microsoft/LAB520-Get-Started-with-Models-in-Microsoft-Foundry-From-First-Inference-to-Deployed-Agent

microsoft/LAB520-Get-Started-with-Models-in-Microsoft-Foundry-From-First-Inference-to-Deployed-Agent

License: MIT

Stars: 0

Forks: 2

Open issues: 1

Created: 2026-06-23T22:39:38Z

Pushed: 2026-06-25T17:37:25Z

Default branch: main

Fork: no

Archived: no

README: --- lab: title: TechLab: Get Started with Models in Microsoft Foundry: From First Inference to Deployed Agent description: level: 200 duration: 120 minutes islab: true primarytopics: ---

@lab.Title

Welcome to Your Lab Environment

To begin, log into the virtual machine using the following credentials: +++@lab.VirtualMachine(Win11-Pro-Base).Password+++

Lab 1: Discover Models in Microsoft Foundry

> Duration: ~10 minutes | Phase: Orientation (UI)

Scenario

You are Serena, a developer at Zava -- a large global home-improvement retailer that operates both online and physical stores. Zava's platform receives thousands of customer product reviews daily from shoppers like Bruno, who is renovating his kitchen. Your task is to build an automated review moderation system that classifies customer reviews before they go live on the site. Eventually, this system will work alongside Cora, Zava's AI shopping assistant, to keep the platform safe and helpful.

In this lab, you will explore the Microsoft Foundry model catalog to find a model that can power Zava's review moderation pipeline.

Objective

Explore the Microsoft Foundry portal to discover available hosted models, understand model capabilities, and identify a model suitable for inference-based tasks like product review moderation.

---

Step 1: Open Microsoft Foundry Portal

Open, Microsoft Foundry Portal +++https://ai.azure.com+++ and sign in with the following Azure credentials using the start building button:

Username: +++@lab.CloudPortalCredential(User1).Username+++

If Prompted for a temporary Access Password 'TAP' +++@lab.CloudPortalCredential(User1).AccessToken+++

If Prompted for a Password: +++@lab.CloudPortalCredential(User1).Password+++

You will land on the Foundry resources page. This is the central hub for managing AI resources.

Ensure the new foundry switch at the top of the screen is turned on for New Foundry.

!IMAGE[newfoundry.png](../../media/newfoundry.png)

If don't get the screen below."You will have to update the current project to the latest Foundry version. Select the project listed to update.

!IMAGE[selectproject.png](../../media/selectproject.png)

Now Select the project from the available projects screen to load the project.

> [+alert] If you are unable to view the project please see below >

1. Switch to old foundry portal by toggling the New Foundry toggle at the top of the page !IMAGE[sla1i1tp.png](../../media/sla1i1tp.png) 2. Click "Continue without feedback" !IMAGE[yc6ix262.png](../../media/yc6ix262.png) 3. Project will be visible in old foundry portal, click on the project !IMAGE[f71vnukz.png](../../media/f71vnukz.png) 4. Once in the project click on the New Foundry toggle button to return to new foundry with the project retained in the new foundry !IMAGE[0dt15v07.png](../../media/0dt15v07.png) !IMAGE[wwez4x1w.png](../../media/wwez4x1w.png)

---

Step 2: Explore the Model Catalog

1. In the main windows, click discover from the top menu. 2. In the discover section browse the available models by clicking view all models -- these are production-ready, hosted models you can use without fine-tuning 4. You can use the filters within the 'Models' page to narrow down the list of models. For example you can filter by Supported features (Agent service, Fine-tuning, etc), Source (Azure OpenAI, Microsoft, Meta, Mistral, etc.) or by Inference Task (Chat Completition, Image Analysis etc). This allows you to quickly filter models based on a specific task or requirement.

Select a Model to view the Model Card

Take note of the following for each model:

| Property | What to Look For | |----------|-----------------| | Publisher | OpenAI, Microsoft, Meta, Mistral, etc. | | Task type | Chat completion, text generation, embeddings | | Deployment options | Serverless API, managed compute | | Pricing tier | Pay-as-you-go, free playground | | Benchmarks | Model performance and stats| | Responsible AI | Prompts and completions are passed through a default configuration of Azure AI Content Safety classification models |

---

Step 3: Identify a Model for This Lab

For this workshop, you need a model that supports chat completion -- the ability to accept a system prompt and user messages and return a structured response.

Recommended models for this lab:

| Model | Publisher | Why | |-------|-----------|-----| | gpt-4.1-mini | OpenAI | Fast, cost-efficient, excellent for classification | | gpt-4.1 | OpenAI | Higher quality, good for complex moderation | | Phi-4 | Microsoft | Strong reasoning, open-weight | >[!Tip] ** gpt-4.1-mini is the best choice for this lab -- it is fast, inexpensive, and well-suited for moderation and classification tasks.

---

Step 4: Check Model Details

Click on your chosen model (e.g., gpt-4.1-mini) to view its detail page:

1. Details Read the model description and capabilities 2. Deployments -- Deployment options 3. Benchmarks -- Review performance metrics 4. License -- The Model License

> You will deploy this model programmatically in Lab 2. For now, just confirm it is available in the catalog and you can see the model card details.

---

Step 5: Explore the Playground (Optional)

1. Select Deploy → Select "gpt-4.1-mini" under Use an existing deployment, which then brings you to the playground for the model deployment. 2. In the instructions, enter:

Instructions You are a product review moderator for Zava, a home-improvement retailer. Classify the following customer review as SAFE, NEEDS_REVIEW, or UNSAFE. Respond with only the classification label.

3. In the chat with model window, enter:

Prompt This paint is garbage and whoever designed it should be fired

4. Click Send and observe the response

This is a preview of the inference pattern you will implement in code during Labs 3 and 4 to moderate Zava product reviews.

---

What You Learned

• ✅ How to navigate the Microsoft Foundry portal
• ✅ How to browse the model catalog
• ✅ How to identify models suitable for chat completion tasks
• ✅ How to check quota and region availability
• ✅ How a model responds to a Zava review...