ExamFinch.com My Study

Cheat sheet

Azure AI-901 Cheat Sheet

Practice this exam

Responsible AI

  • 40-45% of exam
  • Responsible AI Safety Controls Content Safety Human Review Fairness Transparency

Models + Workloads

  • 40-45% of exam
  • Model Basics AI Workloads ML Tasks RAG Model Picker Parameters

Foundry Apps

  • 55-60% of exam
  • Foundry Flow Prompting Agents Deployments Build Picker Playground

Foundry Tools

  • 55-60% of exam
  • Text + Speech Vision + Images Extraction Content Understanding Speech Picker Vision Picker

Quick Facts

  • Exam AI-901
  • Credential Azure AI Fundamentals
  • Status Beta
  • Time 45 minutes
  • Pass 700/1000
  • Questions 40-60 MCQ
  • Fee $99 US
  • Blueprint Apr 15 2026
  • Largest Foundry 55-60%
  • Code Python basics

Responsible AI RE

  • F R P I T A
  • Fairness Reliability+safety Privacy+security Inclusiveness Transparency Accountability

Responsible AI RE

  • Fairness Equitable outcomes
  • Reliability Consistent and safe
  • Privacy Protect user data
  • Security Protect connected systems
  • Inclusiveness Accessible for everyone
  • Transparency Explain limits
  • Accountability People remain responsible
  • Review Humans approve risk

Safety Controls RE

  • Content filters Block harms
  • Hate Identity attacks
  • Sexual Adult content
  • Violence Physical harm
  • Self-harm Self injury
  • Prompt Shields Prompt attack detection
  • Protected material Known text/code
  • Groundedness Checks source support

Token Limits MO

  • Tokens bound cost, context, length
  • Input tokens Output tokens Context cap Max response

RAG vs Fine-tune MO

RAG

  • Retrieves sources
  • Fresh/private data
  • No weight change

Fine-tune

  • Learns examples
  • Changes weights
  • Stable task style
  • Ground vs adapt

Model Picker MO

  • Need chat text → Chat model ( Conversation )
  • Need semantic search → Embeddings ( Vectors )
  • Need image input → Multimodal ( Vision prompt )
  • Need image output → Image generation ( New image )
  • Need private facts → RAG ( Grounding )
  • Need repeated style → Fine-tune ( Examples )
  • Need lower cost → Smaller model ( Latency )
  • Need residency → Region choice ( Compliance )

Exam Snapshot MO

  • AI-901 Azure AI Fundamentals
  • Status Beta exam
  • Updated Apr 15 2026
  • Pass 700/1000
  • Time 45 minutes
  • Format Multiple choice
  • Fee $99 US
  • Provider Pearson VUE
  • Practice test Not yet available

Classification vs Regression MO

Classification

  • Predicts category
  • Spam/not spam
  • Discrete output

Regression

  • Predicts number
  • Demand forecast
  • Continuous output
  • Label vs number

Model Basics MO

  • Token Text unit
  • Context window Input/output limit
  • Embedding Meaning vector
  • Temperature Randomness control
  • Max tokens Output cap
  • System message High-priority rules
  • User message Task request
  • Deployment Callable model
  • Endpoint Request URL

AI Workloads MO

  • Generative AI Creates content
  • Predictive AI Estimates outcomes
  • Agentic AI Acts with tools
  • Text analysis Understands text
  • Speech Processes audio
  • Vision Processes images
  • Extraction Structures content
  • Translation Changes language

ML Tasks MO

  • Classification Predicts category
  • Regression Predicts number
  • Clustering Groups unlabeled data
  • Forecasting Predicts future values
  • Anomaly detection Finds outliers
  • Training Learns from data
  • Inference Uses trained model

Foundry Flow FO

  • Catalog | Deploy | Test | Code
  • Catalog Deployment Playground SDK/app

Catalog vs Deployment FO

Catalog

  • Compare options
  • Model cards
  • Capabilities/cost

Deployment

  • Creates endpoint
  • Named model
  • App callable
  • Choose vs use

Build Picker FO

  • Compare models → Catalog
  • Call model → Deployment
  • Try prompt → Playground
  • Create app → SDK
  • Connect data → Connection
  • Use actions → Agent
  • Inspect behavior → Tracing
  • Share stable agent → Publish

Foundry Flow FO

  • Foundry resource Azure AI container
  • Project Solution workspace
  • Catalog Compare models
  • Deployment Creates endpoint
  • Playground Test prompts
  • View code Copy client call
  • SDK Build app
  • Connection Secures data link
  • RBAC Controls access

