Skip to content

AI-900 study guide ​

Key facts grouped by workload. AI-900 mostly tests whether you can match a scenario to the right kind of AI workload and Azure service.

Key facts by topic ​

Azure AI Services - Which Does What? ​

ServiceCapabilities
VisionImage analysis, OCR, object detection, face detection (attributes)
FaceFace recognition (verify, identify), face detection
LanguageCLU, sentiment, NER, key phrases, PII detection
SpeechSpeech-to-text, text-to-speech
TranslatorText translation, document translation
Content SafetyContent moderation
Document IntelligenceExtract data from forms/documents

Vision Service Limitations ​

CAN DoCANNOT Do
Detect adult contentModerate content
Extract text (OCR)Translate text
Detect facesRecognize faces (use Face service)
Analyze imagesTrain custom models (use Custom Vision)

Face Detection vs Face Recognition ​

Face Detection/AnalysisFace Recognition
Age estimationVerification (1:1)
Glasses detectionIdentification (1:many)
3D orientation-
Gender-
NOT emotion-

Microsoft Responsible AI Principles ​

PrincipleCovers
FairnessTreat all people equally
Reliability & SafetyWorks correctly, safe operation
Privacy & SecurityData protection, secure systems
InclusivenessEmpower everyone, accessible
TransparencyUnderstandable, explainable
AccountabilityPEOPLE responsible for AI

Responsible AI Lifecycle Stages ​

  1. Identify - Detect and prioritize potential harms
  2. Measure - Assess frequency and severity
  3. Mitigate - Implement safeguards (content filtering, prompt engineering)
  4. Operate - Readiness plan, monitoring, incident response

Azure OpenAI Fine-Tuning Facts ​

  • REST API: Supported
  • Data format: JSONL (not CSV)
  • Billing: Hourly hosting charge (not per-token)
  • Storage only (not deployed): Free

Azure AI Foundry ​

ConceptPurpose
AccountInfrastructure, policies, security
ProjectDevelopment workspace, deploy agents
EvaluationsAssess model quality, safety, metrics
PlaygroundInteractive experimentation
Prompt flowVisual workflow design tool for orchestrating multi-step applications
AgentsIntelligent entities built to perform tasks

Deployment Options ​

TypeUse Case
StandardVariable workloads, pay-per-use
BatchAsync processing, large data, non-urgent
ProvisionedMission-critical, low-latency, high volume

ML Fundamentals ​

ConceptDefinition
FeatureInput variable (what you know)
LabelOutput variable (what you predict)
Training setFit the model
Validation setTune hyperparameters
Testing setFinal evaluation

AutoML Support ​

SupportedNOT Supported
ClassificationClustering
Regression-
Time-series forecasting-

Sentiment Analysis Details ​

  • Scores at BOTH document and sentence levels
  • Confidence scores: 0 to 1 (not -1 to 1)
  • All three scores (positive, neutral, negative) add up to 1

Transformer Architecture ​

  • Faster than RNNs due to parallel processing
  • Uses self-attention (not convolutions)
  • Positional encodings preserve token order
  • No sequential dependencies

NLP Workload Types ​

WorkloadExample
Language modelingCommand to smart action
Speech recognitionPhone call transcript
Sentiment analysisMining customer opinions
Entity recognitionFinding people/places
Key phrase extractionMain talking points
TranslationLanguage conversion

Personal study notes written from official documentation. Not affiliated with Microsoft or GitHub. Contains no exam questions.