Microsoft Fabric Tour with Seattle Data & AI
On May 31st, 2025, I attended the inaugural Microsoft Fabric Tour at the Microsoft Reactor in Redmond — an in-depth, all-day exploration of Microsoft Fabric with leading industry experts. This groundbreaking event brought together data engineers, AI practitioners, and business leaders for a comprehensive deep dive into Microsoft’s unified data platform.
Event Overview
The Microsoft Fabric Tour featured an impressive lineup of 12 expert speakers from leading organizations including Microsoft, iLink Digital, TopLine, OmniData, Havens Consulting, and Eide Bailly LLP. The day-long event (9 AM — 6 PM) was structured around six core sessions, two panel discussions, and ample networking opportunities.
The Expert Speaker Lineup
Panel 1 — The Future of Unified Data:
Reid Havens (Founder, Havens Consulting)
Mohammad Ali (Partner Director, Microsoft)
Arindam Chatterjee (Principal Product Manager, Microsoft)
Individual Sessions:
Chris Hetzler (Solution Architect, Eide Bailly LLP) — Replacing ADF and Azure SQL
Sandeep Pawar (Sr Architect, Hitachi Solutions) & Anshul Sharma (Principal PM, Microsoft) — Real Time Intelligence in Fabric
Belinda Allen (Director of Enablement, iLink Digital) — Making Power BI Easy for End Users
Mark Kromer (Principal Product Manager, Microsoft) — Data Integration with Data Factory
Pam Lahoud (Principal PM Manager, Microsoft) — SQL Database in Fabric
Panel 2 — Modern Data Architecture:
Gregory Petrossian (Global Lead, Microsoft Datastax)
Sajay Suresh (Senior Director, Microsoft)
Sanjay Raut (Principal PM Manager, Microsoft)
Treb Gatte (Founder & CEO, TopLine)
Dan Erasmus (VP Sales, OmniData)
What made this event particularly valuable was the combination of expert-led sessions, hands-on insights, panel discussions, and extensive networking opportunities — all while enjoying excellent food and refreshments throughout the day.
Here are my key takeaways from this comprehensive exploration of Microsoft Fabric.
The Foundation: Building Good Data Models
One of the most fundamental insights came from the emphasis on creating quality data models. As highlighted in the presentation, good models require making thoughtful choices about:
What tools you use — Selecting the right technology stack for your specific needs
How you use them — Following best practices and established patterns
The data modeling lifecycle involves six critical phases:
Plan & Design — Strategic planning and architecture decisions
Connect & Transform Data — ETL/ELT processes and data integration
Model Data & Author DAX — Creating semantic models and business logic
Deploy & Manage Changes — Version control and deployment strategies
Consume & Distribute — Making data accessible to end users
Support and Monitor — Ongoing maintenance and performance optimization
Direct Lake: The Game-Changer for Power BI Performance
Understanding Direct Lake Mode
Direct Lake emerged as a standout feature, representing a significant evolution in Power BI storage modes. Unlike traditional Import or Direct Query modes, Direct Lake offers:
On-demand data loading into memory with automatic eviction of unused data
Superior performance for large data volumes compared to Direct Query
Reduced semantic model refresh time and compute costs
Near real-time insights with faster report availability
Key Direct Lake Concepts
The presentation covered several important technical concepts:
Fabric (Premium) Capacity requirements
Parquet file dependency for optimal performance
V-Order optimization for enhanced query performance
Data framing and transcoding capabilities
Cold and warm cache management
Column temperature optimization
Direct Query fallback mechanisms
Best Practices for Direct Lake Implementation
For organizations implementing Direct Lake, the Gold Layer (Lakehouse) approach was recommended:
Implement Star Schema compliant tables for optimal performance
Keep lean columns and rows with appropriate data types
Apply Z-Order and V-Order to fact tables for best storage and retrieval performance
Regular maintenance of gold layer delta tables with Optimize and Vacuum commands
Configure higher bin sizes (500 MB to 1 GB) using
spark.conf.set("spark.databricks.delta.optimizeWrite.binSize")
Direct Lake vs. Alternative Storage Modes
The comparison between storage modes revealed clear use cases:
Direct Query:
✅ Real-time analysis capability
✅ Can handle large data volumes
❌ Slow report rendering
❌ Dependent on data source optimization
❌ Limited support for complex DAX functions
Import:
✅ Best report rendering performance
✅ Supports complex calculations
❌ Data latency issues
❌ Processing complexity
❌ Extended processing time
Direct Lake:
✅ Real-time analysis
✅ Large data volume handling
✅ Fast report rendering performance
