Course home

Search

Search across lessons, key terms, scenarios, practice questions, the glossary, acronyms, and hands-on labs.

Practice assessment

118 exam-style practice questions across the five AI-103 exam domains. Choose your topics, set the count, and start quizzing.

Questions
118
Topics
5
Multi-Answer
16

Choose Mode

Select Topics

Choose one or more topics, or leave all unselected to include every topic.

Quiz Options

Order
Number of Questions
max: 118

Terminology Reference

Plain-language definitions for every key term across all 5 exam domains. Click a domain to expand, or search for any term.

Official Acronym reference

Key acronyms for AI-103 — Microsoft Foundry, Azure AI services, agents, RAG, and search. Master every one for this course.

Total
27
Source
Official

Acronym reference

1 / 26
Acronym
ETL
Stands For
Extract, Transform, Load

Full Acronym Reference

Quick-Match Quiz — Acronyms

Python examples

Every Python script from the course lessons in one place, grouped by module. Pick a lesson from the menu to view its script.

Scripts
30
Source
Lessons

Lesson scripts 


            
          

        

      

    


    
    

      

        
        

          
Chapter Notes

          
Your personal notes for this chapter

        

      

      

        
My Notes

        
        

          
          
          
        

      

    


    
    

      
      
      

      

        
Developing AI Apps and Agents on Azure

        
Structured learning path aligned to Microsoft Learn for Exam AI-103 (skills measured April 2026). Complete all 30 lessons across 5 modules — mapped one-to-one to the five exam skill domains — plus labs, knowledge checks, and practice assessments.

      

      

        
5
Modules

        
40+
Key terms

        
0
Modules completed

        
—
Assessment score

      

      
Course modules

      

        

          
Ch 1Plan and Manage an Azure AI Solution

          
Choose Foundry services & models, deploy/manage/secure, and apply responsible AI.

          

            Microsoft FoundryModel selectionSecurity & monitoringResponsible AI
          

          
0%

          

            
            
          

        

        

          
Ch 2Implement Generative AI and Agentic Solutions

          
Generative apps, RAG, agents, function calling, multi-agent, and optimization.

          

            RAGFoundry SDKAgentsFunction calling
          

          
0%

          

            
            
          

        

        

          
Ch 3Implement Computer Vision Solutions

          
Image/video generation & editing, multimodal understanding, and visual safety.

          

            Image generationInpaintingCaptioningContent Understanding
          

          
0%

          

            
            
          

        

        

          
Ch 4Implement Text Analysis Solutions

          
Language analysis, sentiment, translation, and speech for agentic interactions.

          

            Entities & sentimentTranslatorSpeech (STT/TTS)Structured JSON
          

          
0%

          

            
            
          

        

        

          
Ch 5Implement Information Extraction Solutions

          
Retrieval/grounding pipelines and document extraction with Content Understanding.

          

            Hybrid/vector searchOCRContent UnderstandingRAG ingestion
          

          
0%

          

            
            
          

        

      

    


    
    

      

        
Exam AI-103

        
Skills measured

        
Exam AI-103 — Developing AI Apps and Agents on Azure (skills measured as of April 16, 2026). Passing score: 700. The five skill domains map one-to-one to the five course modules. Content aligns with the official Microsoft Learn path modules listed below.

      

      

        
Domain 1 — Plan and manage an Azure AI solution (25–30%)

        
    
          
  • Choose appropriate Foundry services and models for generative AI and agents
    • 
          
  • Set up AI solutions in Foundry (infrastructure, deployment, CI/CD)
    • 
          
  • Manage, monitor, and secure AI systems (quotas, drift, identity, networking)
    • 
          
  • Implement responsible AI across generative and agentic systems
    • 
          
  • Covered in Module 1
    • 
        

      

      

        
Domain 2 — Implement generative AI and agentic solutions (30–35%)

        
    
          
  • Build generative applications by using Foundry (RAG, SDKs, evaluation)
    • 
          
  • Build agents by using Foundry (tools, function calling, memory)
    • 
          
  • Optimize and operationalize generative AI systems (prompting, observability)
    • 
          
  • Covered in Module 2 (primary implementation domain)
    • 
        

      

      

        
Domain 3 — Implement computer vision solutions (10–15%)

        
    
