ML Models in Fabric: Training, Deployment, and When to Stay on Azure ML

ML Models in Fabric: Training, Deployment, and When to Stay on Azure ML

ML Models in Fabric: Training, Deployment, and When to Stay on Azure ML Episode 22 • 2026-05-29 Microsoft Fabric ships its own MLflow registry — but is it a replacement for Azure Machine Learning? Matthias and Fabia work through the four-layer registry model, PREDICT versus Model Endpoints, the Direct Lake prediction loop, and the architectural question that actually determines the answer: where do your predictions land? What we discuss * A real-world mistake from a pre-Fabric era * The one question that reframes the architectural debate * How we got here — predecessor products and evolution * Why the "obvious" answer is often wrong * A real Reddit/Microsoft Q&A question unpacked * The concrete recommended architecture * F-SKU realism — what this actually costs * When the rejected approach is actually right * Risks of the recommended path * What Microsoft is shipping that changes the calculus * The architectural principle to take home Key takeaways * Where do the predictions land. That question answers the architecture. OneLake plus Power BI Direct Lake — Fabric ML Model, genuinely the right call. REST API for an app — evaluate Endpoints maturity or route to Azure ML. GPU training,... * I'd go further. Already on Databricks with Unity Catalog? Don't migrate. Fabric ML Model is not a migration target for Databricks shops — the platform maturity gap is real. The hybrid that actually works: train on Azure ML with GPU,... * For Power BI shops — yes. PREDICT writes predictions to a Delta table in OneLake, Direct Lake reads it with zero copy, zero scheduled refresh. That eliminates an entire class of ETL work. But only if Power BI is your audience. Resources * ML Experiment [https://learn.microsoft.com/fabric/data-science/machine-learning-experiment?wt.mc_id=AZ-MVP-5003447] * Notebooks [https://learn.microsoft.com/fabric/data-engineering/how-to-use-notebook?wt.mc_id=AZ-MVP-5003447] * Lakehouse [https://learn.microsoft.com/fabric/data-engineering/lakehouse-overview?wt.mc_id=AZ-MVP-5003447] * Direct Lake [https://learn.microsoft.com/fabric/get-started/direct-lake-overview?wt.mc_id=AZ-MVP-5003447] * Code-first AutoML [https://learn.microsoft.com/fabric/data-science/python-automated-machine-learning-fabric?wt.mc_id=AZ-MVP-5003447] * Low-code AutoML [https://learn.microsoft.com/fabric/data-science/low-code-automl?wt.mc_id=AZ-MVP-5003447] * SynapseML [https://learn.microsoft.com/fabric/data-science/synapse-overview?wt.mc_id=AZ-MVP-5003447] * Activator [https://learn.microsoft.com/fabric/real-time-intelligence/data-activator/activator-introduction?wt.mc_id=AZ-MVP-5003447] * Machine learning model in Microsoft Fabric [https://learn.microsoft.com/fabric/data-science/machine-learning-model?wt.mc_id=AZ-MVP-5003447] * What is Data Science in Microsoft Fabric? [https://learn.microsoft.com/fabric/data-science/data-science-overview?wt.mc_id=AZ-MVP-5003447] * Tutorial Part 3: Train and register a machine learning model [https://learn.microsoft.com/fabric/data-science/tutorial-data-science-train-models?wt.mc_id=AZ-MVP-5003447] * Tutorial Part 4: Perform batch scoring and save predictions [https://learn.microsoft.com/fabric/data-science/tutorial-data-science-batch-scoring?wt.mc_id=AZ-MVP-5003447] * Machine learning model scoring with PREDICT [https://learn.microsoft.com/fabric/data-science/model-scoring-predict?wt.mc_id=AZ-MVP-5003447] * Serve real-time predictions with ML model endpoints (Preview) [https://learn.microsoft.com/fabric/data-science/model-endpoints?wt.mc_id=AZ-MVP-5003447] * Train models with scikit-learn in Microsoft Fabric [https://learn.microsoft.com/fabric/data-science/train-models-scikit-learn?wt.mc_id=AZ-MVP-5003447] About the show Built on ElevenLabs [https://elevenlabs.io] voice synthesis. Matthias — cloned voice. Fabia — designed AI co-host. See Matthias live on YouTube (Fabric Friday) [https://www.youtube.com/@yourchannelhere], at his meetups, and at conferences like FabCon [https://fabricconf.com]. Hosted by Matthias Falland — Microsoft Data Platform MVP and community architect behind the Fabric Periodic Table [https://www.fabricperiodictable.com]. New episodes every Friday. Submit your case Have an architecture decision you are wrestling with? DM Matthias on LinkedIn — find him as Matthias Falland [https://www.linkedin.com/in/matthiasfalland/]. Three to five sentences about the decision, your team size, and your current stack. We anonymize before airing. Built on ElevenLabs voice synthesis. Brand design based on fabricperiodictable.com [https://www.fabricperiodictable.com].

Event Schema Set: Contracts That Stop Midnight Breakage

Event Schema Set: Contracts That Stop Midnight Breakage Episode 21 • 2026-05-22 Event Schema Set is Fabric's contract layer for streaming data — but it ships in Preview with real gaps. Matthias and Fabia unpack the retrofit trap, the dead-letter gap everyone worries about, and when Confluent Schema Registry is honestly the better call. What we discuss * A real-world mistake from a pre-Fabric era * The one question that reframes the architectural debate * How we got here — predecessor products and evolution * Why the "obvious" answer is often wrong * A real Reddit/Microsoft Q&A question unpacked * The concrete recommended architecture * F-SKU realism — what this actually costs * When the rejected approach is actually right * Risks of the recommended path * What Microsoft is shipping that changes the calculus * The architectural principle to take home Key takeaways * Treat schemas as append-only contracts. Add fields with defaults — safe. Remove required fields — breaks consumers. Change a type — silent data corruption. Rename a field — silent loss in KQL queries. The system won't stop you. Your... * Fair argument. And honestly? If you're an existing Kafka shop with established Confluent practices — use Confluent. The migration cost isn't worth it. Eventstream can deserialize Confluent-encoded payloads natively. You get Avro plus JSON... * But you operate a separate cluster. Separate auth. Separate billing. If your entire stack is Fabric-native — Eventstream, Notebook, Activator, Eventhouse — the integration is a real win. No client library. No external cluster. Governance... Resources * Schema Registry — known limitations [https://learn.microsoft.com/fabric/real-time-intelligence/schema-sets/schema-registry-limitations?wt.mc_id=AZ-MVP-5003447] * CloudEvents 1.0 [https://github.com/cloudevents/spec] * Use schemas in eventstreams [https://learn.microsoft.com/fabric/real-time-intelligence/schema-sets/use-event-schemas?wt.mc_id=AZ-MVP-5003447] * Real-Time Hub Schemas [https://learn.microsoft.com/fabric/real-time-intelligence/schema-sets/create-manage-event-schemas-real-time-hub?wt.mc_id=AZ-MVP-5003447] * Business Events Concepts [https://learn.microsoft.com/fabric/real-time-hub/business-events/business-events-concepts?wt.mc_id=AZ-MVP-5003447] * Consume Business Events from Activator [https://learn.microsoft.com/fabric/real-time-hub/business-events/consume-business-events-from-activator?wt.mc_id=AZ-MVP-5003447] * Eventhouse [https://learn.microsoft.com/fabric/real-time-intelligence/eventhouse?wt.mc_id=AZ-MVP-5003447] * Confluent Kafka source [https://learn.microsoft.com/fabric/real-time-intelligence/event-streams/add-source-confluent-kafka?wt.mc_id=AZ-MVP-5003447] * Schema Registry in Fabric Real-Time Intelligence (preview) — Overview [https://learn.microsoft.com/fabric/real-time-intelligence/schema-sets/schema-registry-overview?wt.mc_id=AZ-MVP-5003447] * Create and manage event schema sets [https://learn.microsoft.com/fabric/real-time-intelligence/schema-sets/create-manage-event-schema-sets?wt.mc_id=AZ-MVP-5003447] * Create and manage event schemas in schema sets [https://learn.microsoft.com/fabric/real-time-intelligence/schema-sets/create-manage-event-schemas?wt.mc_id=AZ-MVP-5003447] * EventSchemaSet REST API definition [https://learn.microsoft.com/rest/api/fabric/articles/item-management/definitions/eventschemaset-definition?wt.mc_id=AZ-MVP-5003447] * Eventstream Overview — Schema Management section [https://learn.microsoft.com/fabric/real-time-intelligence/event-streams/overview?wt.mc_id=AZ-MVP-5003447] * Multiple-Schema Inferencing in Eventstream (Preview) [https://learn.microsoft.com/fabric/real-time-intelligence/event-streams/process-events-with-multiple-schemas?wt.mc_id=AZ-MVP-5003447] * Eventstream Data Formats: JSON, CSV, Avro [https://learn.microsoft.com/fabric/real-time-intelligence/event-streams/data-formats?wt.mc_id=AZ-MVP-5003447] About the show Built on ElevenLabs [https://elevenlabs.io] voice synthesis. Matthias — cloned voice. Fabia — designed AI co-host. See Matthias live on YouTube (Fabric Friday) [https://www.youtube.com/@yourchannelhere], at his meetups, and at conferences like FabCon [https://fabricconf.com]. Hosted by Matthias Falland — Microsoft Data Platform MVP and community architect behind the Fabric Periodic Table [https://www.fabricperiodictable.com]. New episodes every Friday. Submit your case Have an architecture decision you are wrestling with? DM Matthias on LinkedIn — find him as Matthias Falland [https://www.linkedin.com/in/matthiasfalland/]. Three to five sentences about the decision, your team size, and your current stack. We anonymize before airing. Built on ElevenLabs voice synthesis. Brand design based on fabricperiodictable.com [https://www.fabricperiodictable.com].

Data Activator: Stateful Alerts That Don't Spam Your Team

Data Activator: Stateful Alerts That Don't Spam Your Team Episode 20 • 2026-05-15 Data Activator is Fabric's no-code event detection engine — but most teams build it wrong. Matthias and Fabia unpack the stateful rule model, the billing trap everyone hits once, and when Power Automate is actually the better answer. What we discuss * A real-world mistake from a pre-Fabric era * The one question that reframes the architectural debate * How we got here — predecessor products and evolution * Why the "obvious" answer is often wrong * A real Reddit/Microsoft Q&A question unpacked * The concrete recommended architecture * F-SKU realism — what this actually costs * When the rejected approach is actually right * Risks of the recommended path * What Microsoft is shipping that changes the calculus * The architectural principle to take home Key takeaways * Take-home: the entity hierarchy is the product. * Fair. For low-frequency data — a daily KPI check — it works fine. Where it breaks: ten thousand events per second per rule. Power Automate isn't built for that volume. And a per-flow variable isn't per-object state — you'd need one flow... * Right. Wrong in exactly one place — the state machine. Here's the thing. A stateless rule fires on every matching event. Value greater than twenty-five? Sensor reports every five seconds, stays above twenty-five for an hour — you get seven... Resources * Add Activator to Eventstream [https://learn.microsoft.com/fabric/real-time-intelligence/event-streams/add-destination-activator?wt.mc_id=AZ-MVP-5003447] * Activator from KQL Queryset [https://learn.microsoft.com/fabric/real-time-intelligence/data-activator/activator-alert-queryset?wt.mc_id=AZ-MVP-5003447] * Activator from RTD [https://learn.microsoft.com/fabric/real-time-intelligence/data-activator/activator-get-data-real-time-dashboard?wt.mc_id=AZ-MVP-5003447] * Activator from Power BI [https://learn.microsoft.com/fabric/real-time-intelligence/data-activator/activator-get-data-power-bi?wt.mc_id=AZ-MVP-5003447] * Real-Time