Playground vs SDK FO

Playground

  • Portal test
  • Tune prompts
  • No code

SDK

  • Application code
  • Project client
  • Production path
  • Prototype vs build

Prompting FO

  • System prompt Sets behavior
  • User prompt Asks task
  • Few-shot Shows examples
  • Zero-shot Instruction only
  • Constraints Format boundaries
  • Grounding Adds sources
  • Temperature Varies output
  • Max tokens Limits length

Chat vs Agent FO

Chat

  • Single response
  • No tool plan
  • Prompt driven

Agent

  • Multi-step goal
  • Uses tools
  • Managed runtime
  • Answer vs act

Agents FO

  • Agent Model plus tools
  • Instructions Goal and rules
  • Tools External actions
  • Runtime Runs agent
  • Responses API Agent entry point
  • Memory Conversation state
  • File search Finds project files
  • Publish Stable endpoint
  • Tracing Observe steps

System vs User Prompt FO

System

  • Role rules
  • Higher priority
  • Behavior guard

User

  • Task request
  • Conversation input
  • Lower priority
  • Rules vs request

Workload Inputs TO

  • Text, speech, image, document
  • Language: text Speech: audio Vision: image CU: multimodal

OCR vs Extraction TO

OCR

  • Reads text
  • Lines/words
  • Image/document input

Extraction

  • Returns fields
  • Schema output
  • Confidence scores
  • Text vs fields

Workload Picker TO

  • Analyze text → Language
  • Find text entities → NER
  • Transcribe audio → Speech to text
  • Speak response → Text to speech
  • Translate speech → Speech translation
  • Read image text → OCR
  • Generate image → Image model
  • Extract form fields → Content Understanding
  • Process video → Content Understanding

Text + Speech TO

  • Language Text analytics
  • Sentiment Opinion polarity
  • Key phrases Main terms
  • NER Typed entities
  • Summarization Shorter meaning
  • Speech to text Audio transcript
  • Text to speech Spoken output
  • Speech translation Translated audio/text
  • Diarization Separates speakers

Speech vs Speaker TO

Speech recognition

  • What was said
  • Transcript
  • Captions

Speaker recognition

  • Who spoke
  • Voice traits
  • Azure retired Sep 30, 2025
  • Words vs person

Vision + Images TO

  • Vision Image analysis
  • OCR Reads visible text
  • Object detection Labels locations
  • Image captions Describes scene
  • Multimodal Text plus image
  • Visual input Prompt image reasoning
  • Image generation Creates new image
  • Image safety Flags harmful images

Vision vs Generation TO

Vision

  • Analyzes image
  • OCR/objects
  • Existing content

Generation

  • Creates image
  • Prompt driven
  • New content
  • Inspect vs create

Content Understanding TO

  • Documents Forms and files
  • Images Visual fields
  • Audio Speech insights
  • Video Scenes and moments
  • Analyzer Extraction configuration
  • Schema Desired fields
  • Confidence Reliability score
  • Grounding Source region
  • Markdown Search output
  • JSON Automation output

Common Traps

Facts vs fluency

  • Model sounds confident ≠ Sources prove answers

Catalog vs deployment

  • Catalog lists models ≠ Deployment makes callable

Prompt vs fine-tune

  • Prompt changes context ≠ Fine-tune changes weights

Agent vs chatbot

  • Chat replies once ≠ Agent uses tools

OCR vs extraction

  • OCR reads text ≠ Extraction returns fields

Speech vs speaker

  • Speech recognizes words ≠ Speaker recognizes person

Vision vs generation

  • Vision analyzes images ≠ Generation creates images

Fairness vs accuracy

  • Fairness checks groups ≠ Accuracy averages performance

Last Minute

  • 1. Blueprint: 40-45 / 55-60
  • 2. Pass: 700/1000
  • 3. Beta updated Apr 15 2026
  • 4. Foundry domain is larger
  • 5. FRPITA = responsible AI
  • 6. Catalog then deploy
  • 7. Playground before SDK
  • 8. RAG grounds; fine-tune adapts
  • 9. System prompt sets rules
  • 10. Tokens drive cost and limits
  • 11. Agent = model + tools
  • 12. Language = written text
  • 13. Speech = audio
  • 14. Vision = images
  • 15. Content Understanding extracts fields
  • 16. OCR reads visible text

Microsoft Certified: Azure AI Fundamentals (AI-901)

  • Practice Questions 100 questions Study Guide 15 topics Cheat Sheet 102 facts Flashcards 50 cards Shorts 1 short Videos 2 videos Blogs 3 blogs Free CE Podcast