Enter your search text in the box above

Select a result to preview

Deliver Trusted Data Products for Analytics and AI in Microsoft Fabric

#fabcon2026 #session #friday #data-dev #data-products #ai #learning #microsoft #fabcon2026-session

Description

Microsoft Fabric accelerates analytics and AI at scale but without governed, trusted data, results fall short. This session's demo showcases how data products, and semantic models, are utilized to succeed with data and AI initiatives. Learn how modern AI-driven, DataOps-built MDM delivers trusted data products into OneLake, enabling business and agentic users to succeed with high quality data.

Key Takeaways

My Notes

Action Items

Slides

📥 Download Slides

3/2/26
Delivering Data
Products Across the
Microsoft Ecosystem
Designed in VS Code. Engineered with Copilot.
Published to Fabric. Governed by Purview.
Presented by: Craig Gravina, CTO of Semarchy
Data Without Meaning — And Without Trust
Fabric is a powerful analytics and AI platform. But it works with whatever data arrives.
● Power BI analysts see gd_fname, gd_dept — they don't know what these mean
● Copilot can query tables, but can't accurately answer business questions
● Agent Foundry generates responses, but without context they're unreliable
The gap isn't just understanding the data. It's deeper:
● Data without deduplication — multiple records for the same customer, supplier, product
● Data without quality rules — no validation, no certification, no confidence score
● Data without governance — no lineage, no provenance, no discoverability
● Data without context & semantics — column names like gd_dept instead of "Company Department"
This is what a Data Product solves. Not just semantic context — but trusted,
governed, comprehensive data – engineered to be consumed with confidence.
Data Products Deliver Trusted Data at Scale Through Fabric
A governed, reusable unit of trusted data available through Fabric for enterprise-wide access.
A Data Product delivers:
● Certified golden records — deduplicated, enriched, quality-scored master data.
One truth for every customer, product, supplier, or location.
● Rich semantic context — rich labeling, descriptions, and relationships make data
immediately understandable — by people and by AI.
● Governance from design-time — lineage, provenance, ownership, and policy are
embedded from the moment a Data Product is designed — not bolted on.
Data
Stewards
Business
Users
● Developer & API access — Consumable via REST APIs, enabling direct access to
governed master data by apps, microservices, and custom integrations.
● AI-native access — Vector embeddings, MCP endpoints, and RAG patterns make
Data Products first-class participants in AI architectures — not just tables AI queries
over.
Data
Scientists
Data Products
Application
Developers
GenAI &
Agentic
Not datasets. Not tables. Not integration. Governed, self-describing units of data
designed for consumption — by humans, applications, and AI.
From Design to Consumption — Inside the Microsoft Ecosystem
This is a full lifecycle — not a connector.
Design &
Engineering
Publish &
Govern
Consume
Design — Data products designed
with DataOps discipline in VS Code,
versioned in Git, promoted through
CI/CD pipelines
&
Engineer — AI-enhanced engineering with
GitHub Copilot: MCP for full design context,
RAG for enterprise standards, intent-based
design from business requirements
Publish to Fabric, Govern with Purview — Every data product published to
Purview with lineage, discoverability, and governance metadata from day one
Consume across the Fabric Ecosystem — Trusted data with deep
semantic context available to Power BI, Copilot, Agent Foundry, Data
Warehouse, Data Science, and Real-Time Intelligence
AI Foundry
Design
(VS Code)
AI-Engineering
(Copilot)
Publish
(Fabric)
Govern & Catalog
(Purview)
Copilot
Data Science
Power BI
Data Products Built in VS Code
● Data engineers work in VS Code — not a
proprietary GUI
Modern, DataOps-Driven
design experience (DXP)
Self-Serve, Governed,
Data Products
Operate
● Changes flow through CI/CD pipelines with
automated validation and promotion gates
● Every asset versions, promotes, and rolls
back as a coherent unit
VELOCITY
VALUE
Deliver
Outcomes
● Models, rules, workflows stored in Git
(GitHub, Azure DevOps)
Build
Design
Monitor &
Optimize
Test
integrated with widely-adopted engineering tools,
version control, and CI/CD
Why this matters: Your data engineers already use VS Code and GitHub. Semarchy is the only data
management platform where that's not a workaround — it's the primary design experience.
AI Data Product Engineering with GitHub Copilot
Full AI Engineering for Data Products
Copilot in DXP — AI copilot in VS Code for entity
modeling, rule authoring, workflow design, and SemQL
generation
A multimodal design experience where visual models, declarative
definitions, forms, and AI copilots are over a single design system.
Agentic copilot
design experience
Design MCP — Copilot gets live access to the full data
product design: entities, attributes, relationships, rules,
semantic models. Not just the file open in the editor — the
entire data product.
Knowledge RAG — Copilot accesses your enterprise data
standards, naming conventions, glossaries, and best
practices. AI recommendations are grounded in your
standards, not generic patterns.
Semarchy-native
agentic copilots
embedded in designtime, powered by
project context, MCP
and RAG
Visual Modeling
Declarative Syntax
Form-Based
Intent-Based Design — Describe what you need in
business terms: "I need a supplier master that handles