Hub Set Alerts [https://learn.microsoft.com/fabric/real-time-hub/set-alerts-data-streams?wt.mc_id=AZ-MVP-5003447] * Set Alerts on Anomaly Detection [https://learn.microsoft.com/fabric/real-time-hub/set-alerts-anomaly-detection?wt.mc_id=AZ-MVP-5003447] * What is Fabric Activator? [https://learn.microsoft.com/fabric/real-time-intelligence/data-activator/activator-introduction?wt.mc_id=AZ-MVP-5003447] * Tutorial: Create and activate a Fabric Activator rule [https://learn.microsoft.com/fabric/real-time-intelligence/data-activator/activator-tutorial?wt.mc_id=AZ-MVP-5003447] * Create a rule in Fabric Activator [https://learn.microsoft.com/fabric/real-time-intelligence/data-activator/activator-create-activators?wt.mc_id=AZ-MVP-5003447] * Trigger modeling in Activator [https://learn.microsoft.com/fabric/real-time-intelligence/data-activator/activator-trigger-model?wt.mc_id=AZ-MVP-5003447] * Fabric Activator rules [https://learn.microsoft.com/fabric/real-time-intelligence/data-activator/activator-rules-overview?wt.mc_id=AZ-MVP-5003447] * Detection conditions [https://learn.microsoft.com/fabric/real-time-intelligence/data-activator/activator-detection-conditions?wt.mc_id=AZ-MVP-5003447] * Activator Limitations [https://learn.microsoft.com/fabric/real-time-intelligence/data-activator/activator-limitations?wt.mc_id=AZ-MVP-5003447] * Latency in Activator [https://learn.microsoft.com/fabric/real-time-intelligence/data-activator/activator-latency?wt.mc_id=AZ-MVP-5003447] * Activator Capacity Consumption [https://learn.microsoft.com/fabric/real-time-intelligence/data-activator/activator-capacity-usage?wt.mc_id=AZ-MVP-5003447] About the show Built on ElevenLabs [https://elevenlabs.io] voice synthesis. Matthias — cloned voice. Fabia — designed AI co-host. See Matthias live on YouTube (Fabric Friday) [https://www.youtube.com/@yourchannelhere], at his meetups, and at conferences like FabCon [https://fabricconf.com]. Hosted by Matthias Falland — Microsoft Data Platform MVP and community architect behind the Fabric Periodic Table [https://www.fabricperiodictable.com]. New episodes every Friday. Submit your case Have an architecture decision you are wrestling with? DM Matthias on LinkedIn — find him as Matthias Falland [https://www.linkedin.com/in/matthiasfalland/]. Three to five sentences about the decision, your team size, and your current stack. We anonymize before airing. Built on ElevenLabs voice synthesis. Brand design based on fabricperiodictable.com [https://www.fabricperiodictable.com].

Real-Time Dashboard: When 10-Second Refresh Changes the Architecture

Real-Time Dashboard: When 10-Second Refresh Changes the Architecture Episode 19 • 2026-05-08 Real-Time Dashboard is not Power BI wearing a different hat. Matthias and Fabia unpack the naming collision, permission separation, Activator alert traps, and when you should actually use Power BI DirectQuery instead. What we discuss * A real-world mistake from a pre-Fabric era * The one question that reframes the architectural debate * How we got here — predecessor products and evolution * Why the "obvious" answer is often wrong * A real Reddit/Microsoft Q&A question unpacked * The concrete recommended architecture * F-SKU realism — what this actually costs * When the rejected approach is actually right * Risks of the recommended path * What Microsoft is shipping that changes the calculus * The architectural principle to take home Key takeaways * If someone