Establishing a Semantic Layer for GenAI Success with Radiant Advisors and Cube

Generative AI (GenAI) is posed to be an utterly transformative technology that evolves the way business—and undoubtedly, the world—interacts with information. In a proprietary context, rapid access to data and information when and where its consumers need it radically enhances productivity, efficiency, and the quality of decision making.

Yet, despite many desires, GenAI is not the perfect technological boon to save enterprises from siloed information—at least, not yet. Hallucinations—or the presentation of false, though seemingly accurate, information—threaten to devalue any GenAI implementation if not dealt with.

To discuss the ways GenAI can be augmented for true success, John O'Brien, principal advisor and industry analyst at Radiant Advisors, and David Jayatillake, VP of AI at Cube, joined DBTA’s webinar, Delivering on the promise of AI: Increasing Accuracy with a Semantic Layer, offering the ways that a semantic layer—or a consistent convention of understanding and definition that data adheres to—can position GenAI and large language models (LLM) to live up to their technological intent.

O'Brien emphasized that data unification is the umbrella that a semantic layer lives under, accompanied by API access, data cataloging, data governance, and collaboration. Below this, data integration and data persistence layers work to fulfill the data unification layer. 

“These are capabilities that the architecture needs to deliver an enabling platform to the enterprise,” noted O’Brien. “This whole area of unification—we need these distinct features to help deliver business intelligence and reporting on top of warehouses. We need this to help self-service people find, understand, and trust the data.”

The top challenges that a proper semantic layer solves, according to O’Brien, include the following:

  1. Lack of BI tool performance, where clients are implementing additional data extracts tier for localized performance and UX
  2. Unawareness of enterprise metrics, with confusion caused by duplicating and conflicting metrics underscoring the necessity of a comprehensive platform to search
  3. Semantic layers become data silos when handled within BI tools become closed to other access tools

To ensure that the semantic layer addresses each of these challenges, enforcing openness—where business logic is not tied to a specific BI tool, but is closer to the data—serves to unify a data architecture. Semantics are then shared across groups, moving from local domains to core data, delivering a single point of access for everyone.

Ultimately, O’Brien concluded with these three takeaways:

  • OLAP is critical in analytics capabilities, where dimensional analysis of business measures and metrics are crucial.
  • Having an open semantic data layer prevents data silos in BI and reporting tools from forming.
  • Pre-aggregate and cache dimensional data sets to enhance performance and consistency.

Jayatillake established that, “it’s not only humans that need data; now, it’s services and machines that need data too. Organizations have to be proactive to leverage these huge amounts of data that they’ve collected in order to innovate, gain insights, and maintain their competitiveness.”

Between data stack complexity, the proliferation of siloed data due to disconnected data apps, and inconsistent business definitions, enterprises are unable to extract the true value of their data, especially as it relates to GenAI implementation.

With Cube Cloud, Cube aims to offer a model once, deliver anywhere consistency through its universal semantic layer. Cube Cloud unifies—creates a single source of truth with consistent metrics—governs—centralizes and enforces fine-grained access controls—and optimizes—achieves faster, cost-effective results—enterprise data.

Jayatillake further delved into the details of GenAI and how a semantic layer addresses its core challenges.

While LLMs offer an easy, excitable form of information retrieval and ingestion, hallucinations are bad—and in some cases, dangerous—in reality. A semantic layer fundamentally works to mitigate against hallucinations because it knows about the world that you have defined, and declines retrieval when asked about something unrelated to its domain.

Using GenAI to request data from a semantic layer ultimately invites a level of transparency and consistency needed to ensure that GenAI applications fulfill their design intention, explained Jayatillake.

For the full discussion of semantic layers and GenAI, featuring example architectures, demos, and more, you can view an archived version of the webinar here.