Artificial Intelligence (AI) refers to systems designed to perform tasks that typically require human intelligence such as reasoning, problem-solving, perception, and language understanding. 

An AI application is a software solution that uses AI techniques such as computer vision, speech, and information extraction—to perform tasks that typically require human-like intelligence. These applications can understand, reason, learn, and respond to inputs in a way that feels more adaptive than traditional software. 

AI applications are powered by machine learning (ML) models, which are mathematical systems trained to recognize patterns in data and make predictions or generate outputs. ML models are the engines inside an AI application. When you interact with an AI application, the model performs inference, meaning it applies what it learned during training to new input.

AI applications are:

• Model-powered: They use trained models to process inputs and generate outputs, such as text, images, or decisions.

• Dynamic: Unlike static programs, AI apps can improve over time through retraining or fine-tuning.

Some examples of AI applications for different industries include:

• Healthcare: AI-powered diagnostic tools that analyze medical images (such as X-rays or MRIs) and help doctors detect diseases more accurately and quickly.

• Finance: Fraud detection systems that use AI to monitor transactions in real time and identify suspicious activity, helping prevent financial crimes.

• Retail: Personalized recommendation engines that analyze customer behavior and preferences to suggest products, improving the shopping experience.

• Manufacturing: Predictive maintenance solutions that use AI to monitor equipment and forecast when machines are likely to fail, reducing downtime and maintenance costs.

• Education: Intelligent tutoring systems that adapt to each student’s learning style and pace, providing customized feedback and support to enhance learning outcomes.

In this module, you learn how Microsoft enables you to build AI applications with the latest technology, securely, and at scale. While the model is the engine, AI applications also need security, networking, hosting, data storage, application logic, and user interfaces. Microsoft provides all the infrastructure and services needed to support enterprise-scale AI development. The module gives you a foundation in how Azure streamlines AI application development, integrates with Microsoft Foundry, and enables rapid innovation.

Developing AI apps on Azure

The AI applications we use and trust today rely on robust security, networking, hosting, scaling, data storage, and AI capabilities. Azure gives you everything you need to build scalable and secure AI applications quickly and using common technologies of your choice. 

Security and networking

The foundation of AI applications is security and networking. Azure is secure by design, offering built in identity, access control and network isolation to protect your AI solutions.

AI applications require strong protection and stable connectivity. Azure's security tools protect your AI applications from unauthorized access and threats. For example, Azure Entra ID ensures that only the right people and services can access your AI resources. It enables role-based access control (RBAC) to limit access to model deployments, resources, and data.

Azure also protects an AI application's secrets. A secret is any sensitive value that your AI application must keep hidden because it grants access to a system, service, or data. Secrets can include API keys, database connection strings, OAuth tokens, passwords, and more. A key is a type of secret that is usually a long, randomly generated string—that authenticates your request when you call an endpoint (a URL). In Azure, secrets are typically stored in Azure Key Vault, not in code or GitHub.

For example, suppose you're building an AI chatbot with Azure.

1. Your application calls the model's endpoint.

2. Your request includes a key to authenticate the call.

3. The key is stored in Azure Key Vault as a secret.

4. Your application retrieves the secret at runtime using a secure method (managed identity).

Azure's security tools cover identity, secret, and data protection, compliance, threat detection, monitoring, and security controls such as firewalls. Its networking services ensure the application runs reliably, privately, and efficiently across cloud and hybrid environments.

Hosting and scaling

Applications run on computers or environments known as a host. In cloud contexts, a host can be a virtual machine (VM), providing the compute, memory, and networking your application needs to execute.

To host and scale your applications, you can use Azure Kubernetes Service (AKS) for containerized workloads or Azure App Service to modernize and deploy web apps quickly. AKS orchestrates, or manages, a large number of containers, which hold what your code needs to run. Azure App Service hosts web applications, APIs, and background jobs. These services accelerate time to market while ensuring reliability.

Scaling your applications means to automatically or manually adjust the amount of compute power your app uses—usually by adding or removing instances. In cloud applications, 'instances' refer to copies of your application running at the same time.

Two types of scaling:

• Scale out (horizontal): Add more instances

• Scale up (vertical): Increase CPU/memory on the existing instance

Cloud platforms like Azure make scaling automatic based on CPU usage, number of requests, or custom metrics.

Data storage

Models : Foundry supports thousands of models—including rich first-party, third-party, and open-source options—directly from its unified model catalog. Developers can access Azure-hosted OpenAI models such as the latest GPT‑5 family (GPT‑5, GPT‑5-mini, GPT‑5-nano, GPT‑5-chat/5.2-chat) with extensive multimodal and reasoning capabilities, alongside specialist models from Anthropic (e.g., Claude Opus 4.6/4.5, Sonnet), Mistral, Cohere, Meta LLaMA, DeepSeek, xAI’s Grok, Black Forest Labs, and gated (enterprise-governed) Hugging Face models.

Users can browse thousands of models—ranging from massive foundation models to lightweight, domain-specific variants—evaluate them via built-in leaderboards and playgrounds, and manage deployments directly in Foundry. Full lifecycle support enables deployment per region, customizable deployment types (standard, provisioned, batch), version control, and governance support with Responsible AI and content safety.

Agents : At the core of Microsoft Foundry is an agent‑first approach that lets developers build intelligent, task‑oriented agents directly within their Foundry projects. These agents can reason over inputs, call tools, interact with data, and automate workflows using the platform’s built‑in orchestration. Foundry handles the underlying coordination—including message threading, tool execution, safety controls, and observability—so developers can focus on designing the agent’s goals and capabilities. Using either low‑code or code‑first workflows, teams can create multi‑agent systems that work with project resources such as documents, datasets, search indexes, and connections to external systems, including integrations like Azure Functions or Microsoft Fabric.

