A production Azure AI application stores API keys in environment variables. A security audit flags this as a concern. What is the recommended improvement?

Switch to managed identity with DefaultAzureCredential. Managed identity with DefaultAzureCredential is the most secure authentication method. It eliminates all credential management — no keys to store, rotate, or protect. Azure automatically handles token issuance and rotation. This is the recommended approach for all production deployments.

What is the content safety severity scale used by Azure AI Content Safety?

0-7. Azure AI Content Safety uses a 0-7 severity scale. 0-1 is Safe, 2-3 is Low, 4-5 is Medium, and 6-7 is High. This is frequently tested — do not confuse it with 0-3 or 0-10 scales.

Which of the following statements about Azure OpenAI is TRUE?

It requires a dedicated resource with --kind OpenAI. Azure OpenAI requires its own dedicated resource created with --kind OpenAI. It CANNOT be accessed through a multi-service resource. Content filters cannot be completely disabled (minimum safety filters are always active). DALL-E 3 only supports n=1 (one image per request).

You need to implement a conversational bot that handles both FAQ questions and flight booking actions. Which architecture should you use?

CLU orchestration routing to Custom Question Answering (FAQ) and CLU domain model (actions). The correct architecture is a CLU orchestration model that routes user utterances to the appropriate backend: Custom Question Answering for FAQ questions and a CLU domain model for structured actions (like flight booking with intent/entity extraction). This combines the strengths of both services.

A company needs their Azure AI application to work when the internet is temporarily unavailable. Which deployment approach should they use?

Deploy Azure AI containers on-premises or at the edge. Azure AI containers can be deployed on-premises or at the edge (Azure IoT Edge) for scenarios requiring offline capability. The AI models run locally, providing inference without internet. However, periodic connectivity is still needed for billing/metering purposes.

Key Takeaways

Remember the six domains and their weights: Plan/Manage (15-20%), Content Moderation (10-15%), Vision (15-20%), NLP (25-30%), Knowledge Mining (10-15%), Generative AI (10-15%).
Deprecated service names: LUIS → CLU, QnA Maker → Custom Question Answering, Form Recognizer → Document Intelligence, Cognitive Search → AI Search.
Security priority: managed identity > Entra ID tokens > Key Vault stored keys > environment variable keys > hardcoded keys (never).
Content Safety severity scale: 0-7 (not 0-3 or 0-10). Four categories: violence, self-harm, sexual, hate.
For production scenarios: always use Standard tier (SLA), private endpoints (network security), managed identity (authentication), and content filtering (safety).

Exam Day Quick Reference

Quick Answer: This section is a last-minute review of the most critical facts for the AI-102 exam. Focus on service selection, deprecated names, security hierarchy, content safety settings, and production best practices.

Critical Service Selection Rules

Text extraction from images → Azure AI Vision (Read API / OCR)
Field extraction from forms/invoices → Azure AI Document Intelligence
Sentiment, NER, key phrases → Azure AI Language
Custom intent/entity recognition → CLU (Conversational Language Understanding)
FAQ Q&A → Custom Question Answering
Text generation, summarization, chat → Azure OpenAI Service
Image generation → Azure OpenAI (DALL-E 3)
Search with AI enrichment → Azure AI Search
Text/image moderation → Azure AI Content Safety
Face detection/verification → Azure AI Face
Video analysis → Azure AI Video Indexer
People counting/tracking → Azure AI Vision (Spatial Analysis)
Speech to text → Azure AI Speech
Translation → Azure AI Translator

Security Hierarchy (Most to Least Secure)

Managed Identity — No credentials in code, automatic token management
Entra ID Service Principal — Token-based, rotatable
Keys in Key Vault — Centralized, audited, rotatable
Keys in Environment Variables — Better than hardcoded, but not ideal
Keys in Configuration Files — Risk of source code exposure
Hardcoded Keys — NEVER acceptable

Critical Numbers to Remember

Item	Value
Passing score	700/1000
Exam duration	120 minutes
Content Safety severity scale	0-7
Content Safety categories	4 (violence, self-harm, sexual, hate)
OCR languages (print)	164
OCR languages (handwriting)	9
CLU max utterances	50,000 per project
GPT-4o context window	128K tokens
1 token ≈	0.75 English words
embedding-3-large dimensions	3,072
DALL-E 3 images per request	1
Composed model max components	100
Fine-tuning min examples	10

Common Exam Traps

Trap	Correct Answer
Using LUIS for intent recognition	Use CLU (LUIS is deprecated)
Using QnA Maker for FAQ bots	Use Custom Question Answering
API keys for production auth	Use managed identity
Public endpoints for production	Use private endpoints
Disabling all content filters	Never disable; create custom configurations
Content Safety scale is 0-10	Scale is 0-7
Azure OpenAI in multi-service resource	Requires dedicated OpenAI resource
Face API emotion recognition	Retired — no longer available
DALL-E 3 n=4 (multiple images)	Only n=1 is supported
Translator uses standard AI endpoint	Uses api.cognitive.microsofttranslator.com
Temperature + top_p both adjusted	Adjust one OR the other, not both
Fine-tuning for real-time knowledge	Use RAG for up-to-date information

Azure AI Engineer Associate

7.5 Exam Day Quick Reference

Key Takeaways

Exam Day Quick Reference

Critical Service Selection Rules

Security Hierarchy (Most to Least Secure)

Critical Numbers to Remember

Common Exam Traps

Azure AI Engineer Associate

1Introduction

2Domain 1: Plan and Manage an Azure AI Solution (15-20%)

3Domain 2: Implement Content Moderation Solutions (10-15%)

4Domain 3: Implement Computer Vision Solutions (15-20%)

5Domain 4: Implement Natural Language Processing Solutions (25-30%)

6Domain 5: Implement Knowledge Mining and Document Intelligence Solutions (10-15%)

7Domain 6: Implement Generative AI Solutions (10-15%)

8Exam Review: Cross-Domain Topics and Advanced Practice

7.5 Exam Day Quick Reference

Key Takeaways

Exam Day Quick Reference

Critical Service Selection Rules

Security Hierarchy (Most to Least Secure)

Critical Numbers to Remember

Common Exam Traps