Plan, Create & Deploy a Microsoft Foundry Service

Terms introduced in this topic

These words will appear in exam questions about resource creation and deployment. Learn them here before the concepts below use them.

What to decide before provisioning any Azure AI resource

Exam questions often describe a scenario where a company has made a wrong planning decision and asks what should have been done differently. These are the decisions you must plan before creating a resource.

Step-by-step: Create an Azure AI Language resource

This walkthrough creates the Azure AI Language service — the one we used in Topic 1's Python example. The steps are almost identical for any Azure AI service.

How to store your API key safely

Before we write any code that uses your newly created resource, you need to know how to store your key safely. An environment variable is a value stored in your operating system — outside of your code — that your Python program can read. This way your key never appears in your script.

Step 1 — Set the environment variable

Run this in your terminal before running your Python script:

# On Windows Command Prompt:
set LANGUAGE_KEY=your-actual-key-here
set LANGUAGE_ENDPOINT=https://your-resource-name.cognitiveservices.azure.com/

# On Mac/Linux terminal:
export LANGUAGE_KEY=your-actual-key-here
export LANGUAGE_ENDPOINT=https://your-resource-name.cognitiveservices.azure.com/

Step 2 — Read the variable in Python

Python's built-in os module lets you read environment variables. The os module is part of Python itself — no installation needed.

# "import os" loads Python's built-in operating system module
# This module lets you interact with your operating system, including reading environment variables
import os

# os.environ is a dictionary of all environment variables on your system
# .get("LANGUAGE_KEY") retrieves the value stored under that name
# If the variable doesn't exist, .get() returns None instead of crashing
key = os.environ.get("LANGUAGE_KEY")

# Same for the endpoint URL
endpoint = os.environ.get("LANGUAGE_ENDPOINT")

# Now "key" and "endpoint" hold your values, but they never appear as text in your code
print(key)       # This would print your actual key at runtime — but only you can run this
print(endpoint)  # And your actual endpoint URL

How to install any Azure AI SDK

Every Azure AI service has its own Python SDK package you install with pip. Here are the packages the exam most commonly references — you need to recognise which package belongs to which service.

Complete setup script — Language service

Below is the complete, production-ready setup. Notice how it uses environment variables for the key and endpoint rather than hard-coding them. A new Python concept appears here: if/else — explained in the comments.

# ── Import modules ─────────────────────────────────────────────────────

# os gives us access to environment variables
import os

# sys lets us exit the program if something goes wrong
import sys

# The Azure Language SDK client
from azure.ai.textanalytics import TextAnalyticsClient

# The credential helper for API key authentication
from azure.core.credentials import AzureKeyCredential


# ── Read credentials from environment variables ────────────────────────

key      = os.environ.get("LANGUAGE_KEY")
endpoint = os.environ.get("LANGUAGE_ENDPOINT")


# ── Safety check (new concept: if/else) ───────────────────────────────

# "if" checks a condition. If the condition is True, the indented block runs.
# "not key" means "if key is empty or None (doesn't exist)"
# "or" means EITHER condition can trigger the error
if not key or not endpoint:
    # print() displays a message to the user
    print("Error: Please set LANGUAGE_KEY and LANGUAGE_ENDPOINT environment variables.")
    # sys.exit(1) stops the program with an error code (1 means something went wrong)
    sys.exit(1)


# ── Create the client ──────────────────────────────────────────────────

# Wrap the key in a credential object
credential = AzureKeyCredential(key)

# Create the client — our connection to the Language service
client = TextAnalyticsClient(endpoint=endpoint, credential=credential)

print("✅ Client created successfully. Ready to call the Language service.")

Models vs Services — and how to pick a model

For most Azure AI services (Language, Vision, Speech), Microsoft manages the underlying models for you — you just call the service and it uses the best available model automatically.

However for Azure OpenAI, you must explicitly choose and deploy a model. Here are the models you need to know for the exam:

Three ways to deploy an Azure AI service

The exam presents scenarios asking which deployment option best fits the requirements. There are three options and each has a clear use case.

Container deployment — what the exam tests

The exam frequently asks about running Azure AI service containers. There is a specific Docker run command pattern you must recognise. Below is the Language service sentiment container as an example.

# This is a Docker command (not Python) — you run it in your terminal
# "docker run" starts a new container from an image
# --rm  means: delete the container automatically when it stops (no cleanup needed)
# -it   means: interactive terminal (you can see output as it runs)
# -p 5000:5000  maps port 5000 on your machine to port 5000 in the container
#              This lets you call the service at http://localhost:5000

docker run --rm -it -p 5000:5000 \
  mcr.microsoft.com/azure-cognitive-services/textanalytics/sentiment \
  # ↑ The Docker image location — from Microsoft Container Registry (mcr.microsoft.com)
  Eula=accept \
  # ↑ REQUIRED: You must accept the licence agreement. Container won't start without this.
  Billing=https://your-resource.cognitiveservices.azure.com/ \
  # ↑ Your endpoint — used for billing reporting back to Azure (even offline containers bill)
  ApiKey=your-api-key
  # ↑ Your API key — proves this container is authorised to bill your subscription

ACI vs AKS vs Azure App Service — when to use which

Default endpoints and naming conventions

The exam shows you endpoint URLs and asks which service they belong to, or asks you to identify the correct endpoint for a given service. Learn the patterns.

Integrating Foundry Services into a CI/CD pipeline

CI/CD stands for Continuous Integration / Continuous Delivery. It's the practice of automatically testing and deploying your code every time you make a change. The exam wants you to know that this is possible with Azure AI services and which tools are used — not the deep implementation details.

Two CI/CD tools you need to know

High availability with multiple regions

A single Azure region can experience outages. For mission-critical applications, you deploy your Azure AI service to multiple regions and use traffic routing to send users to the nearest healthy instance.

What this topic is tested on

Test your understanding

5 questions. Select an answer to see the explanation immediately.