generation (RAG) relies on traditional data sources, such as databases that cannot provide live telemetry at scale, or on streaming platforms, which lack descriptive information about the system. As a result, AI lacks the continuously updated, contextualized view of a live system needed to effectively monitor and analyze it.
ScaleOut Digital Twins™ addresses that gap by creating a fast, scalable data layer for AI RAG that eliminates bottlenecks in accessing live data. It uses an object-oriented, digital twin model hosted in memory to track all system components in real time. This gives AI RAG instant access to live data by interrogating component properties within the digital twin. The in-memory digital twin is hosted on a cluster of servers and scales as needed to match the size and complexity of a live system. This ensures that data access remains fast for very large digital twins with thousands or even millions of digital twin objects.
Starting with version 7, digital twins can now incorporate metadata that describe the overall model, individual digital twin objects, and their properties. This metadata includes details about property units and expected min/max values. This rich metadata gives AI RAG deep context about the system’s structure and its intended behavior to aid its analysis.
By examining a digital twin, AI RAG has immediate access to both the state of a live system and its structure. These exciting new capabilities enable generative AI reason about complex, live systems and identify emerging issues or opportunities as they assist personnel in managing these systems.
For example, consider the use of a digital twin to manage a large data center. The digital twin uses in-memory objects to track individual components within the data center, such as servers, routers, switches, racks, and power supplies. It periodically receives telemetry from these components and maintains contextual information, such as model numbers, specifications, and connections to other components. Starting with version 7, each digital twin object can also host metadata that explains the role of each component, its expected behavior, and the structure of the component hierarchy. As shown in the diagram below, a generative AI model can use the digital twin’s scalable, in-memory compute platform to instantly access all this information and reason about health of the live system:
