Ravi Chandu Edru/ learn
← All topics

Learn

Ontology & Fabric IQ

From what counts as one thing, to an AI agent that can reason over your whole business.

29 of 29 concepts published

Build it in the Ontology Lab

Start here

The concepts read in order, and the lab is one build you keep open and advance as you go. Read a module, then open the lab and build the step it points to. By the end you have modeled an ontology, materialized its graph, and grounded an agent on it, by hand.

  1. Read in order

    Each page is one idea with a diagram, a small interactive, and a check-yourself. The pages bridge into each other, so top to bottom is the intended path.

  2. Keep one lab open

    The lab is a single build, not one per module. Open it once and advance the same build. Every concept links to the exact lab step where you practice it.

  3. Mark what you finish

    Each page has a mark-as-learned toggle. It saves in your browser, so this page shows how far you have gotten.

Before you build anything

Open the lab and run the guided tour once, so you see the whole map before you make a piece of it. Then load the Lakeshore cold-chain demo for real data, and build in manual mode. Manual is slower and teaches more than auto.

Open the Ontology Lab

Your progress

0 of 29 learned · 29 to go

Tick a concept here, or mark it learned on its page, and the meter fills in. Saved in this browser.

00

The Gap

Foundations · 0/3 learned

Why tables and semantic models hit a wall for AI agents, and what closes it.

  1. The map is not the territory

    Overview

    Your data is a map built for the questions you already had. An AI agent needs to go where it was never drawn.

    Read →
  2. Fabric IQ & the intelligence layer

    Overview

    How Fabric IQ turns a data estate into a shared, reasoning-ready model.

    Read →
  3. OneLake: the unified foundation

    Overview

    One governed lake so every concept binds to the same source of truth.

    Read →
01

Things

Foundations · 0/4 learned

The first modeling decision: what counts as one kind of thing.

  1. Entity

    Ontology

    A concept declared above your tables so every source means the same thing.

    Read →
  2. Property

    Ontology

    A named fact about an entity, with a declared type and unit.

    Read →
  3. Identity & keys

    Ontology

    What makes two rows the same thing — the entity key.

    Read →
  4. Entity instance

    Ontology

    A concrete occurrence, with provenance and the time it was true.

    Read →
02

Connections

Foundations · The payoff · 0/3 learned

How things relate — and how meaning travels across them.

  1. Relationship

    Ontology

    A typed, directional link with cardinality and its own attributes.

    Read →
  2. Foreign keys & relationship binding

    Ontology

    Wiring a relationship to real key columns across sources.

    Read →
  3. Traversal

    Ontology

    Multi-hop reasoning: Order → Shipment → Sensor → Breach.

    Read →
03

Grounding

The payoff · 0/3 learned

Connecting the model to real, governed, quality-checked data.

  1. Data binding

    Ontology

    Binding concepts to OneLake — lakehouse, eventhouse, semantic model.

    Read →
  2. Data quality & provenance

    Ontology

    Nullability, ranges, uniqueness, and lineage at the concept layer.

    Read →
  3. Time-series binding

    Ontology

    A second, streaming binding — telemetry alongside static facts.

    Read →
04

The Graph

The payoff · The frontier · 0/5 learned

Relationships become first-class, queryable, and traversable.

  1. Knowledge graph

    Graph

    Instances as nodes, relationships as edges — context made explicit.

    Read →
  2. Labeled property graph

    Graph

    The graph model Fabric uses: labels and properties on nodes and edges.

    Read →
  3. Asserted vs. derived edges

    Graph

    Facts you stored vs. links a rule inferred.

    Read →
  4. Cost & materialization

    Graph

    A graph is materialized and metered: you pay for a floor and for uptime, asked or not.

    Read →
  5. Graph query (GQL)

    Graph

    Ask relationship questions by pattern — no SQL joins.

    Read →
05

The Semantic Layer

The payoff · The frontier · 0/3 learned

Assemble the whole — and see how ontology and semantic models relate.

  1. Ontology

    Ontology

    Catalog + bindings + graph + query, assembled into one shared model.

    Read →
  2. Power BI semantic model

    Power BI semantic model

    Measures, hierarchies, dimensions — the curated analytics layer.

    Read →
  3. Semantic model vs. ontology

    Power BI semantic model

    When to reach for each — and generating one from the other.

    Read →
06

Agents & Action

The frontier · 0/6 learned

Both people and AI reason over the model — and act on it.

  1. NL2Ontology

    Data agent

    Turn a business question into a structured, grounded query.

    Read →
  2. Grounded data agent (GraphRAG)

    Data agent

    Answers grounded in the ontology instead of guessed from tables.

    Read →
  3. When graphs lose

    Data agent

    The graph advantage is regime-dependent: on simple fact retrieval, plain retrieval matches it for a fraction of the tokens.

    Read →
  4. Operations agent

    Operations agent

    Monitor live data, detect anomalies, trace impact across the graph.

    Read →
  5. Actions & rules

    Operations agent

    Governed actions the agent can take — the last mile of the loop.

    Read →
  6. Who can see it, and who acts

    Graph

    A viewer sees the whole graph, and the agent acts as its creator: two trust models meeting at one item.

    Read →
07

Breadth: Plan & Integrations

Breadth · 0/2 learned

Speak to the whole product — planning and where the ontology surfaces.

  1. Plan

    Plan

    No-code planning and forecasting over the same trusted foundation.

    Read →
  2. Integrations

    Integrations

    Surfacing the ontology in M365 Copilot, Foundry, and Copilot Studio.

    Read →