asks 'what's happening right now' — Real-Time Dashboard. * But you lose permission separation. You lose tile-as-query simplicity. And your team will absolutely blame the network when the DirectQuery report takes four seconds to load at scale. Different tools, different tradeoffs. * Fair argument. Power BI can connect to KQL via DirectQuery. You get DAX measures, RLS, the full semantic model. And in Premium, automatic page refresh goes as low as five seconds. So if your team already lives in Power BI — that's a legitimate path. Resources * KQL Database [https://learn.microsoft.com/fabric/real-time-intelligence/create-database?wt.mc_id=AZ-MVP-5003447] * KQL Queryset [https://learn.microsoft.com/fabric/real-time-intelligence/create-query-set?wt.mc_id=AZ-MVP-5003447] * Real-Time Hub [https://learn.microsoft.com/fabric/real-time-hub/real-time-hub-overview?wt.mc_id=AZ-MVP-5003447] * Activator on RTD [https://learn.microsoft.com/fabric/real-time-intelligence/data-activator/activator-get-data-real-time-dashboard?wt.mc_id=AZ-MVP-5003447] * Anomaly Detection [https://learn.microsoft.com/fabric/real-time-intelligence/anomaly-detection?wt.mc_id=AZ-MVP-5003447] * Power BI + KQL [https://learn.microsoft.com/power-bi/connect-data/real-time-intelligence-sample?wt.mc_id=AZ-MVP-5003447] * Fabric Map [https://learn.microsoft.com/fabric/real-time-intelligence/map/create-map?wt.mc_id=AZ-MVP-5003447] * What is Real-Time Dashboard? [https://learn.microsoft.com/fabric/real-time-intelligence/real-time-dashboards-overview?wt.mc_id=AZ-MVP-5003447] * Create a Real-Time Dashboard [https://learn.microsoft.com/fabric/real-time-intelligence/dashboard-real-time-create?wt.mc_id=AZ-MVP-5003447] * Real-Time Dashboard Permissions [https://learn.microsoft.com/fabric/real-time-intelligence/dashboard-permissions?wt.mc_id=AZ-MVP-5003447] * Use Parameters in Real-Time Dashboards [https://learn.microsoft.com/fabric/real-time-intelligence/dashboard-parameters?wt.mc_id=AZ-MVP-5003447] * Customize Real-Time Dashboard Visuals [https://learn.microsoft.com/fabric/real-time-intelligence/dashboard-visuals-customize?wt.mc_id=AZ-MVP-5003447] * Activator Limitations [https://learn.microsoft.com/fabric/real-time-intelligence/data-activator/activator-limitations?wt.mc_id=AZ-MVP-5003447] * Generate Real-Time Dashboard with Copilot [https://learn.microsoft.com/fabric/fundamentals/copilot-generate-dashboard?wt.mc_id=AZ-MVP-5003447] * Copilot-assisted Real-Time Data Exploration (Preview) [https://learn.microsoft.com/fabric/real-time-intelligence/dashboard-explore-data?wt.mc_id=AZ-MVP-5003447] About the show Built on ElevenLabs [https://elevenlabs.io] voice synthesis. Matthias — cloned voice. Fabia — designed AI co-host. See Matthias live on YouTube (Fabric Friday) [https://www.youtube.com/@yourchannelhere], at his meetups, and at conferences like FabCon [https://fabricconf.com]. Hosted by Matthias Falland — Microsoft Data Platform MVP and community architect behind the Fabric Periodic Table [https://www.fabricperiodictable.com]. New episodes every Friday. Submit your case Have an architecture decision you are wrestling with? DM Matthias on LinkedIn — find him as Matthias Falland [https://www.linkedin.com/in/matthiasfalland/]. Three to five sentences about the decision, your team size, and your current stack. We anonymize before airing. Built on ElevenLabs voice synthesis. Brand design based on fabricperiodictable.com [https://www.fabricperiodictable.com].

