Azure AI Fundamentals Study Guide & Cheat Sheet

Microsoft Azure AI Fundamentals (AI-900) Exam Guide

Exam domains

Who it’s for: Technical and non-technical candidates who want a foundational grasp of AI and machine learning concepts and need to recognize which Azure AI service fits a given scenario - no coding required. AI-900 is the 2026 update that replaces AI-900 (which retires June 30, 2026), covering the same core skill areas (AI workloads and responsible AI, machine learning, computer vision, NLP, and generative AI) with refreshed, current Azure service names and a stronger generative-AI focus.

Study & test-day tips

• AI-900 is the 2026 refresh of AI-900 (retiring June 30, 2026) - use updated study materials and learn the current names: Azure AI services (formerly Cognitive Services), Azure AI Foundry (formerly Azure AI Studio), Azure AI Vision (formerly Computer Vision), Azure AI Language (formerly Text Analytics/LUIS), and Azure AI Document Intelligence (formerly Form Recognizer).
• Memorize Microsoft's six responsible-AI principles cold - fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability - and practice matching a one-line scenario to the right principle, the single most reliable point-scorer.
• Generative AI is the heaviest area (20-25%), so over-invest there: know tokens, prompts/completions, embeddings, the Transformer, Azure OpenAI Service (GPT, embeddings, DALL-E), and Azure AI Foundry's model catalog.
• Drill 'which service?' scenario questions - the exam constantly tests whether you pick Azure AI Vision vs. Document Intelligence vs. Face, or Azure AI Language vs. Translator vs. Speech - so learn each service by the problem it solves.
• Lock in ML vocabulary: distinguish regression (predicts a number) from classification (predicts a category) from clustering (groups unlabeled data), and know features vs. labels and training vs. validation data.
• Know what each Azure Machine Learning capability is for - automated ML (auto-tries algorithms), the designer (no-code drag-and-drop), compute/data assets, the model registry, and endpoints for deployment.
• Watch for distractors that mix up vision tasks: image classification (what is in the image) vs. object detection (what and where) vs. OCR (reading text) vs. facial detection - the wrong answer is usually a real feature applied to the wrong job.
• Pace yourself at roughly a minute per question; flag anything uncertain and return to it. There is no penalty for guessing, so never leave a question blank.
• Use the free Microsoft Learn AI Fundamentals learning paths and a sandbox of the Azure portal / Azure AI Foundry to see the services for real - recognizing the portal experience helps with terminology questions.
• Read each scenario for the key verb or noun ('extract text from invoices' = Document Intelligence, 'translate in real time' = Translator/Speech, 'generate an image from a prompt' = DALL-E in Azure OpenAI) and answer based on the service's primary purpose, not surface keywords.

Cheat sheet

AI workloads & responsible AI

• Six responsible-AI principles: fairness, reliability and safety, privacy and security, inclusiveness, transparency, accountability.
• Fairness = treat all groups equitably; inclusiveness = empower people of all abilities; transparency = users understand how/why the system decides; accountability = humans remain responsible for the system.
• Common AI workloads: computer vision, natural language processing, document/knowledge mining, and generative AI.
• Anomaly detection flags unusual data points; knowledge mining extracts insights from large volumes of unstructured content.
• Azure AI services (formerly Cognitive Services) is the umbrella for prebuilt vision, language, speech, and decision APIs.

Machine learning concepts

• Regression predicts a numeric value; classification predicts a category/label; clustering groups unlabeled data by similarity.
• Features are the input variables (X); the label is the value being predicted (Y).
• Training data builds the model; validation/test data measures how well it generalizes to unseen data.
• Deep learning uses multi-layer neural networks; the Transformer architecture underpins modern LLMs.
• Supervised learning uses labeled data (regression, classification); unsupervised learning uses unlabeled data (clustering).

Azure Machine Learning

• Automated ML (AutoML) automatically tries algorithms and hyperparameters to find the best model.
• The designer is a no-code, drag-and-drop canvas for building ML pipelines.
• Compute resources run training and inference; data assets are reusable, versioned references to data.
• The model registry stores and versions trained models for governance and reuse.
• Endpoints deploy models so applications can call them for real-time or batch predictions.

Computer vision on Azure

• Azure AI Vision (formerly Computer Vision): image classification, object detection, captioning, tagging, and OCR for reading printed/handwritten text.
• Image classification = what the image is; object detection = what objects are present and where (bounding boxes).
• Azure AI Face: detect, analyze, and recognize human faces and facial attributes.
• Azure AI Document Intelligence (formerly Form Recognizer): extract text, key-value pairs, and tables from forms, invoices, and receipts.
• OCR reads text from images and documents; prebuilt and custom models are available for documents.

