Schedule

We are used to managing data before deploying AI: carefully collecting, cleaning and structuring it. But that is changing. AI now helps to improve data itself: automatically enriching, validating, integrating and documenting it. We are moving from static management to dynamic improvement: AI brings data to life and changes how we deal with it

Topics and discussion points:

• Data becomes a living, self-learning system through AI.
• Errors and missing data can be automatically detected and corrected.
• Data sources will soon merge automatically.
• Classification, documentation and compliance are increasingly supported in real time.
• The promise of less manual work seems within reach, but how far are we really from achieving it?

Hallucinations from AI can destroy trust in BI outputs. This live, technical session walks through building an LLM-powered analytics assistant that only answers from governed, verified data. Using Snowflake Cortex Semantic Models, Cortex Analyst, and Cortex Search, we’ll map business terms to actual definitions, auto-generate safe SQL and trace every step for auditability. You’ll see the full stack in action, with architecture diagrams and code patterns you can implement.

Key takeaways:

• How to ground LLMs in your semantic layer
• How to integrate text-to-SQL safely in BI workflows
• How to make AI outputs traceable and defensible
• How to connect unstructured docs and structured data in one system
• How to design observability into AI so you know when it fails.

Semantical context is needed for both human and AI users to understand what the data they’re seeing is about and how they should use it. Semantic layers are the technological solution our engineering teams currently like to propose as a solution. But how much of the work we do around semantic layers actually has to do with semantics? And what about all the other data in the organization that never goes through your semantic layer technology? While the intent of semantic layers and the modeling involved is good, many implementations focus on the technologies and fail to consider the actual semantics. We will explore this seemingly paradoxical problem and consider a path forward, toward real business context being applied not only to individual datasets, but at enterprise scale.

Topics and discussion points:

• Current role of the semantic layer as query abstraction
• Where do we get the semantics from – the ugly truth
• Semantic modeling maturity curve from cargo cults to universal semantics
• Discovering actual business context with the help of conceptual modeling
• Semantics beyond a single layer – linking between solution-level datasets and enterprise-level semantics
• Semantics beyond analytics – AI use cases and more.

Many organizations struggle with complex data challenges. Examples include tracking data usage (both transactional and analytical), properly managing and maintaining historical data, synchronizing source systems, reconstructing events (operational lineage), making data and reports accessible via metadata, streamlining data exchange, and preparing data for AI applications.
Often, the solution is sought in reference architectures based on, for example, a data warehouse, data lake, data lakehouse, or data fabric. While valuable, these architectures do not fully address the challenges mentioned above. They focus only on part of the data journey and fail to solve the core problems.
To truly tackle these challenges, a data architecture must cover the entire data journey: from source to insight. Only a holistic approach can achieve this. During this session, we will discuss a data architecture that spans the full data journey. The previously mentioned architectures may play a role within that architecture, but only as components of a larger whole.

This session will cover, among other topics:

• Three types of IT systems: source systems, compensation systems and analytical systems
• The positioning of data warehouses, data lakes, and data lakehouses as compensation systems
• An overview of the Delta data architecture
• How source systems can be made future-proof by “wrapping” them with additional modules
• The importance of abstraction and data minimization within a data architecture
• The role of metadata as the driving force behind a modern data shop.

The data lake landscape is undergoing a fundamental transformation. Traditional Hive tables are giving way to a new generation of open table formats—Apache Iceberg, Apache Hudi, Delta Lake, and emerging contenders like DuckDB—each promising to solve the inherent challenges of managing massive datasets at scale.
But which format fits your architecture? This session cuts through the marketing noise to deliver practical insights for data architects and engineers navigating this critical decision. We’ll explore how these formats tackle schema evolution, time travel, ACID transactions, and metadata management differently, and what these differences mean for your data platform’s performance, reliability, and total cost of ownership.
Drawing from real-world implementations, you’ll discover the hidden complexities, unexpected benefits, and common pitfalls of each approach. Whether you’re modernizing legacy Hive infrastructure, building greenfield data lakes, or evaluating lakehouse architectures, you’ll leave with a clear framework for choosing and implementing the right open table format for your specific use case—and the confidence to justify that decision to stakeholders.

Highlights:

• Format Face-Off: Direct comparison of Hive, Iceberg, Hudi, Delta Lake, and Ducklake capabilities across critical dimensions including ACID guarantees, partition evolution, and query performance optimization
• Real-World Battle Scars: Lessons learned from production deployments including migration strategies, performance tuning insights, and cost implications at petabyte scale
• Ecosystem Integration Deep-Dive: How each format plays with modern compute engines (e.g. Spark, Flink, Trino, Presto, DuckDB) and cloud platforms, plus vendor lock-in considerations
• The Hidden Costs: Beyond storage and compute—examining operational overhead, team expertise requirements, and long-term maintenance implications of your format choice
• Decision Framework: A practical methodology for evaluating which open table format aligns with your organization’s data architecture, workload patterns, and strategic goals.

Data Mesh, coined by Zhamak Dehghani, is a framework for federated data management and governance that gets a lot of attention from large organizations around the world facing problems with bottlenecked data teams and sprawling solution spaces. While the core principles of Data Mesh are well established in the literature, and practical implementation stories have started to emerge giving meat to the theoretical bones, some questions remain.
One of the biggest challenges is managing business context across multiple domains and data products. In this session, we will discuss how data modeling can be used to enable both within-domain design of understandable and discoverable data products as well as cross-domain understanding of domain boundaries, overlaps, and possibly conflicting business concepts. The well-known best practices of conceptual and logical data models prove their worth in this modern de-centralized framework by enabling semantic interoperability across different data products and domains, as well as allowing the organization to maintain a big picture of their data contents.

Topics and discussion points:

• Data Mesh basics – the paradigm and its principles
• Pros and cons of the Mesh approach
• Why context matters – semantic interoperability
• Data modeling as the key to context and semantics
• Modeling at two levels – data products and domains
• Maintaining the big picture and why the “Enterprise Data Model” is still a valid topic.

Are your data governance efforts stuck in endless debate cycles or only looks good on paper, with little to show for it? The Data Governance Sprint™ is a proven, accelerated method to establish practical data governance foundations in just five weeks. This session introduces a structured, workshop-based approach that moves beyond theory and delivers tangible outcomes: clear roles, a business glossary, an operating model, and early wins that build momentum. Designed for data leaders and practitioners, this methodology helps you overcome alignment struggles, engage stakeholders, and demonstrate measurable progress—fast.

• Understand why governance initiatives often fail—from lack of momentum to disengaged stakeholders—and how a sprint approach addresses these barriers
• Explore the 5-week Data Governance Sprint™ structure—a time-boxed, repeatable method to design, build, and validate core governance components
• Learn practical facilitation techniques to turn unproductive meetings into high-impact workshops that align business and IT around shared goals
• Prototype and test minimum viable governance deliverables such as a business glossary, lightweight operating model, and data stewardship roles
• See real-world applications and lessons learned from organizations that applied the sprint to accelerate adoption, avoid pitfalls, and sustain change
• Leave with an actionable roadmap for launching or rebooting a governance initiative in your own organization, showing quick wins while laying a foundation for long-term success.