Getting Started With AI in Azure

This lesson explains how Azure underpins modern AI applications. We’ll cover Azure structure and identity, introduce Microsoft Foundry and its lifecycle support, and describe how Foundry projects combine models, agents, tools, and knowledge to build production AI solutions. Finally, you’ll see how applications access Foundry models and services via endpoints and SDKs.

Azure fundamentals: identity, resources, and governance

Microsoft Azure is a cloud platform that provides infrastructure and services so organizations can build, deploy, and operate applications without managing the physical data center. A critical piece of that environment is Microsoft Entra ID (identity and access management). Entra ID answers two fundamental questions for Azure:

Who or what is requesting access?
What is that identity allowed to do?

Core Entra ID concepts:

Users — individual human identities.
Groups — collections of users for simplified access control.
Applications and service identities — identities for apps and automated services.
Permissions and roles — rules that define allowed actions.

Use cases and examples:

Identity Component	Use Case	Example
Users	Human access and MFA	`alice@contoso.com`
Groups	Grant permissions to multiple users	`SupportTeam` group
Service identities	Non-human app authentication	`my-app-service-principal`
Roles & permissions	Fine-grained access control	Built-in role `Contributor`

Azure resources are arranged in a hierarchy that affects access, billing, and governance. The typical hierarchy is:

Level	Purpose	Example
Tenant	Organization identity boundary in Microsoft Entra ID	`contoso.onmicrosoft.com`
Subscription	Billing and resource boundary; used to separate dev/test/prod	`Contoso-Prod-Sub`
Resource group	Logical container for related resources	`rg-customer-support-app`
Resource	Individual services like VMs, storage accounts, or AI services	`storageacct01`, `foundry-resource`

Think of the hierarchy as: tenant → subscription → resource group → individual resources.

A Microsoft Azure diagram titled "A Cloud Platform for Modern Applications" showing a tenant and Microsoft Entra ID boundary, subscriptions and resource groups that organize resources like compute, networking, databases and AI services. It visualizes identity/security and billing/governance layers.

Microsoft Foundry — a unified AI platform

Microsoft Foundry is a unified platform that brings models, compute, storage, and AI services together so developers can build, test, deploy, and manage complete AI applications and agents. Foundry supports the full AI application lifecycle:

Development — configure models, tune behavior, and assemble application components.
Deployment — publish AI capabilities to users or downstream services.
Management — monitor, update, scale, and observe applications in production.

A Foundry resource exposes four capability areas: model access, compute, storage, and AI services. Those capabilities are assembled within Foundry projects to produce end-to-end AI solutions.

A presentation slide titled "Microsoft Foundry: A Unified Platform for AI Applications and Agents" showing nested boxes for "Microsoft Foundry Resource" (model access, compute, storage, AI services) and "Microsoft Foundry Project" (an isolated development environment). Inside the project box are three colored buttons labeled "Building," "Testing," and "Managing," with a small "© Copyright KodeKloud" at the bottom.

Foundry capability areas (purpose and examples):

Capability	Purpose	Example
Model access	Connect apps to models for inference or generation	LLMs for chat, summarization
Compute	Run model inference or fine-tuning workloads	GPU clusters, managed inference nodes
Storage	Persist datasets, files, and configuration	Blob storage, managed datasets
AI services	Specialized building blocks	Speech-to-text, translation, NER APIs

Foundry projects and the core building blocks

Inside a Foundry resource, you create Microsoft Foundry projects — isolated development environments for building, testing, and managing AI applications and agents. Projects are useful for experimentation and iterative development before broader release. Within a Foundry project, four core components combine to form production AI applications:

Models — the AI systems that process inputs and produce outputs (e.g., language models).
Agents — orchestrators that use models, tools, and knowledge to accomplish multi-step tasks.
Tools — connectors that let agents interact with external systems or APIs in a controlled, auditable way.
Knowledge — domain-specific data (documents, databases, knowledge bases) that grounds model responses in accurate, organization-specific information.

