-->

Super Early Bird Pricing for KMWorld 2026 Available for a Limited Time!
Register NOW for November 16-19. Use code SUPERSAVINGS.

Evolution of Semantic Layers and Knowledge Graphs: Pivotal Components for KM

Article Featured Image

Since metadata is typically organized by silo, collaboration across departments is required in order to reconcile the differences. “The departments should come to an agreement on common vocabularies and concepts,” continued Blumauer. “For one thing, AI agents cannot operate across repositories unless this step is carried out.” In addition, getting agreement on terminology allows human users to obtain consistent answers to their inquiries.

Historically, there has been little collaboration between those in IT, who manage structured data, and those in business units, who manage content. However, one of the lasting benefits of going through the process of developing knowledge graphs is that the two groups start talking with each other. “If they do not, the organization ends up with a shallow semantic layer,” Blumauer cautioned. “A knowledge graph should have more than common metadata schemas. And a shallow layer cannot support autonomous agents that can produce meaningful results.”

Closing In on Convergence

As the need for real-time analytics and agentic readiness has increased, organizations have been trading in their legacy systems for more rapid and flexible ones. PL Developments manufactures and distributes over-the-counter products and other consumer healthcare domestically and globally. Its data warehouse and BI tools were unwieldy, slow, and not ready for AI. In addition, the system lacked the security controls that would allow broader access to analytics by employees throughout the enterprise. The company selected Dremio, an agentic-ready data lakehouse, to replace its legacy system.

PL Developments’ architecture included both a Postgres database for operational data and an Iceberg data lakehouse for large datasets. Dremio’s Intelligent Query Engine allowed users to search both simultaneously. In addition, the new architecture provided the security that PL Developments needed in order to allow access by more workers. Dremio’s semantic layer provided the interface that users needed to work with the underlying data.

An LLM application was implemented that could generate customer communications automatically by combining order histories, cancellation patterns, and production schedules. Food and Drug Administration (FDA) databases could be queried to find FDA warning letters and place products on hold immediately. With Dremio’s Open Catalog, enterprise-level permissions enforced access policies across all the AI and BI platforms. AI Cursor was integrated with Dremio’s MCP server, allowing the AI to access and interact with external data sources. The net result of these changes was a dynamic information ecosystem that could act on its own as well as support workers.

Dremio was founded about a decade ago for the purpose of accelerating the speed of business analytics, first as a federated engine and then evolving into one of the few Iceberg-native data lake engines. “The founders built Arrow, a tool for in-memory analytics, made it open source, and donated it to Apache,” said Rahim Bhojani, CTO of Dremio. “Apache Arrow is now foun- dational to modern analytics platforms. From the get-go, we knew our customers wanted to access and analyze data simul- taneously from multiple databases, so we developed methods of federating and virtualizing the data. With the use of an open table format like Iceberg, there is no need to copy and store the data in a proprietary format; we just fetch the metadata, and the data remains in place. This reduces costs, increases performance, and improves reliability.”

KMWorld Covers
Free
for qualified subscribers
Subscribe Now Current Issue Past Issues