Career upgrade: Learn practical AI skills for better jobs and higher pay.
Level up
Cheat sheet

Azure AI-900 Cheat Sheet

AI Workloads

15-20%of exam

Responsible AIAI WorkloadsDocument ProcessingAnomaly DetectionUse Cases

Machine Learning

15-20%of exam

RegressionClassificationClusteringTraining DataAzure ML

Computer Vision

15-20%of exam

Image AnalysisOCRFaceCustom VisionDocument Intelligence

NLP Workloads

15-20%of exam

LanguageSpeechTranslatorSentimentKey Phrases

Generative AI

20-25%of exam

Azure OpenAIPromptsRAGCopilotResponsible GenAI

Quick Facts

Exam
AI-900
Credential
Azure AI Fundamentals
Time
45 min
Pass
700/1000
Level
Fundamentals
Retires
Jun 30 2026
Prereq
None

Responsible Six

Fair safe private inclusive transparent accountable

FairnessReliabilityPrivacyInclusivenessTransparencyAccountability

Workload Picker

  1. Images or videoVision
  2. Forms or invoicesDocument Intelligence
  3. Text sentimentLanguage
  4. Speech audioSpeech
  5. Translate textTranslator
  6. Generate contentAzure OpenAI

Responsible AI

Fairness
Avoid bias
Reliability
Work safely
Privacy
Protect data
Inclusiveness
Serve everyone
Transparency
Explain behavior
Accountability
Human responsibility

AI Workloads

Vision
Images/video
NLP
Human language
Speech
Audio language
Document processing
Forms/text extraction
Anomaly detection
Unusual patterns
Generative AI
Creates content
Recommendation
Suggest items

ML Targets

Number, label, group

RegressionClassificationClustering

Regression vs Classification

Regression

  • Numeric target
  • Price/quantity
  • Continuous output

Classification

  • Category target
  • Yes/no
  • Class label

Number vs label

ML Picker

  1. Predict priceRegression
  2. Predict categoryClassification
  3. Group customersClustering
  4. Image recognitionDeep learning
  5. Train without codeAutoML
  6. Serve predictionsEndpoint

ML Concepts

Feature
Input variable
Label
Target value
Training set
Learn patterns
Validation set
Tune model
Test set
Final check
Model
Learned function
Inference
Use model

Classification vs Clustering

Classification

  • Known labels
  • Supervised
  • Predict class

Clustering

  • No labels
  • Unsupervised
  • Find groups

Labels vs groups

ML Techniques

Regression
Predict number
Classification
Predict class
Clustering
Find groups
Deep learning
Neural networks
Transformer
Attention architecture
AutoML
Automate training
Metric
Measure performance

Azure ML

Workspace
ML container
Designer
Visual pipelines
Automated ML
Model search
Compute
Training resources
Dataset
Data asset
Endpoint
Model serving

Vision vs Document

Vision

  • Images
  • Objects
  • Captions

Document

  • Forms
  • Tables
  • Fields

Scene vs form

Vision Picker

  1. Read image textOCR
  2. Describe imageImage Analysis
  3. Detect objectsObject detection
  4. Classify custom imagesCustom Vision
  5. Extract form fieldsDocument Intelligence
  6. Detect facesFace

Vision Services

Image Analysis
Tags/captions
OCR
Read text
Face
Detect faces
Custom Vision
Custom classifier
Object detection
Locate objects
Spatial analysis
People movement
Content safety
Moderate content

Document Intelligence

Prebuilt model
Common documents
Custom model
Your forms
Layout
Text/tables
Invoice
Billing fields
Receipt
Purchase fields
Confidence
Extraction score

Language Stack

Text, speech, translate

LanguageSpeechTranslator

Speech vs Translator

Speech

  • Audio input
  • Transcription
  • Voice output

Translator

  • Text input
  • Language conversion
  • No audio

Audio vs text

Language Services

Sentiment
Opinion score
Key phrases
Important terms
Entity recognition
Named things
PII detection
Sensitive text
Summarization
Shorten text
Question answering
Knowledge answers
Conversational language
Intent/entities

Speech + Translate

Speech-to-text
Audio to words
Text-to-speech
Words to audio
Speech translation
Audio translated
Translator
Text translation
Custom voice
Brand voice
Pronunciation
Speaking assessment

RAG Flow

Retrieve facts before generation

SearchGroundGenerateCite

Prompt vs RAG

Prompt

  • Instruction
  • Context words
  • Model input

RAG

  • Retrieve facts
  • Ground output
  • External data

Ask vs ground

GenAI Picker

  1. Need factual answersRAG
  2. Need search memoryEmbeddings
  3. Need task assistantCopilot
  4. Need custom appAzure OpenAI
  5. Need better outputPrompt engineering
  6. Need safety controlsContent filters

Generative AI

LLM
Text generation
Prompt
Input instruction
Completion
Generated output
Grounding
Use facts
RAG
Retrieve then generate
Embeddings
Vector meaning
Copilot
AI assistant
Azure OpenAI
Hosted models

OpenAI vs Copilot

Azure OpenAI

  • Build apps
  • Model APIs
  • Custom prompts

Copilot

  • Ready assistant
  • Product embedded
  • Task help

Platform vs assistant

Common Traps

Regression vs classification

Regression predicts numbers Classification predicts labels

Vision vs OCR

Vision sees images OCR reads text

Translator vs Speech

Translator handles text Speech handles audio

Prompt vs grounding

Prompt instructs model Grounding supplies facts

Responsible AI

Fairness reduces bias Accountability stays human

AI-900 timing

Exam retires June Finish before retirement

Last Minute

  1. 1.AI-900 retires Jun 30 2026
  2. 2.GenAI is 20-25%
  3. 3.Regression = number
  4. 4.Classification = label
  5. 5.Clustering = unlabeled groups
  6. 6.OCR = read image text
  7. 7.Language = text analytics
  8. 8.Speech = audio
  9. 9.RAG = retrieve then generate
  10. 10.Responsible AI has six principles
Same family resources

Explore More Microsoft Azure Certifications

Continue into nearby exams from the same family. Each card keeps practice questions, study guides, flashcards, videos, and articles in one place.