AI-Driven Data Modeling with Tableau Semantics: A Principal Architect’s Perspective

Editor's note: This interview was originally featured on the Salesforce Engineering blog as part of the “Engineering Energizers” Q&A series, which highlights engineering leaders who are shaping the future of technology.

Lior Ebel, Principal Architect of Software Engineering at Salesforce, leads the architecture of Salesforce’s semantic layer, Tableau Semantics. This advanced engine enables businesses to query and model data with unparalleled efficiency and accuracy. Tableau Semantics powers Data Cloud, Agentforce, and the agentic analytics platform Tableau Next.

In this interview, you’ll discover how the team addresses challenges including optimizing large-scale semantic queries, integrating AI for intuitive data modeling, and ensuring seamless scalability without compromising trust and security—all contributing to smarter, data-driven decision making.

What is your team’s mission?

Lior: Our team empowers businesses with a unified and intuitive way to model and analyze their data. We focus on advancing Tableau Semantics, the core data querying and modeling engine for Tableau Next. Our primary goal is to simplify how organizations define data relationships and business metrics, creating a single source of truth for consistent and reliable insights.

Through Semantic Data Models (SDMs), users can structure their data for high-level, intuitive queries. This approach reduces the complexity of traditional SQL and accelerates the development of data-driven applications. The semantic layer query engine translates these queries into efficient SQL based on SDM definitions.

Here is a high-level example of what such a translation may look like:

Leveraging cutting-edge technology like generative AI, we integrate AI-driven agents and natural language capabilities. This enables users to query data conversationally without needing technical expertise. Our ultimate goal is to make data accessible, actionable, and valuable while ensuring a seamless, intelligent, and scalable analytics experience.

What was the most significant technical challenge your team faced recently?

Lior: One of our most significant technical challenges was ensuring that semantic layer queries perform efficiently at a B2C scale, where vast amounts of data must be processed quickly and accurately.

To address this, the team chose a SQL generator-based approach for the semantic layer query execution engine instead of building our own query processor. This strategic decision allows us to leverage the scalability and optimization features of underlying data warehouses, avoiding the need to reinvent performance tuning.

We also emphasize optimizing the SQL queries generated by the semantic layer. By using efficient joins, applying early filtering, and minimizing query complexity, we follow best practices in SQL design. Additionally, we test and refine SQL patterns early in development, proactively identifying and resolving performance bottlenecks.

This dual focus on architecture and SQL optimization enables us to handle large-scale B2C data, delivering fast, reliable, and scalable analytics.

How do you manage challenges related to scalability?

Lior: Scalability presents several complex challenges, including growing data volumes, increasing customer usage, rising query loads, and expanding teams interacting with the system. Each of these requires targeted strategies to address effectively.

To manage system load, we employ a two-pronged approach: top-down and bottom-up. During the design and planning phases, we identify and mitigate potential bottlenecks through collaboration and by leveraging our team’s expertise. This proactive top-down strategy is complemented by stress-testing during development. Simulating high loads helps us uncover bottlenecks that may not have been evident earlier, ensuring we’re ready for unforeseen scaling challenges.

To support the growing number of teams interacting with the system, we prioritize stability and usability. Thorough documentation, intuitive APIs, and a self-service design enable both internal and external teams to scale their usage independently. This reduces the need for direct support, allowing for seamless scaling while maintaining performance and usability standards.

What ongoing research and development efforts are aimed at improving your project’s capabilities?

Lior: The team has been focused on integrating AI into the semantic layer to create more intelligent and scalable data experiences within Salesforce. Over the past year, our efforts have centered on leveraging Semantic Data Models to support autonomous and assistive agentic experiences. By grounding these AI tools in an organization’s specific business definitions, we ensure that responses to data queries are consistent, accurate, and aligned with unique metrics, thereby reducing issues like LLM hallucinations.

AI is also being utilized to improve the process of creating and refining semantic models. AI-powered assistants can suggest best practices for defining business logic and recommend efficient data model structures, streamlining SDM development as complexity increases. As an example, Tableau Semantics may suggest a formula for calculating your ROI in response to a free-text request like “create an ROI field please.”

Positioning the semantic layer as an AI-powered foundation enables smarter decision making. Advancing these capabilities ensures reliable, intuitive experiences across the Salesforce ecosystem, empowering businesses to make data-driven decisions with greater confidence.

How do you balance the need for your project’s rapid deployment with maintaining high standards of trust and security?

Lior: Balancing rapid deployment with high standards of trust and security is a core priority, especially given the nature of Salesforce’s customer base. We integrate trust and security into every step of the development process, supported by Salesforce’s robust foundational infrastructure. This foundation allows us to deploy quickly without compromising the critical elements of security and trust.

The semantic layer is built on Hyperforce, which abstracts much of the complexity around security and compliance. By leveraging this infrastructure, we avoid reinventing security measures, enabling both speed and reliability. A solid infrastructure forms the backbone of maintaining high security standards.

Beyond infrastructure, rigorous quality assurance and validation processes are crucial. Every release undergoes extensive testing to ensure it meets the highest standards. This combination of trusted infrastructure and diligent testing allows for rapid development while keeping security and customer trust as top priorities.

How do you gather feedback from users and how does it influence future development?

Lior: The semantic layer serves two primary user groups: external users who interact with it as a product, and internal users—Salesforce developers who use it in their workflows.

For external users, the product team engages in ongoing conversations with key stakeholders to gather feedback and address pressing pain points. This ensures the product aligns with user needs.

Internal users (e.g. business Intelligence app developers) provide feedback through open communication channels, such as Slack for asynchronous input and face-to-face meetings as needed. Software development is inherently human-centered, so understanding the challenges faced by both developers and end users is crucial for building effective solutions. Often, the most compelling technical challenges come directly from user pain points, as these represent problems users can’t solve on their own.

Development priorities are heavily influenced by user feedback to enhance the product’s real-world utility. However, proactive innovation is also critical, as it allows us to anticipate future needs and push boundaries beyond existing requests.

Learn more about Tableau Semantics 

Tableau Semantics enriches data and AI with business context to ensure accurate insights, streamlined data modeling, and simplified metric governance for smarter decisions at scale. Dive into the features and benefits of Tableau Semantics to learn more.