Skip to content
L-SYSTEM

recommendedbeginner

AI-901: Microsoft Azure AI Fundamentals — Exam Prep with Microsoft Foundry

A preparation course for Microsoft exam AI-901 (Microsoft Azure AI Fundamentals), written for software engineers in Big 4 and other professional-services firms who already write SQL, C# and Python and now need a working grasp of AI on Azure — and the certificate that proves it. AI-901 replaced AI-900 in 2026 and changed its character: more than half of the exam is about building with Microsoft Foundry — deploying models, writing prompts, building agents, and calling text, speech, vision and Content Understanding capabilities from lightweight Python apps. The course follows the official skills outline (April 2026 revision) domain by domain, with the same weights: AI concepts and capabilities (40–45%) and implementing AI solutions with Foundry (55–60%). Every topic is taught the same way: the problem that makes you reach for the capability, the capability itself, and what it costs — in tokens, quotas, latency, data residency and risk. Hands-on labs walk through the Foundry portal and the Python SDK step by step; if your corporate subscription blocks Foundry, each lab has a read-the-code path so you can still practise every exam objective. The running case is Aldervane Advisory LLP, a fictional professional-services firm building an internal Service Desk agent, extracting data from invoices and timesheets, and transcribing client meetings — where confidentiality of client data is never optional. Practice uses the question formats of the real exam: single and multiple choice, matching, Yes/No statement sets and code completion. Visual sketchnotes summarise each concept on one colourful page. Optional units cover the same apps in C#, quotas and cost, evaluation and tracing, and client-data governance.

Path through the course

· 24 units

any order · units can be taken in any order you likeView: map · steps

  1. ~33.8 h

  2. Exam

    closed · enrol on the course

    • final · in the browser, without the assistant

      Exam

The path in words
  • Exam map and the Foundry landscape — ~1.3 h
  • AI workloads: from scenario to workload — ~1 h
  • How generative AI models work — ~1 h
  • Choosing a model in the Foundry catalog — ~1 h
  • Deployment options and configuration parameters — ~1.3 h
  • Responsible AI: six principles and Foundry guardrails — ~1.3 h
  • Text analysis: key phrases, entities, sentiment, summaries — ~1 h
  • Speech: recognition, synthesis, translation and audio models — ~1 h
  • Computer vision and image generation — ~1 h
  • Information extraction with Azure Content Understanding — ~1 h
  • Prompt engineering: system and user prompts that work — ~1.5 h
  • Lab: deploy a model and work with it in the Foundry portal — ~1.5 h
  • Lab: a lightweight chat client with the Foundry SDK (Python) — ~2 h
  • Build and test a single agent in the Foundry portal — ~2 h
  • Lab: a lightweight client app for the ENGAGE helpdesk agent — ~1.5 h
  • Lab: Build a lightweight app with text analysis — ~1.5 h
  • Lab: Speech — spoken prompts and the Azure Speech SDK — ~1.5 h
  • Lab: Vision — images in, images out — ~1.5 h
  • Lab: Content Understanding — documents, images, audio and video — ~1.5 h
  • Exam rehearsal: logistics, strategy and a mixed practice set — ~2.5 h
  • Client-data governance in Foundry (Aldervane) — optional; ~1.5 h
  • The same apps in C#/.NET — optional; ~1.5 h
  • Quotas, tokens, cost and monitoring — optional; ~1.5 h
  • Evaluation, tracing and red teaming in Foundry — optional; ~1.5 h

Programme · 24 units

1Exam map and the Foundry landscape

Describe the two AI-901 domains, their weights and the exam format (700 pass mark, no labs)

~1.3 h
2AI workloads: from scenario to workload

Name the six AI workloads in AI-901 and the input and output of each

~1 h
3How generative AI models work

Explain tokens and why billing, quotas and context windows are counted in tokens

~1 h
4Choosing a model in the Foundry catalog

Distinguish models sold directly by Azure from models from partners and community

~1 h
5Deployment options and configuration parameters

Compare serverless API and managed compute deployment options

~1.3 h
6Responsible AI: six principles and Foundry guardrails

Map a short scenario to the right responsible AI principle: fairness, reliability and safety, privacy and security, inclusiveness, transparency or accountability

~1.3 h
7Text analysis: key phrases, entities, sentiment, summaries

Choose the text analysis technique for a scenario: language detection, key phrase extraction, named entity recognition, entity linking, PII detection, sentiment analysis, opinion mining or summarization

~1 h
8Speech: recognition, synthesis, translation and audio models

Distinguish speech recognition (speech to text) from speech synthesis (text to speech) and speech translation in a scenario

~1 h
9Computer vision and image generation

Choose the computer vision task for a scenario: image classification, object detection, segmentation, OCR, captioning and tagging, face detection or face recognition

~1 h
10Information extraction with Azure Content Understanding

Distinguish OCR, layout analysis and field extraction, and say why OCR alone does not produce structured data

~1 h
11Prompt engineering: system and user prompts that work

Distinguish the system prompt from the user prompt and decide where a given rule belongs

~1.5 h
12Lab: deploy a model and work with it in the Foundry portal

Deploy a model from the Foundry model catalog and tell a model from a deployment

~1.5 h
13Lab: a lightweight chat client with the Foundry SDK (Python)

Build a Python chat client with DefaultAzureCredential, AIProjectClient and get_openai_client

~2 h
14Build and test a single agent in the Foundry portal

Explain what a Foundry prompt agent is: a model, instructions and tools packaged under a name

~2 h
15Lab: a lightweight client app for the ENGAGE helpdesk agent

Call an existing Foundry agent from Python with an agent_reference in responses.create

~1.5 h
16Lab: Build a lightweight app with text analysis

Choose between Azure Language in Foundry Tools and a deployed generative model for a text analysis scenario

~1.5 h
17Lab: Speech — spoken prompts and the Azure Speech SDK

Choose between an Azure Speech pipeline, a multimodal audio model and Voice Live for a voice scenario

~1.5 h
18Lab: Vision — images in, images out

Tell apart a multimodal chat model (image in, text out) and an image generation model (text in, image out) in a scenario

~1.5 h
19Lab: Content Understanding — documents, images, audio and video

Explain analyzers, schemas and fields in Azure Content Understanding in Foundry Tools, and choose prebuilt or custom analyzers

~1.5 h
20Exam rehearsal: logistics, strategy and a mixed practice set

Describe the AI-901 exam format: time, question types, passing score, breaks, retakes and scheduling

~2.5 h
21Client-data governance in Foundry (Aldervane)

Choose a deployment type that meets a client's data-processing residency requirement

~1.5 h
22The same apps in C#/.NET

Map the Python Foundry, Language and Speech packages to their .NET equivalents

~1.5 h
23Quotas, tokens, cost and monitoring

Explain how TPM quota and RPM limits are granted and assigned to deployments

~1.5 h
24Evaluation, tracing and red teaming in Foundry

Choose between evaluation, tracing and red teaming for a given question

~1.5 h

Recipes of this course

all recipes of the course (1)