AI-901 Azure AI Fundamentals

Course overview

AI-900 asked you to describe AI. AI-901 asks you to build with it. Microsoft retired AI-900 on June 30, 2026; AI-901 earns the same Azure AI Fundamentals certification, but 55–60% of it is hands-on implementation in Microsoft Foundry, and Microsoft expects you to read basic Python. This course gets you there even if you've never written code.

Exam at a glance

Already passed AI-900? Your certification stays valid — Fundamentals certifications don't expire. Take AI-901 if you want the hands-on Foundry skills employers are now asking for.

Learning outcomes

1. Explain the six Microsoft responsible AI principles and apply them to a real design.
2. Explain how generative models work: tokens, embeddings, transformers, training vs. inference, grounding.
3. Choose a model by capability, and choose deployment options and parameters (temperature, top-p, max tokens).
4. Identify AI workloads: generative, agentic, text analysis, speech, computer vision, image generation, information extraction.
5. Write effective system and user prompts.
6. Deploy a model in the Foundry portal and build a lightweight Python chat client.
7. Create, test and call a single agent.
8. Build lightweight text-analysis, speech, vision and image-generation apps.
9. Extract information from documents, images, audio and video with Azure Content Understanding.

Syllabus

Assessment

Module checks 20% · 6 builds with evidence 45% · Capstone 20% · Practice exam 15%.

What you need

• An Azure subscription with permission to create a Foundry resource and project (free account/credits or Azure for Students). Model usage is billed per token; this course's labs typically cost a few dollars total. Set a budget alert first.
• Python 3.10+ and VS Code (or GitHub Codespaces), Azure CLI for az login.

Module 0 — Python primer for AI builders

Microsoft says AI-901 candidates need Python syntax knowledge. You need to read code like this and understand what it does — not write a framework.

# variables and strings
name = "RoSeé"
greeting = f"Hello, {name}!"          # f-string inserts variables

# lists and dictionaries
messages = [
    {"role": "system", "content": "You are a helpful tutor."},
    {"role": "user", "content": "What is the cloud?"},
]
print(messages[1]["content"])          # -> What is the cloud?

# functions
def ask(question: str) -> str:
    return f"You asked: {question}"

# loops and conditions
for m in messages:
    if m["role"] == "user":
        print("User said:", m["content"])

# imports and environment variables (keep secrets OUT of code)
import os
endpoint = os.environ["AZURE_AI_ENDPOINT"]

# error handling
try:
    result = 10 / 0
except ZeroDivisionError as e:
    print("Caught:", e)

Setup once:

python -m venv .venv
source .venv/bin/activate          # Windows: .venv\Scripts\activate
pip install openai azure-identity azure-ai-projects azure-ai-textanalytics azure-cognitiveservices-speech python-dotenv
az login

Dope Translation: Python is the recipe. Variables are labeled containers. A list is a line of containers. A dictionary is a container with labeled compartments — like a pill organizer: "Monday: …, Tuesday: …". A function is a recipe you can call by name instead of repeating every step.