NLP & speech on Azure

• Azure AI Language (formerly Text Analytics/LUIS): key phrase extraction, named entity recognition, sentiment analysis, language detection, and question answering.
• Azure AI Translator: real-time text translation across many languages.
• Azure AI Speech: speech-to-text (transcription), text-to-speech (synthesis), speech translation, and speaker recognition.
• Sentiment analysis scores text as positive/negative/neutral; entity recognition identifies people, places, organizations, and dates.
• Conversational language understanding interprets user intent and entities to power bots and assistants.

Generative AI on Azure

• Azure OpenAI Service provides GPT models (text/chat), embeddings (semantic similarity), and DALL-E (image generation).
• Tokens are the chunks of text models process; a prompt is the input, the completion is the generated output.
• Embeddings turn text into numeric vectors used for search, similarity, and grounding (RAG).
• Azure AI Foundry (formerly Azure AI Studio) is the unified platform to build, deploy, and manage generative AI, with a model catalog of foundation models.
• Built-in responsible-AI tooling includes content filtering, groundedness checks, and safety evaluations to reduce harmful or fabricated output.

Glossary

Artificial intelligence (AI)

Software that performs tasks normally requiring human intelligence, such as recognizing images, understanding language, or making predictions.

Machine learning (ML)

A subset of AI in which models learn patterns from data rather than being explicitly programmed with rules.

Deep learning

Machine learning that uses multi-layered neural networks to model complex patterns, powering vision and language tasks.

Transformer

A neural network architecture using attention mechanisms; the foundation of modern large language models.

Regression

A supervised ML task that predicts a continuous numeric value, such as a price or temperature.

Classification

A supervised ML task that predicts which category or class an item belongs to.

Clustering

An unsupervised ML task that groups similar, unlabeled data points together.

Features and labels

Features are the input variables a model learns from; the label is the value the model is trained to predict.

Training and validation data

Training data is used to build the model; validation (or test) data measures how accurately it predicts unseen cases.

Azure Machine Learning

Azure's cloud platform for building, training, deploying, and managing ML models at scale.

Automated ML (AutoML)

An Azure ML feature that automatically tries multiple algorithms and settings to find the best-performing model.

Azure ML designer

A no-code, drag-and-drop interface in Azure Machine Learning for building ML pipelines visually.

Model registry

A versioned store in Azure ML for managing, tracking, and reusing trained models.

Endpoint

A deployed model exposed as a callable service so applications can request predictions in real time or batch.

Azure AI services

The umbrella brand (formerly Cognitive Services) for prebuilt vision, language, speech, and decision APIs.

Azure AI Vision

Service (formerly Computer Vision) for image classification, object detection, captioning, and OCR.

Object detection

A vision task that identifies objects in an image and locates each with a bounding box.

Optical character recognition (OCR)

Technology that reads printed or handwritten text from images and documents into machine-readable text.

Azure AI Face

Service that detects, analyzes, and recognizes human faces and facial attributes in images.

Azure AI Document Intelligence

Service (formerly Form Recognizer) that extracts text, key-value pairs, and tables from forms, invoices, and receipts.

Natural language processing (NLP)

AI that interprets, analyzes, and generates human language in text or speech.

Azure AI Language

Service (formerly Text Analytics/LUIS) for sentiment analysis, entity recognition, key phrase extraction, and language understanding.

Sentiment analysis

An NLP task that classifies text as positive, negative, or neutral in tone.

Named entity recognition

An NLP task that identifies entities such as people, places, organizations, and dates within text.

Azure AI Translator

Service that performs real-time text translation across many languages.

Azure AI Speech

Service for speech-to-text, text-to-speech, speech translation, and speaker recognition.

Generative AI

AI that creates new content - text, images, code, or audio - in response to a prompt.

Large language model (LLM)

A generative model trained on vast text data to understand and produce human-like language.

Token

A chunk of text (word or sub-word) that a language model processes; usage and limits are measured in tokens.

Prompt and completion

The prompt is the input you give a generative model; the completion is the output it generates in response.

Embedding

A numeric vector representation of text that captures meaning, used for search, similarity, and grounding.

Azure OpenAI Service

Azure service providing access to OpenAI models including GPT (text/chat), embeddings, and DALL-E (image generation).

Azure AI Foundry

Unified platform (formerly Azure AI Studio) for building, deploying, and managing AI and generative AI solutions, including a model catalog.

Responsible AI

Microsoft's framework of six principles - fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability - for building trustworthy AI.