          
  • Design and implement image- and video-generation solutions
    • 
          
  • Design and implement multimodal understanding workflows
    • 
          
  • Implement responsible AI for multimodal content
    • 
          
  • Covered in Module 3
    • 
        

      

      

        
Domain 4 — Implement text analysis solutions (10–15%)

        
    
          
  • Apply language model text analysis (entities, sentiment, JSON, translation)
    • 
          
  • Implement speech solutions (STT, TTS, custom speech, speech translation)
    • 
          
  • Covered in Module 4
    • 
        

      

      

        
Domain 5 — Implement information extraction solutions (10–15%)

        
    
          
  • Build retrieval and grounding pipelines (semantic/hybrid/vector search, OCR)
    • 
          
  • Extract content from documents with Content Understanding (Markdown, fields)
    • 
          
  • Covered in Module 5
    • 
        

      

      

        
Microsoft Learn path modules (by module)

        
Modules from the Azure AI Apps and Agents Developer Associate learning path, mapped to this course.

        
    
          
  • Module 1 — Plan and prepare to develop AI solutions on Azure; Select, deploy, and evaluate Microsoft Foundry models; Implement a responsible generative AI solution in Microsoft Foundry
    • 
          
  • Module 2 — Develop a generative AI chat app with Microsoft Foundry; Develop generative AI apps that use tools; Optimize generative AI model performance with Microsoft Foundry; Develop AI agents with Microsoft Foundry and Visual Studio Code; … (+8 more agent modules)
    • 
          
  • Module 3 — Develop a vision-enabled generative AI application; Generate images with AI; Generate videos with Microsoft Foundry; Analyze images with Content Understanding; Create a multimodal analysis solution with Azure Content Understanding; Create an Azure Content Understanding client application
    • 
          
  • Module 4 — Analyze text with Azure Language in Foundry Tools; Develop a text analysis agent with the Azure Language MCP server; Develop a speech-capable generative AI application; Create speech-enabled apps with Azure Speech in Microsoft Foundry Tools; Develop a speech agent with the Azure Speech MCP server; Develop an Azure Speech Voice Live Agent in Microsoft Foundry; Translate text and speech with Microsoft Foundry Tools
    • 
          
  • Module 5 — Extract data with Azure Document Intelligence; Create a knowledge mining solution with Azure AI Search
    • 
        

      

      

        
      

    

    
    

      

        
Module 1

        
Plan and Manage an Azure AI Solution

        
Plan and manage Azure AI solutions on Microsoft Foundry: select services and models, deploy and secure workloads, and apply responsible AI across generative and agentic systems.

      

      

        
Learning objectives

        
    
          
  • 1.1 — Choose appropriate Foundry services and models (LLM, SLM, multimodal) for generative AI and agents
    • 
          
  • 1.1 — Choose retrieval/indexing methods and agent memory, tool, and knowledge services
    • 
          
  • 1.2 — Design infrastructure, choose deployment options, configure deployments, and integrate CI/CD
    • 
          
  • 1.3 — Manage quotas, scaling, and cost; monitor performance, drift, and grounding; secure with managed identity and private networking
    • 
          
  • 1.4 — Configure safety filters and guardrails, RAI evaluators, auditing, and agent governance
    • 
          
  • Foundation module — planning, deployment, security, and responsible AI.
    • 
        

      

      

        

          
Lessons

          Lesson 1 of 3
        

        
Each lesson aligns to a Microsoft Learn module and includes official terminology, worked examples, and Python samples.

        

        
Loading lessons…

        

          0 of 3 lessons completed
          

            

          

        

        

          
          
        

      

      

        
Scenario exercises

        
Select a scenario to view the recommended approach.

        

          
A team wants one Azure platform that unifies models, agents, and tools with shared RBAC, networking, tracing, and policy. What should they use?
Answer
Microsoft Foundry — a unified PaaS that groups agents, models, and tools under one Azure resource provider namespace with unified RBAC, networking, monitoring, and policy.

          
You already have an Azure OpenAI resource. How do you adopt Foundry without losing your endpoint and keys?
Answer
Upgrade the Azure OpenAI resource to a Foundry resource — this preserves the existing endpoint, API keys, and state.

          
An on-device assistant must run cheaply in a resource-constrained environment. LLM or SLM?
Answer
A small language model (SLM) such as Phi-4 — designed for low-latency, on-device, or resource-constrained scenarios. Reserve large models (GPT-5/GPT-4.1) for complex multi-step reasoning.

