Training course

This is a free, self-paced reading course for AI-900: Microsoft Azure AI Fundamentals, the fundamentals-level credential that asks you to recognize AI workload types, match them to the right Azure service, and reason about Microsoft’s Responsible AI principles. There is no coding in it. The exam does not require Python, deep learning experience, or statistics; it requires clean vocabulary and the judgment to pick the smallest service that solves a stated business problem. This course is written to build exactly that.

The course is organized as one module per official exam domain, in the order Microsoft publishes them, and each module carries the domain’s published weight. On AI-900 that weighting is not cosmetic. Generative AI is the single heaviest domain at 20-25%, and because the exam is short, a shaky module 5 can cost you eight or nine questions on its own. Reading in domain order with the weights visible tells you where the time belongs before you have wasted any of it.

AI-900 Fundamentals level 5 modules Domain-weighted Self-paced Free account

What the course covers

Artificial intelligence workloads and considerations

Module 1 · 15-20%

The common AI workload types, prediction and forecasting, anomaly detection, computer vision, natural language processing, document intelligence, knowledge mining, and generative AI, and how to match each to a real business scenario. The module then works through Microsoft’s six Responsible AI principles: fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability, and how each one surfaces in the design and operation of a system. Scenario stems here usually turn on which principle is most directly at risk.

Fundamental principles of machine learning on Azure

Module 2 · 15-20%

Regression, classification, and clustering and what separates them; features, labels, training and validation datasets, and supervised versus unsupervised learning. The module covers reading a confusion matrix and interpreting the standard evaluation metrics (accuracy, precision, recall, F1, and ROC/AUC for classification; MAE, RMSE, and R-squared for regression) then turns to the Azure tooling: Azure Machine Learning studio, Designer’s drag-and-drop pipelines, Automated ML, and notebooks, and when each is the right choice. Vocabulary carries more weight here than math.

Computer vision workloads on Azure

Module 3 · 15-20%

The core vision tasks and the output each one produces: image classification, object detection, semantic segmentation, OCR, and the distinction between face detection, face verification, and face recognition. The module maps those tasks onto the services: Azure AI Vision for prebuilt image analysis and OCR, Azure AI Custom Vision for training on your own labeled images, Azure AI Face for face-specific work, and Azure AI Document Intelligence for structured extraction from invoices, receipts, and forms, because most items in this domain come down to picking between services that all sound plausible.

Natural language processing workloads on Azure

Module 4 · 15-20%

The Azure AI Language feature set, sentiment analysis, key phrase extraction, named entity recognition, PII detection, language detection, custom question answering, and conversational language understanding for intent and entity extraction, alongside Azure AI Translator for text and speech translation and Azure AI Speech for text-to-speech, speech-to-text, and speaker recognition. The module practices the pairing the exam actually asks for: a business scenario such as categorizing reviews, redacting identifiers, or routing by language, and the smallest service that solves it.

Generative AI workloads on Azure

Module 5 · 20-25%

The heaviest domain on the current exam. What generative AI is and how large language models work conceptually, tokens, prompts, embeddings, temperature, and the transformer idea; Azure OpenAI Service and its GPT and DALL-E model families; grounding and retrieval-augmented generation as the standard defense against hallucination; and the Microsoft Copilot lineup, covering Microsoft 365 Copilot, Copilot Studio, GitHub Copilot, and Copilot in Windows, plus Azure AI Foundry as the platform for building and evaluating generative applications. Content filters and responsible generative AI practices are tested consistently.

How to use it

Read a module, then test that domain immediately on the AI-900 practice exam rather than saving all the questions for the end. AI-900 rewards recognizing a stem pattern, and the gap between having read about Azure AI Custom Vision and being able to choose it over Azure AI Vision under time pressure only shows up when you answer questions. Because scenario stems routinely cross domain boundaries (a policy chatbot touches both NLP and generative AI, a defect-detection kiosk touches vision and responsible AI) a second pass over the whole course reads differently from the first, and is worth the time. The course modules themselves open once you are signed in to a free Certifym account.

Start with module 5 if you are short on time and confident everywhere else; it carries the most marks and it is the newest material. For exam logistics (item count, timing, the 700-of-1000 passing threshold, and who the credential is aimed at) see the AI-900 certification guide.

AI-900: Microsoft Azure AI Fundamentals

One module per AI-900 exam domain, weighted the way Microsoft weights them.

A self-paced reading course for AI-900: Microsoft Azure AI Fundamentals, the fundamentals-level credential that asks you to recognize AI workload types, match them to the right Azure service, and reason about Microsoft's Responsible AI principles. There is no coding in it. The exam does not require Python, deep learning experience, or statistics; it requires clean vocabulary and the judgment to pick the smallest service that solves a stated business problem.