multi-country, supports hierarchy, and feeds procurement"
— and AI engineers the complete data product.
This is not AI-assisted. This is AI engineering.
Data Products Published as Native Fabric Assets
The keyword is publish — not integrate. A data product is published to Fabric the same way a software artifact is
published to a registry. It becomes a native asset in the target ecosystem.
Golden records — Certified, deduplicated, enriched master data lands in a Fabric Mirrored Database, continuously
synchronized
Semantic model — A custom Fabric Semantic Model populated with the full business vocabulary: entity labels, attribute
labels, descriptions, relationships. Every column has meaning. Every table has context.
Governance metadata — Lineage, discoverability, and data product ownership published to Microsoft Purview
Two Paths, Same Output:
CLI (GA) — Command-line driven, scriptable, fits
CI/CD pipelines. Ideal for DevOps engineers and
automation-first teams. Published directly from VS
Code or your deployment pipeline.
Fabric Workload (proposed) — Native Fabric UX
with a guided wizard, discoverable in the Workload
Hub. For Data Platform Owners, Fabric Admins, and
Analytics Engineers who want a Fabric-native
experience.
Both produce the same result: a Fabric Mirrored Database with golden records and a
custom Fabric Semantic Model enriched from SDP's governance layer.
Why Semantic Context Changes Everything
The Difference Between Querying and Reasoning
Without Semantic Context
Copilot sees gd_emp with columns
gd_fname, gd_dept, sd_loc. It
can query the data — but can't
confidently explain what it
represents or answer business
questions accurately.
With semantic context from a governed Data Product
Copilot sees "Employee" with "First Name," working in a "Company
Department" at a “Corporate Location, ”and a description: "A verified,
deduplicated employee record representing a unique individual who
works within the organization." It accurately answers: "How many
employees are working in marketing and where in what offices?" —
because it understands the data.
This isn't enrichment. It's the difference between AI that can query tables and AI that can reason
with business context and understanding.
Consume: Value Across Fabric
Every Fabric Service Benefits
Once a data product is published, it's consumable across the entire Fabric ecosystem — not as raw data, but as trusted,
semantically rich content that every service understands:
● Power BI — Reports built on golden data with business-friendly labels and descriptions. No column decoding. Copilot-ready
from day one.
● Copilot — Conversational analytics grounded in semantic context. Accurate answers because the data carries meaning,
not just schema.
● Agent Foundry — Data Agents consuming master data with full understanding. AI that reasons about business data, not
just queries tables.
This fuels the ecosystem to provide operational & advanced analytics
● Data Warehouse — Certified foundation dimensions with unambiguous business definitions. The trusted base layer for
analytics.
● Data Science — Semantically rich, trusted datasets. Skip the 80% prep time and start with governed golden records.
● Real-Time Intelligence — Event-driven analytics enriched with master data context
This is what it means to be a Fabric citizen — not an external system pushing data
through a connector, but a native participant in the ecosystem.
Govern: Microsoft Purview
Every Data Product Is Governed
Publishing to Fabric includes populating Microsoft Purview — automatically, as part of the publish:
Lineage — Trace data from source systems
through SDP mastering into Fabric. End-to-end
provenance.
Discoverability — Data products findable
through Purview search and browse. If it's not
in Purview, it doesn't exist to the governance
team.
Business metadata — Descriptions,
ownership, classifications, and data product
context. Not just technical metadata —
business meaning.
API access — Purview entries link back to
SDP APIs for deeper exploration and
governance operations
Governance isn't bolted on after the fact. It's part of the publish. Every data product — whether
published via CLI or Workload — lands in Purview the same way, from day one.
Microsoft Validates the Approach
"This collaboration enables Microsoft Power BI dashboards built
directly on enriched, trustworthy golden data records in Microsoft
OneLake with the depth of context provided by semantic models."
— Dipti Borkar, VP and GM, OneLake and Fabric ISVs, Microsoft
What to Remember
Three Key Takeaways
Data Products designed with AI in Microsoft tools. VS Code, GitHub Copilot, MCP,
RAG — this is AI engineering for data, not just "AI-assisted"
Published with semantic context that makes AI work. Fabric gets governed business
vocabulary at enterprise scale, automatically — Copilot and Agent Foundry consume data
with understanding
Native Fabric presence, governed by Purview. SDP is a Fabric citizen, not an external
system pushing data through a connector
In Closing: The Semarchy Data Platform
Trusted Data Products — Designed in VS
Code, Engineered with Copilot, Published
to Fabric, Governed by Purview.
● Website: semarchy.com
● Microsoft Partnership:
semarchy.com/partners/microsoft
Sound off.
The mic is all yours.
Influence the product roadmap.
Join the Fabric User Panel
Join the SQL User Panel
Share your feedback directly with our
Fabric product group and researchers.
Influence our SQL roadmap and
ensure it meets your real-life needs
https://aka.ms/JoinFabricUserPanel
https://aka.ms/JoinSQLUserPanel