Skip to content

AI-102 study guide ​

Key facts by service. AI-102 goes a level deeper than AI-900: SDK and REST specifics, resource configuration, security and the order of operations in a pipeline.

Key facts by topic ​

ConceptKey Facts
Index Propertiesfilterable=WHERE, sortable=ORDER BY, facetable=counts, retrievable=include in results
Indexer Setup OrderData Source → Index → Indexer → Schedule
Indexer PropertiesskillsetName + outputFieldMappings for enrichments
REST OperationsRun=execute now, Reset=full re-index next run, Update=change definition
SecuritySeparate data sources for folders, Entra ID security filters for trimming
Query API KeysUse management.azure.com endpoint + Microsoft.Search provider + createQueryKey
Hybrid SearchBM25 + vector scoring for best keyword + semantic relevance
Partitions vs ReplicasPartitions=query speed, Replicas=concurrent load (2 replicas=99.9% SLA read-only)

Knowledge Store ​

Projection TypeStorageUse Case
TablesAzure Table StorageTransactional/tabular data, structured rows
ObjectsAzure Blob StorageJSON documents from enrichments
FilesAzure Blob StorageBinary files, images (JPEG, PNG)
PropertyPurpose
sourcePath to enrichment (e.g., /document/tableprojection/keyPhrases/*)
tableNameAzure Table storage table name
storageContainerFor blob/object projections

Azure OpenAI / Generative AI ​

ConceptKey Facts
Chat Completion APIclient.chat.completions.create() - GPT-4, GPT-4o, o3 models
Completion APIclient.completions.create() - GPT-35-Turbo-Instruct only
Model Parameter= deployment name in Azure OpenAI (not base model name)
Fine-tuning FormatJSONL (JSON Lines) only - NOT XML, JSON, or CSV
Fine-tuning StepsConvert to JSONL → Upload to Blob Storage → POST request
Embedding Modeltext-embedding-ada-002 (NOT GPT-4)
Multimodal ImagesBase64 data URI or URL with SAS token in image_url field
Usage Objectprompt_tokens + completion_tokens + total_tokens

Azure AI Foundry & Agents ​

ConceptKey Facts
Agent ComponentsLanguage Model + Instructions + Tools
Blocklistsai.azure.com → Safety + security → Blocklists tab → Add blocklist
Agent ToolsBing Search=public web, Logic Apps=external workflows, Code Interpreter=Python
File UploadCorrect connection string + Storage Blob Data Owner role on agents-blobstore
Conversation AccessFull access to structured threads (user-agent and agent-agent)
Data StorageDoes NOT auto-manage all data - requires external connections

Custom Vision ​

ConceptKey Facts
Use CaseTrain your own model for custom objects (logos, animals, products)
Portalcustomvision.ai (NOT Azure portal)
Object DetectionReturns location (bounding boxes) + classification
ClassificationLabels only, no location info
Edge DeploymentUse Compact domains (General compact, etc.)
Training MethodTrainProject() - returns iteration object
Polling Statusiteration.Status == "Training" while in progress
TensorFlow Exportmodel.pb (ProtoBuf) + labels.txt
Metrics APIGetIterationPerformance (not GetImagePerformances)
IterationDoesn't guarantee improvement - need more/better training data

Azure AI Vision ​

ConceptKey Facts
Endpoint Format{resourceName}.cognitiveservices.azure.com
visualFeaturesTags=word list, Categories=86-class taxonomy, Adult=nudity/violence, Brands=logos
detailsCelebrities, Landmarks
OCR PollingGetReadResultAsync, check Running AND NotStarted status
vs Custom VisionAzure AI Vision = pre-trained; Custom Vision = train your own

Document Intelligence ​

ConceptKey Facts
Use CaseExtract key-value pairs, text, tables from forms/documents
Custom TemplateRule-based, fixed layouts, consistent format
Custom NeuralDeep learning, mixed layouts, format variations
Composed ModelsGroup existing models → one model ID → auto-classification
Export FormatJSON only (not XML)
Access MethodsDocument Intelligence Studio + SDKs (C#, Python, Java, JavaScript)
Min Training Docs5 documents for custom models
vs NERDocument Intelligence = structured extraction; NER = entity recognition in text

Azure AI Language ​

FeatureKey Facts
CLU UtteranceUser input text (the "text" field)
CLU IntentWhat user wants to do (action category)
CLU OffsetCharacter position where entity starts
CLU Training SplitManual split = control specific utterances; Auto = percentage-based only
CLU FailoverCreate resource FIRST → Export → Import → Train/Deploy
Language DetectioncountryHint defaults to "US", stateless service, confidence 0-1
Sentiment "Mixed"Requires at least one positive AND one negative sentence
QA Refresh URLUpdates from source, retains manual edits
QA Alternative QuestionsDifferent phrasings to improve search
QA Follow-up PromptsGuide users through conversation path
QA MetadataCategorize/tag Q&A pairs
KB ExportEdit knowledge base (not Deploy) → Export as TSV