Tools : Foundry offers a comprehensive suite of Azure services—such as speech, vision, language, document intelligence, and more. These Foundry Tools provide AI capabilities that can be built into web or mobile applications in a way that's straightforward to implement. There're over a dozen different services that can be used separately or together to add AI power to applications. For example, you could use Azure Vision to analyze images, Azure Language to summarize text, classify information, or extract key phrases, and Azure Speech to convert speech to text and text to speech.

Knowledge : Foundry IQ provides a permission‑aware, multi‑source knowledge layer that gives agents accurate, grounded answers using an organization’s own data. It lets you create a configurable knowledge base made up of internal and external knowledge sources—such as Azure Blob Storage, SharePoint, OneLake, or public web data—and automatically handles indexing, document chunking, vector embeddings, and metadata extraction. When an agent queries the knowledge base, Foundry IQ uses agentic retrieval to break the question into subqueries, search multiple sources in parallel, and return relevant, citation‑backed information while enforcing user permissions and Microsoft Purview sensitivity labels. This ensures that agents can draw from trusted, up‑to‑date content and only return information the user is authorized to see, providing a reliable knowledge foundation for enterprise AI workflows.

The assets for your AI solution are organized within a project. Each project is contained within a Foundry resource, which provides model hosting and the services your apps and agents need in Azure.

Foundry resources and projects

To get started with Foundry, you need to create a Foundry resource, which provides model hosting and the services your apps and agents need. You can create a Foundry resource in the Azure portal, Foundry portal, or programmatically with scripting.

A Foundry resource is the Azure resource that provides the platform capabilities. A Foundry resource provides access to:

• Models (Microsoft, partner, and OpenAI‑compatible)

• Foundry’s agent service

• Deployment governance

• Monitoring & observability

• Security boundaries

• Quotas and operational controls

A Foundry project is a workspace inside that resource where you build AI apps, agents, and evaluations. A Foundry Project lets you build and manage:

• Agents

• Evaluations

• Files and datasets

• Vector indexes

• Flows (AI logic)

• Connections

• Project‑specific settings

You might have one Foundry resource for a team or department, and many Foundry projects inside it, each focused on a separate AI use case.

Foundry portal

The Foundry portal provides a modern web-based interface for developing, testing, and operating AI solutions. This is where you'll spend a lot of your time when working with models, agents, and other assets.

In the new Foundry portal you can discover models and tools, build agents, manage the operation of those agents, and much more. At any time, you can get help with Ask AI agent helper. The Ask AI experience uses specialized sub‑agents to answer questions and help with tasks across Microsoft Foundry. It can guide you through documentation, explain model catalog capabilities, troubleshoot issues, and manage model deployments, quotas, and operations. It also compares and analyzes models, interprets monitoring dashboards, and supports end‑to‑end evaluation workflows for language models and agents. 

Using Foundry portal for application development

When you're building applications on Azure, Foundry provides a powerful and versatile platform for development. A general name for applications (that may or may not have AI capabilities) is client applications. A client application is a program that a user interacts with on their device (like a phone, laptop, or browser) that sends requests to a server and displays the results.

Consider the following workflow for using Foundry portal to develop an AI application:

1. Sign into Foundry portal using your Azure subscription and create a Foundry project.

2. In Foundry, pick a model from the Model Catalog and deploy it.

3. In Foundry, experiment with the model in the Playground. You can use the Playground to write prompts, test model responses, configure parameters. 

4. Use the configured model in your own client application.

An AI client application utilizes a model, data, and application logic to process and return results. The app logic is the code or workflow that sends requests to the model, receives the response, and processes and transforms results. The entire process is known as a client–server interaction. A client-server system provides the foundation for how users interact with AI systems, how requests are processed, and how results are delivered.

Responsibilities of the client:

1. Present a UI or CLI

2. Collect user input (text, voice, images)

3. Format the input into a prompt or API request

4. Send a request to the server (model endpoint)

5. Display the returned output

The client requests the model for results, which is hosted by the server, or the back end. In Foundry, the server is your model deployment.

Responsibilities of the server:

1. Receive the prompt

2. Run inference on the model

3. Apply system instructions, safety, context, and more

4. Return the generated output (for example: text, image, audio, or structured JSON)

Next, let's take a look at how clients connect to Foundry models and how to use Foundry endpoints.

Using Microsoft Foundry endpoints

In Foundry, you can define the models and agents that you want to use in custom AI applications. Since Foundry resources are cloud-based, you can consume them as Application Programming Interfaces (APIs) across internet connections through programmatic interfaces. 

Understand endpoints

Like most cloud services, Microsoft Foundry resources are accessed through an API endpoint, representing a service entry point. The endpoint has a unique HTTP address, like a website, but it's for client application code rather than human users with a web browser. When you view the endpoint for your model, it looks something like:

https://<foundry-project>-resource.cognitiveservices.azure.com/openai/deployments/gpt-4o/chat/completions?api-version=2024-05-01-preview

The interfaces provided at the endpoint are known as Representational State Transfer Interfaces, or REST interfaces for short.

To keep your Foundry resources secure, the endpoint is protected. Applications can only access it if they present the correct API key or a token confirming that your Microsoft Entra ID credentials are valid. The model endpoint and key can be found in the Foundry Playground's details page.