❌ Row volume limits
❌ High cardinality column limits
❌ Works with Parquet files only
Real-Time Intelligence: The Future of Data Processing
The Evolution to Real-Time
The event emphasized the transition from traditional batch processing to real-time intelligence, moving from “days and hours to minutes and seconds.” This shift enables:
Always-on insights for continuous monitoring
Insights-driven action for immediate response
Precision in the moment for time-sensitive decisions
Real-Time Intelligence Architecture
Microsoft Fabric’s Real-Time Intelligence hub provides a comprehensive solution with six core components:
Ingestion — Event streaming and data capture
Analytics — Real-time data processing
Digital Twin Builder — IoT and sensor data modeling
Dashboards — Live visualization and monitoring
Rules — Automated trigger and alert systems
Actions — Automated response mechanisms
Eventstream Capabilities
The new Eventstream features include:
Enhanced connectivity to streaming and discrete sources
MQTT, SAP HANA DB, Weather data feeds, and Solace PubSub+ support
Out-of-box transformations including field management and aggregations
SQL reference data integration
Content-based routing to multiple destinations including Eventhouse and Activator
CI/CD and Public API support for enterprise deployment
Fabric Data Agents: Making Data Conversational
AI-Powered Data Interaction
One of the most exciting developments is Fabric Data Agents, which transforms how users interact with data through natural language processing. These agents:
Enable conversational data access across OneLake (Lakehouse, Warehouse, Eventhouse, Semantic Model)
Preserve conversation context even with disparate data sources
Create endpoints for consumption within Fabric and external services like CoPilot Studio, Azure AI Foundry, and custom applications
Maintain existing security rules including Row-Level Security (RLS) and Column-Level Security (CLS)
Support programmatic management through SDK for deployment and evaluation
Data Agent Integration Components
The architecture includes three main pillars:
Semantic Model — Star Schema, business-friendly names, descriptions, and synonyms
Data Agents — Rich instructions and regular evaluation processes
Integration Points — Within Fabric, CoPilot Power BI, Azure AI Foundry, Teams, and CoPilot Studio
Data Factory Evolution: Comprehensive Data Movement
Pipeline Canvas Architecture
Microsoft Fabric’s Data Pipeline Canvas offers a three-tier approach:
Orchestration — Pipeline service coordination
Replicator — Replication service management
ADMS — Advanced Data Movement service
Enhanced Data Movement Capabilities
The roadmap includes several exciting developments:
Available Today:
Public APIs and CI/CD support in Copy Job
Upsert to SQL Database & Override to Fabric Lakehouse tables in Copy Job
20+ connectors supported in Copy Job
Usability and monitoring improvements in Copy Job
VNET Data Gateway support in Copy activity and Copy Job
Coming Soon:
Native Change Data Capture (CDC) in Copy Job
Additional copy patterns including merge, reset incremental copy to initial full load
Support for time-partitioned data and incremental copy from specific time windows
Aggregated monitoring with alerts in Copy Job
Pipeline integration with Copy Job
Variable Libraries Integration with Copy Job
No-Code Data Pipelines
The platform now offers intuitive, cloud-based user experiences that can:
Orchestrate data movement, transformation, cleansing, and control flow
Support triggered and scheduled execution models
Leverage AI-powered Copilots for enhanced productivity
Parameter Support Enhancement
A significant improvement is the addition of parameter support in Dataflows & Pipelines, which was identified as the top requested feature for metadata-driven data integration. This enables:
Dynamic pipeline configuration using parameters
Flexible connection specification for different environments
Metadata-driven ETL processes for scalable data integration
Modern Data Architecture: Pre and Post-Fabric
Traditional Architecture Challenges
The presentation highlighted the complexity of pre-Fabric architectures, where Team 3 typically managed:
High data volumes and complexity from multiple sources
Separate tools — Tabular Editor for models (.bim) and Power BI Desktop for reports (.pbix)
Complex collaboration workflows among developers
Multiple deployment stages through Azure Repos and Azure Pipelines
Separate workspace management for development, test, pre-production, and production environments
Fabric’s Unified Approach
Microsoft Fabric simplifies this through:
Unified platform eliminating tool fragmentation
Integrated development experience within the Fabric environment
Streamlined deployment processes with built-in version control