A slide titled "Microsoft Foundry: A Unified Platform for AI Applications and Agents" showing core components. It highlights four boxes—Models (browse, evaluate, fine‑tune, deploy), Agents, Tools, and Knowledge.

Models provide the intelligence. Agents combine model reasoning with planning and action. Tools are the safe interfaces to external systems. Knowledge ensures replies are grounded in your enterprise context.

A slide titled "Microsoft Foundry: A Unified Platform for AI Applications and Agents" showing core components — Models, Agents (highlighted), Tools, and Knowledge — with a description that agents use models to reason, plan, and complete tasks.

Tools are intentionally scoped: they enable agents to perform actions such as searching a knowledge store, calling an API, or updating a record — but only through controlled, auditable operations. Knowledge sources (SharePoint, internal docs, databases, product catalogs) provide the context that prevents hallucinations and aligns outputs with company policies.

A slide titled "Microsoft Foundry: A Unified Platform for AI Applications and Agents" showing core components—Models, Agents, Tools—and a highlighted "Knowledge" box that says "Connect agents to enterprise data for grounded responses."

Example — customer support agent:

Model: understands and parses natural language customer requests.
Agent: decides steps (e.g., check order, verify policy, initiate refund).
Knowledge: product and refund policies (e.g., refund within 7 days).
Tools: fetch order details and call the order management API to issue a refund.

This composition enables end-to-end automation or a hybrid human-in-the-loop workflow.

Accessing Foundry models and services: endpoints and SDKs

Applications access Foundry models and AI services via secure endpoints — HTTPS URLs that act as the front door to a model or service. To call an endpoint you need:

Where to connect — the endpoint URL (the endpoint).
How to authenticate — the credential or identity used to prove access.

Foundry supports authentication options such as Microsoft Entra ID tokens (recommended for managed identity-based access) and API keys (simple secret-based access). Always protect credentials and use secure secret stores.

Protect authentication credentials: store API keys and tokens in secure stores (Key Vault, managed identities, or other secret stores) and avoid embedding them in application code or public repositories.

When calling an endpoint, interactions typically follow a request/response pattern:

Build a request with the operation, required headers (including authentication), and JSON payload as needed.
Send the request to the endpoint over HTTPS.
Receive a response with the HTTP status and JSON results.

Example HTTP request payload (simplified):

POST https://your-foundry-endpoint.azure.com/v1/infer
Authorization: Bearer <ACCESS_TOKEN>
Content-Type: application/json

{
  "model": "your-model-id",
  "input": "Summarize the following customer complaint..."
}

SDKs make these interactions easier by handling authentication, request construction, retries, and response parsing for you. Use language-specific SDKs (Python, JavaScript, etc.) to reduce boilerplate and improve developer productivity. SDK benefits:

Benefit	What it provides
Authentication helpers	Managed identity, token acquisition, and key handling
Request helpers	Build and send JSON requests, handle retries
Response parsing	Deserialize results into native objects
Higher-level API	Shorter, idiomatic code vs. raw REST calls

A diagram titled "Microsoft Foundry Endpoints: Access Models and AI Services" showing a central endpoint (https://xxxx) used with either a Microsoft Entra ID token or an API key. The endpoint connects to an SDK that handles Request details (operation, headers, JSON body) and Response details (status, headers, JSON results).

Next steps: explore in the Azure Portal

Now that you understand tenants, subscriptions, resource groups, Microsoft Entra ID, and the Microsoft Foundry concepts (resources, projects, models, agents, tools, knowledge, endpoints, and SDKs), the next step is to explore these resources in the Azure Portal or the Foundry UI. Inspect a Foundry resource, create a project, and try connecting a sample application to a model endpoint using an SDK and a managed identity. Links and references

Microsoft Entra ID overview: https://learn.microsoft.com/entra/identity
Azure resource organization: https://learn.microsoft.com/azure/architecture/cloud-adoption/
Azure Key Vault: https://learn.microsoft.com/azure/key-vault/
Microsoft Foundry (platform docs) — consult your subscription or tenant documentation for the Foundry UI and API endpoints.

AI in Azure