The course is organized as one module per official exam domain, in the order Microsoft publishes them, and each module carries the domain's published weight. You work through the AI workload types and the six Responsible AI principles, the machine learning vocabulary and the evaluation metrics behind it, computer vision on Azure, natural language processing and speech, and finally generative AI — the heaviest domain on the current exam, covering language models, prompts, grounding and retrieval-augmented generation, Azure OpenAI Service, Azure AI Foundry, and the Microsoft Copilot family.

Every lesson ends with its key terms and further reading, so the course works as a reference after the first pass as well as a syllabus during it. Certifym is an independent study publisher and is not affiliated with, endorsed by, or sponsored by Microsoft Corporation.

Module 1: Artificial intelligence workloads and considerations 19% of exam

The vocabulary module. Before you can pick an Azure service you have to name the workload correctly, and most of the service-selection questions on AI-900 are really workload-identification questions wearing a costume. This module walks the workload types Microsoft publishes — prediction and forecasting, anomaly detection, computer vision, natural language processing, document intelligence, knowledge mining, and generative AI — and gives you a scenario test for each.The second half is Microsoft's six Responsible AI principles: fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability. Exam stems here rarely ask you to define a principle. They describe a system behaving badly and ask which principle is most directly at risk, so each principle is taught with the failure that signals it.

1.1 What artificial intelligence is, and what it is not 5 min Free preview
1.2 Prediction and forecasting workloads 5 min 🔒
1.3 Anomaly detection workloads 5 min 🔒
1.4 Computer vision workloads 5 min 🔒
1.5 Natural language processing workloads 5 min 🔒
1.6 Document intelligence and knowledge mining 5 min 🔒
1.7 Generative AI as a workload type 5 min 🔒
1.8 Responsible AI: fairness, reliability and safety 5 min 🔒
1.9 Responsible AI: privacy and security, inclusiveness 5 min 🔒
1.10 Responsible AI: transparency and accountability, and applying the six 5 min 🔒

Module 2: Fundamental principles of machine learning on Azure 19% of exam

The vocabulary-and-judgment module. AI-900 does not ask you to do mathematics, but it does ask you to use the terms precisely: features and labels, training and validation, supervised and unsupervised, regression against classification against clustering. This module builds that vocabulary and then teaches you to read the evaluation output — a confusion matrix, precision against recall, and the regression metrics — well enough to say whether a model is fit for a stated purpose.The second half is the Azure tooling. Azure Machine Learning studio and the workspace it sits on, automated machine learning for finding a good model without writing code, Designer for drag-and-drop pipelines, and notebooks for full control. The exam's recurring question is which of those four a described team should use, so each is taught by what it assumes about the person using it.

2.1 How machine learning works: features, labels, and training 5 min 🔒
2.2 Supervised and unsupervised learning 5 min 🔒
2.3 Regression 5 min 🔒
2.4 Classification 5 min 🔒
2.5 Clustering 5 min 🔒
2.6 Deep learning and neural networks, conceptually 5 min 🔒
2.7 Evaluating classification models: the confusion matrix and its metrics 5 min 🔒
2.8 Evaluating regression models: MAE, RMSE, and R-squared 5 min 🔒
2.9 Azure Machine Learning studio and the workspace 5 min 🔒
2.10 Automated ML, Designer, and notebooks: choosing an authoring approach 5 min 🔒

Module 3: Computer vision workloads on Azure 19% of exam

The service-selection module. Most computer vision items on AI-900 present a business scenario and four Azure services that all sound plausible, and the decision turns on two questions: what shape of output does the scenario need, and are the categories generic or specific to this organization. This module teaches the vision tasks by their output — classification, object detection, segmentation, OCR — and then maps them onto Azure AI Vision, Azure AI Custom Vision, Azure AI Face, and Azure AI Document Intelligence.Face work gets its own lesson because the exam distinguishes detection, verification, and recognition carefully, and because Microsoft restricts access to some facial capabilities on Responsible AI grounds. The module closes with a decision procedure you can run against any vision stem.

3.1 How computers process images 5 min 🔒
3.2 Image classification 5 min 🔒
3.3 Object detection 5 min 🔒
3.4 Segmentation and general image analysis 5 min 🔒
3.5 Azure AI Vision: the prebuilt service 5 min 🔒
3.6 Optical character recognition 5 min 🔒
3.7 Azure AI Custom Vision 5 min 🔒
3.8 Face detection, verification, and recognition 5 min 🔒
3.9 Azure AI Document Intelligence 5 min 🔒
3.10 Choosing the right computer vision service 5 min Free preview

Module 4: Natural language processing workloads on Azure 19% of exam

The language module, and the one whose boundary with generative AI the current exam probes hardest. It covers the Azure AI Language feature set — sentiment analysis, key phrase extraction, named entity and PII detection, language detection, summarization, custom classification, custom question answering, and conversational language understanding — alongside Azure AI Speech for transcription, synthesis, and speaker recognition, and Azure AI Translator for text and speech translation.The exam's habit in this domain is to describe a business situation and offer several services that could technically all work. The expected answer is the smallest service that solves the stated problem, so each lesson closes on the scenario language that identifies it, and the module ends with a decision procedure and the two-service pipelines that recur in stems.

