Exam Prep: Putting It All Together | Guided by A Guide to Cloud

You’ve made it

Simple explanation

You’ve completed 25 prior modules covering every AI-901 exam objective. This module ties it all together.

Think of this as your final revision session. We’ll review the most important concepts, highlight common exam traps, and give you a strategy for exam day. The goal isn’t to re-teach everything — it’s to make sure you can recognise the right answer quickly when the exam asks.

Domain 1 quick reference

Responsible AI — the six principles

Keyword in Scenario	→ Principle
Bias, discrimination, demographic groups	Fairness
Accuracy, edge cases, error handling	Reliability & Safety
Data handling, encryption, compliance	Privacy & Security
Explainability, citations, “am I talking to AI?”	Transparency
Accessibility, multiple languages, diverse users	Inclusiveness
Governance, oversight, impact assessment	Accountability

AI model selection

Scenario	→ Model Choice
Complex reasoning, multimodal	GPT-4o
Simple classification, cost-sensitive	Phi-4-mini (SLM)
Image generation	GPT-image-1.5
Document search / RAG	Embedding model
Speech tasks	Azure AI Speech

Configuration parameters

Parameter	Low Value →	High Value →
Temperature	Deterministic, consistent	Creative, varied
Top-p	Focused word choices	Diverse word choices
Max tokens	Short responses	Long responses

AI workload types

Workload	Key Differentiator
Generative	Creates NEW content
Agentic	Takes ACTIONS (not just creates)
Text analysis	UNDERSTANDS existing text
Speech	Converts BETWEEN spoken and written
Vision	ANALYSES existing images
Extraction	Pulls STRUCTURED DATA from unstructured sources

Domain 2 quick reference

Foundry ecosystem

Component	What It Is
Foundry resource	Top-level Azure resource for AI
Project	Workspace within Foundry (models, agents, keys)
Playground	Interactive testing — no code needed
Foundry Tools	Dedicated AI services (Language, Speech, Vision, Content Understanding)
Model catalog	Browse and deploy models from multiple providers

API patterns

Pattern	When to Use
Chat Completions	Simple Q&A, summarisation, text tasks
Responses API (Agents)	Multi-step tasks, tool use, autonomous actions
Azure AI Language	High-volume text analysis with structured output
Azure AI Speech	Dedicated speech-to-text and text-to-speech
Azure AI Vision	High-volume image analysis
Content Understanding	Document, image, audio, video extraction

Common exam traps

Top 10 exam traps

Generative vs Agentic: If the AI is taking actions (booking, emailing), it’s agentic — not just generative
OCR vs Content Understanding: OCR reads text; Content Understanding extracts structured fields
Temperature 0 ≠ accurate: It means deterministic (same input → same output), not necessarily correct
System prompt vs user prompt: System = developer sets rules; User = end user asks questions
Models are stateless: Full conversation history must be sent with each API call (chat completions)
Foundry ≠ Azure OpenAI only: Foundry includes ALL Azure AI services, not just OpenAI models
SLM vs LLM: If the question mentions “cost-effective” or “edge” → think small model
Pre-built vs custom models: Use pre-built for standard documents; custom for your specific formats
Content filters are ON by default: You configure them, not enable them
Human-in-the-loop: Low confidence = human review, not automatic rejection

Exam day strategy

Strategy	Why
Read the full question	Key words often appear at the end
Eliminate wrong answers first	Usually 2 options are clearly wrong
Watch for “BEST” vs “possible”	Multiple answers may work; pick the BEST one
Scenario keywords → service	Map keywords to the right service/workload
Flag and return	Don’t get stuck — flag difficult questions and come back
Time management	45 min ÷ 50 questions ≈ 54 seconds each — keep moving

🎬 Video walkthrough

🎬 Video coming soon

Exam Prep Review — AI-901 Module 26

~15 min