Prepare your environment

Use this checklist before you start the labs in your own environment. The supported path uses a Windows computer, browser-based cloud services, and PowerShell for the event simulator.

For hosted delivery, the lab provider prepares the same tools, access, and settings before learners start. This page describes the required starting state, not the resources built during the exercises. Use it for environment preparation; licensing, software installation, and administrator approvals aren't part of the L400 challenge's six-hour duration.

Prepare your computer

Use a Windows 11 computer with the following software and access:

Install PowerShell 7 and Python 3.13 before you start. Include pip and add Python to PATH during Python installation.

1. Open PowerShell and confirm the installed versions:

   $PSVersionTable.PSVersion
   python --version
   python -m pip --version
   

2. Install the event simulator dependency from this repository's main branch:

   python -m pip install --requirement https://raw.githubusercontent.com/MicrosoftLearning/mslearn-unify-data/main/Allfiles/L300-unify-data/event-simulator/requirements.txt
   

3. Confirm that the dependency is available:

   python -m pip show azure-eventhub
   

The Fabric notebooks run in Fabric, not on your computer. You don't need a local Spark installation, SQL Server, Git, Visual Studio Code, or Azure CLI for the L300 exercise.

Confirm licenses and permissions

Use a work or school Microsoft Entra account for the lab services. Use the same account and tenant for Fabric and the Copilot experience used by your exercise: Copilot Studio for L300, or Microsoft 365 Copilot for L400.

The exercise creates the workspace, data, agents, and Foundry deployment. You don't need to create those items in advance.

Review the administrator requirements in Fabric data agent tenant settings. Complete these approvals before the lab rather than changing tenant policies during it.

Prepare the L300 customer-feedback extensions

To complete the sentiment and enrichment tasks as well as the core path, prepare:

• An Azure OpenAI resource with a deployed chat-completion model that supports the lab's sentiment request, plus its resource name and deployment name.
• Access from Fabric SQL to the Azure OpenAI endpoint. The administrator responsible for role assignments can grant Cognitive Services OpenAI User on that resource to the Fabric SQL managed identity used by the database when you create it.
• Permission to create database-scoped credentials, tables, and stored procedures in the lab SQL database.
• A Fabric notebook runtime with the supplied synapse.ml.spark.aifunc module and permission to use its AI functions.
• Access to SQL database mirroring, OneLake shortcuts, and ontology creation in the selected Fabric region and capacity.

Azure OpenAI and AI-function usage must have approved billing and available quota. Use synthetic lab data only.

Record your environment values

Keep the following non-secret values available for the exercise. In your own environment, record the resources and settings you prepared. In hosted delivery, the provider supplies these values.

Don't record passwords, access tokens, or connection strings in this table. You obtain the eventstream connection details during the exercise and use them only in the current PowerShell session.

Confirm readiness

• [ ] PowerShell 7, Python 3.13, pip, and the event simulator dependency are available.
• [ ] Browser sign-in and the required network connections are allowed.
• [ ] Your account has the required licenses, paid Fabric capacity access, and workspace-creation permissions.
• [ ] Required tenant settings, cross-service access, regional availability, and model quota are prepared.
• [ ] The customer-feedback prerequisites are prepared for the full exercise.
• [ ] Your environment values and resource ownership boundaries are recorded.

Prepare for the L400 challenge

Prepare the common computer, Fabric, and Foundry requirements above and the Microsoft 365 Copilot requirements below. The Copilot Studio-specific requirements above apply to L300. L400 publishes Fabric data agents directly to Microsoft 365 Copilot. The L400 challenge doesn't use the L300 extension's prepared Azure OpenAI resource, SQL managed-identity credential, or sentiment stored procedure. It uses a Foundry model called by a supplied Python client.

The solutions use distinct agents and source selections in one Fabric workspace. Planning tables are populated once and reused. Prepare access for all three solutions, but leave resource creation to the learner:

Fabric SQL automatically mirrors supported tables into OneLake. Don't require Azure SQL Database or separate mirroring configuration. The business solution validates propagation of a source update through a shortcut; the ontology solution later adds sentiment in the same database.

Before delivery, verify sign-in, simulator dependency installation, local SQL authentication, model invocation, notebook AI functions, and direct Fabric data-agent publication and use in Microsoft 365 Copilot with an equivalent identity. Use disposable provider resources for checks requiring a database or model; don't prebuild the learner's solution. Supply the assigned account, capacity, subscription, region, cleanup boundary, and support contact through the environment's resource information panel.

Release the source assets and L400 ZIP to the repository's main branch before delivery. The setup dependency command, learner download, and curation notebook all use MicrosoftLearning/mslearn-unify-data/main. Confirm that those URLs return the released files; local files alone don't make these download paths available.

Check the model retirement schedule before delivery. Treat gpt-5-mini as an example, not a permanent model requirement. Supply approved model choices and available quota for the assigned region. An operational replacement must support Foundry Agent Service and Fabric IQ tool calling. A sentiment replacement must support the OpenAI Responses API, the supplied client's minimal reasoning effort, and its output-token parameter. Verify a completed response with a valid sentiment label and an operational Fabric IQ call before approving a replacement. If the replacement requires different client parameters, update and retest the supplied client and package before delivery rather than asking learners to troubleshoot model compatibility. Don't use deprecated models unavailable to new customers, such as gpt-4.1-mini, as a new-subscription dependency.

The 360-minute allocation is an author estimate, not a pilot-verified duration. Reuse infrastructure without merging the agents' source scopes: the operational agent uses planning and streaming evidence, while the business and ontology agents use their relevant transactional and curated data. The ontology solution extends the business solution and doesn't require streaming sources. Validate all required service paths, reset procedures, cleanup, and self-paced timing before release.

Continue to either Build an end-to-end Caldova data agent solution or Solve Caldova launch-readiness challenges with grounded agents. Download the corresponding learner package and create the lab resources as instructed there.