Speech Services ​

ConceptKey Facts
recognize_once()Max 15 seconds, single operation
start_continuous_recognition()Unlimited duration, ongoing
SpeechRecognitionLanguageSource language (what you're listening for)
AddTargetLanguageTranslation output languages
PhraseListGrammar.addPhraseImprove recognition of specific terms
SSMLAdjust pitch, rate, pronunciation, volume
AudioDataStreamStreaming for lower latency
Tuning fileAudio Content Creation tool for fine-tuning
Batch TranscriptionSpeech to Text REST API or Azure Speech CLI (not SDKs)
Whisper Limit25MB file size maximum
Custom KeywordBasic (minutes, demos) vs Advanced (48 hours, production)
VisemesLip-sync animation (position of lips, jaw, tongue)

Content Safety ​

ConceptKey Facts
Content FiltersBlock harmful content (hate, sexual, violence)
Prompt ShieldsDetect adversarial prompts (jailbreaks, prompt injections)
Prompt Shields ScopeScans user prompts + documents (RAG sources), NOT model outputs
AnalyzeTextOptionsClass for creating Content Safety request (not AnalyzeTextOp)
Packageazure.ai.contentsafety (sync), azure.ai.contentsafety.aio (async)
ExportFilter config exportable to Python, Java, C#

Semantic Kernel ​

ConceptKey Facts
Concurrent PatternParallel independent tasks
Sequential PatternOrdered pipeline/workflow
Handoff PatternDynamic escalation/transfer of control
OpenAPI PluginAuto-generates functions from Swagger/OpenAPI spec

Containers & Infrastructure ​

ConceptKey Facts
Docker Port Mapping-p host:container (e.g., -p 8080:5000)
Container ParamsEula=accept, Billing=endpoint URL, ApiKey=API key
Key RotationSwitch to Key2 → Reset Key1 → Switch back to Key1

Observability ​

ConceptKey Facts
Tracing Implementation"opentelemetry" (not logging.DEBUG)
Console ExportConsoleSpanExporter() for STDOUT
Azure MonitorAzureMonitorTraceExporter (needs connection string)
OTLP ExportOTLPSpanExporter (needs endpoint)

More key facts ​

Azure AI Search ​

ConceptKey Facts
Indexing PipelineDocument Cracking → Field Mappings → Skillset Execution → Output Field Mappings → Push into Index
PartitionsSplit data across compute - improves individual query speed
ReplicasLoad balancing - handles more concurrent users
Indexer SetupCreate data source → Define index → Create indexer → Schedule indexer
OCR SkillsetOnly needed for images, not database content

Azure OpenAI ​

ConceptKey Facts
Use CaseCreative content generation (descriptions, stories, code, summaries)
Fine-tuning FormatJSONL (JSON Lines) - NOT XML, JSON, or CSV
Fine-tuning StepsConvert to JSONL → Upload to Blob Storage → POST to fine_tuning/jobs
REST APISynchronous responses - NO webhooks needed
API IntegrationGet endpoint/key → Build payload → POST request → Parse JSON response

DALL-E (Azure OpenAI) ​

FeatureDetails
Output FormatPNG only (not JPEG)
Image Sizes1024x1024, 1024x1792, 1792x1024
View CodeYes - prefilled Python code available in playground

Azure AI Document Intelligence ​

ConceptKey Facts
Use CaseExtract key-value pairs, text, tables from forms/documents
Pre-built ModelsInvoices, receipts, tax forms, IDs
Custom ModelsTrain for unique document types (min 5 training docs)
Composed ModelsGroup multiple models → single model ID → auto-classification
NOT forGeneral text analysis (use Azure AI Language instead)

Azure AI Language ​

ConceptKey Facts
Use CaseAnalyze and understand text
FeaturesNER, sentiment analysis, key phrases, summarization, language detection
NOT forForm extraction (use Document Intelligence)
NOT forCreative content generation (use Azure OpenAI)
CLU Flat ArchitectureMax ~50 intents, single feature
CLU OrchestrationMultiple child apps + orchestration for 50+ intents

Azure AI Vision ​

ConceptKey Facts
Endpoint Formathttps://{region}.api.cognitive.microsoft.com/vision/v2.0/analyze
Region NameUse exact name: westus, eastus (not westu, not resource name)
visualFeaturesDescription, Tags, Categories, Objects, Faces, etc.
detailsLandmarks, Celebrities
OCR MethodsGetReadResultAsync for read results, check Running AND NotStarted status

Azure AI Speech ​

MethodMax DurationUse Case
recognize_once()15 secondsShort utterances
recognize_once_async()15 secondsShort utterances (async)
start_continuous_recognition()UnlimitedLong audio (10s - minutes)
addPhrase()N/AImprove recognition of custom terms

Azure AI Foundry ​

ConceptKey Facts
Hub SecuritySettings inherited by projects
Project RegionMUST be same region as parent hub
Cross-region ConnectionsProjects CAN connect to Azure OpenAI in different regions

Azure AI Containers (Disconnected) ​

SourceSupported?
Azure AI Foundry model catalogNo

Knowledge Mining ​

PhaseDescription
1. IngestImport and aggregate structured/unstructured data
2. EnrichApply AI skills (OCR, NER, key phrases, translation)
3. ExploreMake searchable, visualize, discover patterns

Cost Management ​

Alert TypeTriggers When
Actual Cost AlertActual spending reaches % threshold
Forecasted AlertML predicts spending will exceed % threshold
Budget ResetMonthly budgets reset at end of each month

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