Real-Time Hub: The Yellow Pages Your Streams Were Missing

Real-Time Hub: The Yellow Pages Your Streams Were Missing Episode 18 • 2026-05-01 Matthias and Fabia unpack Fabric's Real-Time Hub — the tenant-wide catalog that sits above Eventstream, Eventhouse, and Activator. They tackle why it feels redundant until it doesn't, dig into a real Reddit question about skipping the Hub entirely, and lay out the four-layer real-time stack every architect should internalize. What we discuss * A real-world mistake from a pre-Fabric era * The one question that reframes the architectural debate * How we got here — predecessor products and evolution * Why the "obvious" answer is often wrong * A real Reddit/Microsoft Q&A question unpacked * The concrete recommended architecture * F-SKU realism — what this actually costs * When the rejected approach is actually right * Risks of the recommended path * What Microsoft is shipping that changes the calculus * The architectural principle to take home Key takeaways * So — today's lesson. The Hub is not a processing engine. It's not a new Eventstream. It's the inventory layer that streaming has always been missing. Pattern dictates platform — if your pattern is discovery at organizational scale, this is... * I mean, fair question. If every stream you have lives in one workspace and one team owns them all — the Hub's discoverability value is close to zero. You already know what exists. Same if you're publishing streams to non-Fabric consumers... * Right. And... that's actually fine for small setups. The connector list is identical — same Azure Event Hubs tile, same Kafka tile, same CDC tiles. Both paths end up creating an eventstream artifact. But here's the thing. Eventstream is... Resources * managed private endpoint [https://learn.microsoft.com/fabric/real-time-intelligence/event-streams/set-up-private-endpoint?wt.mc_id=AZ-MVP-5003447] * Eventstream Overview [https://learn.microsoft.com/fabric/real-time-intelligence/event-streams/overview?wt.mc_id=AZ-MVP-5003447] * KQL Database [https://learn.microsoft.com/fabric/real-time-intelligence/create-database?wt.mc_id=AZ-MVP-5003447] * Activator Overview [https://learn.microsoft.com/fabric/real-time-intelligence/data-activator/activator-introduction?wt.mc_id=AZ-MVP-5003447] * Real-Time Dashboard [https://learn.microsoft.com/fabric/real-time-intelligence/dashboard-real-time-create?wt.mc_id=AZ-MVP-5003447] * Schema Sets [https://learn.microsoft.com/fabric/real-time-intelligence/schema-sets/create-manage-event-schema-sets?wt.mc_id=AZ-MVP-5003447] * Digital Twin Builder [https://learn.microsoft.com/fabric/real-time-intelligence/digital-twin-builder/tutorial-0-introduction?wt.mc_id=AZ-MVP-5003447] * Real-Time Hub Overview [https://learn.microsoft.com/fabric/real-time-hub/real-time-hub-overview?wt.mc_id=AZ-MVP-5003447] * Get Started with Real-Time Hub [https://learn.microsoft.com/fabric/real-time-hub/get-started-real-time-hub?wt.mc_id=AZ-MVP-5003447] * Supported Sources [https://learn.microsoft.com/fabric/real-time-hub/supported-sources?wt.mc_id=AZ-MVP-5003447] * Add Azure Event Hubs Source [https://learn.microsoft.com/fabric/real-time-hub/add-source-azure-event-hubs?wt.mc_id=AZ-MVP-5003447] * Add Azure IoT Hub Source [https://learn.microsoft.com/fabric/real-time-hub/add-source-azure-iot-hub?wt.mc_id=AZ-MVP-5003447] * Get Azure Blob Storage Events [https://learn.microsoft.com/fabric/real-time-hub/get-azure-blob-storage-events?wt.mc_id=AZ-MVP-5003447] * Create Streams from Workspace Item Events [https://learn.microsoft.com/fabric/real-time-hub/create-streams-fabric-workspace-item-events?wt.mc_id=AZ-MVP-5003447] * Create Streams from OneLake Events [https://learn.microsoft.com/fabric/real-time-hub/create-streams-fabric-onelake-events?wt.mc_id=AZ-MVP-5003447] About the show Built on ElevenLabs [https://elevenlabs.io] voice synthesis. Matthias — cloned voice. Fabia — designed AI co-host. See Matthias live on YouTube (Fabric Friday) [https://www.youtube.com/@yourchannelhere], at his meetups, and at conferences like FabCon [https://fabricconf.com]. Hosted by Matthias Falland — Microsoft Data Platform MVP and community architect behind the Fabric Periodic Table [https://www.fabricperiodictable.com]. New episodes every Friday. Submit your case Have an architecture decision you are wrestling with? DM Matthias on LinkedIn — find him as Matthias Falland [https://www.linkedin.com/in/matthiasfalland/]. Three to five sentences about the decision, your team size, and your current stack. We anonymize before airing. Built on ElevenLabs voice synthesis. Brand design based on fabricperiodictable.com [https://www.fabricperiodictable.com].