Simplified workspace management with consistent governance
OneLake: Zero ETL Data Unification
Shortcut and Mirroring Sources
OneLake provides seamless data integration through two main approaches:
Generally Available:
Azure SQL Database
Azure Data Lake Store
Microsoft OneLake
Google Cloud Storage
Snowflake
Amazon S3
Microsoft Dataverse
S3 Compatible (cloud/on-premises)
Public Preview:
Azure Cosmos DB
Azure SQL MI
Azure PostgreSQL
Databricks Catalog
Architecture Under the Hood
The technical architecture reveals how Fabric integrates with existing Azure infrastructure:
Azure SQL fleets with provisioning, auto-tiering, scaling, balancing, and routing
Mirroring capabilities for near real-time replication
SQL analytics endpoint (read-only) for query access
Delta Parquet in OneLake for unified storage
Integration with Fabric Data Warehouse, Lakehouse, Power BI, and Fabric Spark
Comprehensive Data Storage Options
Fabric provides four distinct storage options to meet different needs:
OneLake (Lakehouse) — Unstructured and semi-structured data with OLAP and ML capabilities, accessed primarily through Spark notebooks and jobs
Real-Time Intelligence (Eventhouse) — Unstructured, semi-structured, and structured data for real-time event processing, accessed through KQL (Kusto)
Fabric Data Warehouse — Structured data with OLAP capabilities, accessed primarily with T-SQL
Fabric Databases — Structured OLTP data, accessed primarily with T-SQL
Advanced Analytics with SQL Database in Fabric
The presentation showcased how Fabric enables advanced analytics scenarios through integrated pipelines that connect:
External data sources → Pipeline processing → SQL database → OneLake integration → Notebook analysis → Power BI visualization
This creates a seamless flow from operational systems to analytical insights, supporting both business users and analytics professionals.
Metadata-Driven Pipelines
A significant advancement is the support for metadata-driven pipelines that enable:
Dynamic pipeline configuration based on external metadata
Parameterized data integration for flexible ETL processes
Regional DevOps integration for distributed data management
API for GraphQL integration for modern application architectures
Key Takeaways from the Microsoft Fabric Tour
Direct Lake represents a significant leap forward in Power BI performance, offering the best of both Import and Direct Query modes with near real-time capabilities
Real-Time Intelligence enables organizations to move from batch to streaming analytics with comprehensive tooling, supporting the transition from “days and hours to minutes and seconds”
Fabric Data Agents democratize data access through natural language interfaces while maintaining enterprise security standards
Unified architecture dramatically simplifies previously complex multi-tool workflows, eliminating the need for separate development environments
OneLake’s zero-ETL approach reduces integration complexity and significantly improves time-to-value for data initiatives
Comprehensive storage options (Lakehouse, Eventhouse, Data Warehouse, Database) provide flexibility for different data types and use cases
Parameter support in pipelines addresses the top-requested feature for metadata-driven data integration
No-code/low-code capabilities make advanced data engineering accessible to a broader range of users
Final Thoughts
The inaugural Microsoft Fabric Tour was an exceptional learning experience that showcased how Microsoft is revolutionizing the data platform landscape. The combination of expert insights from 12 industry leaders, hands-on demonstrations, and extensive networking opportunities provided attendees with both strategic vision and practical implementation guidance.
For organizations considering their data modernization journey, Microsoft Fabric presents a compelling unified solution that addresses many traditional pain points while enabling advanced analytics scenarios. The platform’s evolution toward real-time capabilities, AI-powered features, and simplified user experiences positions it as a strong contender for enterprises looking to unify their data estate.
The event demonstrated that we’re truly entering the “Age of AI” where data fuels intelligence, and platforms like Microsoft Fabric are making it easier than ever to harness that power effectively.
The insights shared in this post reflect the presentations from the Microsoft Fabric Tour Seattle 2025 event held on May 31st, 2025, at the Microsoft Reactor in Redmond. Special thanks to Seattle Data & AI for organizing this exceptional event and to all the expert speakers who shared their knowledge and experiences. For the most current information on Microsoft Fabric capabilities and features, please refer to the official Microsoft documentation.