4.1 How text is represented: tokens, frequency, and embeddings 5 min 🔒
4.2 Sentiment analysis and opinion mining 5 min 🔒
4.3 Key phrase extraction and summarization 5 min 🔒
4.4 Named entity recognition, PII detection, and entity linking 5 min 🔒
4.5 Language detection 5 min 🔒
4.6 Conversational language understanding 5 min 🔒
4.7 Custom question answering 5 min 🔒
4.8 Azure AI Speech: transcription, synthesis, and speaker recognition 5 min 🔒
4.9 Azure AI Translator 5 min 🔒
4.10 Choosing the right language service 5 min 🔒

Module 5: Generative AI workloads on Azure 24% of exam

The heaviest domain on the current exam and the newest material on it. This module builds generative AI from the ground up — what a language model does, what tokens and embeddings are, the transformer idea, and how prompts, system messages, and generation settings steer behavior — then turns to the Azure products: Azure OpenAI Service, Azure AI Foundry, and the Microsoft Copilot family including Copilot Studio.The through-line is judgment rather than trivia. You are taught grounding and retrieval-augmented generation as the standard defense against hallucination, how to choose between prompt engineering, RAG, and fine-tuning, and what responsible generative AI requires in practice — content filters, evaluation, human oversight, and disclosure. Specific model names, context sizes, and prices move too fast to memorize and are deliberately not the subject here.

5.1 What generative AI is 5 min 🔒
5.2 Language models, tokens, and embeddings 5 min 🔒
5.3 The transformer idea 5 min 🔒
5.4 Prompts, system messages, and completions 5 min 🔒
5.5 Temperature and the other generation settings 5 min 🔒
5.6 Azure OpenAI Service 5 min 🔒
5.7 Grounding and retrieval-augmented generation 5 min 🔒
5.8 Prompt engineering, RAG, or fine-tuning: choosing an approach 5 min 🔒
5.9 Azure AI Foundry 5 min 🔒
5.10 The Microsoft Copilot family 5 min 🔒
5.11 Copilot Studio and building custom copilots 5 min 🔒
5.12 Responsible generative AI and content filters 5 min 🔒

Frequently asked questions about the AI-900 training course

Is the AI-900 training course free?

Yes. The course costs nothing to read and opens once you are signed in to a free Certifym account, no payment and no card.

How is the course structured?

One module per official AI-900 exam domain, in Microsoft’s published order, with each module weighted to the domain’s published percentage range. Within each module the material is broken into short lessons, followed by key terms and further reading.

Does this replace Microsoft’s official training?

No. Microsoft Learn publishes free official learning paths for AI-900, and they are the authoritative source. This course is an independent study companion, written to be read quickly and to slot alongside practice questions, not a substitute for the official study guide.

Do I need a technical background before starting?

No. AI-900 is aimed at technical and non-technical candidates alike, and it has no prerequisite certification. The course assumes no Python, no deep learning experience, and no statistics, only that you are comfortable reading about software and business scenarios. If you have never used Azure at all, the AZ-900 fundamentals material gives you the platform vocabulary first, though it is not required.

What should I do after finishing the course?

Move to the AI-900 practice exam and work until you are clearing the bar consistently in every domain rather than leaning on your strong ones. After the credential, the usual next steps are the role-based AI tracks, Azure AI Engineer or Azure Data Scientist, or the AI-200 developer course if you are building on Azure rather than analyzing with it.

Is the course current with the latest AI-900 outline?

The course is built against the outline as it stands after the January 2026 refresh, which added a substantial block on generative AI, Azure OpenAI Service, and the Microsoft Copilot products, now the heaviest domain at 20-25%. Material written before that refresh misses a large portion of what is on the exam. Microsoft can revise the outline at any time; verify the current study guide on Microsoft Learn before you sit the exam.

Trademark notice & independence. Certifym.net is operated by Certifym Exam Services, LLC and is not affiliated with, endorsed by, or sponsored by Microsoft Corporation. Microsoft®, Azure®, Microsoft Certified, Microsoft 365 Copilot, and AI-900 are trademarks or registered trademarks of Microsoft Corporation. Use of these marks is solely to identify the certification for which these study materials are intended. AI-900 and the Microsoft Certified: Azure AI Fundamentals credential are administered solely by Microsoft, and the exam skills outline is the property of Microsoft Corporation; candidates should review the official, current study guide directly on Microsoft Learn.

All course content, questions, answers, and explanations on Certifym are original content created for study purposes. They are authored to align with the publicly documented AI-900 exam skills outline and are not actual Microsoft training materials or examination questions, nor are they represented as such. No material on this site is drawn from Microsoft’s proprietary item bank. Studying with these materials does not guarantee a passing result on any live certification exam. Exam requirements, format, domain weights, and renewal policies are set by Microsoft and may change; always verify current details on Microsoft Learn before scheduling your exam.