          
You need the best balance of capability and cost for a production chat workload. Which model family fits?
Answer
GPT-4.1
  • positioned as the best balance of capability and cost for production. GPT-4.1 mini is the fastest/cheapest
  • GPT-5 is the most capable for the hardest reasoning.

          
A solution must authenticate to Foundry with no secrets stored in code or config. What do you use?
Answer
Managed identity (Microsoft Entra ID) with keyless credentials — assign Cognitive Services User on the Foundry resource so DefaultAzureCredential obtains a token without storing API keys.

          
You want to call a model by name without first creating a deployment. Which Foundry feature enables that?
Answer
Instant access (preview) — call any supported model by name without creating a deployment first. Otherwise you configure a model deployment with a chosen capacity/quota.

          
Azure AI Content Safety scores a message at severity 6 for the Violence category. What does the default policy do?
Answer
It blocks the request. Content Safety uses a 0/2/4/6 scale (Safe/Low/Medium/High) across four harm categories — Hate & fairness, Sexual, Self-harm, Violence — and the default threshold blocks at Medium and above.

          
A malicious instruction is hidden inside a document your RAG app ingests. Which guardrail detects it?
Answer
Prompt Shields — specifically indirect (cross-domain) prompt injection detection, which flags malicious instructions embedded in documents the model processes. It also covers direct jailbreak attacks.

        

      

      
      

        
Hands-on labs

        

          
Official Microsoft Learn hands-on exercises
Azure labs

          
Do these on the official Microsoft Learn lab site (opens in a new tab). They use a real Azure subscription and the Microsoft Foundry portal.

          
        

        

          
Match Foundry Building Blocks
Click & Match

          
Click a card to select it, then click the correct category zone.

          

        

        

          
Order the CI/CD Deployment Flow
Click & Order

          
Arrange the model/agent deployment pipeline steps in order.

          

        

        

          
Responsible AI Terms
Fill in the Blank

          
Type the correct Responsible AI term for each definition.

          

        

      

      

        
Key terms

        
Review key terms from each Learn module. Select a card to view the definition, implementation notes, and code sample.

        

          
          1 / 3
          
        

        

          

            

              
Learn module

              
Plan and prepare to develop AI solutions on Azure

              
Connect to a Foundry project

            

            

              
Definition, notes, and code

              

              
Use the Foundry SDK with DefaultAzureCredential and your project endpoint — no API keys in code.

              

              
              

            

          

        

      

      

        
Knowledge check

        
After each attempt you can review recommended lessons below your score and retake the assessment.

        

        

      

      

        
Notes

        
        
 

      

      

        
      

    


    
    

      

        
Module 2

        
Implement Generative AI and Agentic Solutions

        
Build generative and agentic solutions with Microsoft Foundry: RAG, the Foundry SDK, agents with tools and memory, multi-agent orchestration, and operational tuning.

      

      

        
Learning objectives

        
    
          
  • 2.1 — Deploy/consume models; implement RAG; design tool-augmented, multistep reasoning flows
    • 
          
  • 2.1 — Evaluate apps for fabrications, relevance, quality, and safety; connect apps to a Foundry project
    • 
          
  • 2.2 — Define agent roles, goals, and tool schemas; integrate retrieval, function-calling, and memory
    • 
          
  • 2.2 — Integrate agent tools; orchestrate multi-agent and autonomous flows with approval controls
    • 
          
  • 2.3 — Tune prompts and parameters; add reflection/self-critique; set up observability
    • 
          
  • Core implementation module for generative AI, RAG, and agents.
    • 
        

      

      

        

          
Lessons

          Lesson 1 of 12
        

        
Each lesson aligns to a Microsoft Learn module and includes official terminology, worked examples, and Python samples.

        

        
Loading lessons…

        

          0 of 12 lessons completed
          

            

          

        

        

          
          
        

      

      

        
Scenario exercises

        
Select a scenario to view the recommended approach.

        

          
An LLM keeps inventing facts about your internal policies. What pattern grounds it in your data?
Answer
Retrieval-Augmented Generation (RAG) — retrieve relevant chunks from a search index and pass them as grounding context so the model answers from your content, with citations.

          
Responses are too random and inconsistent for a compliance summarizer. Which parameter do you lower?
Answer
Temperature (and/or top_p) — lower values make output more deterministic and focused. Raise temperature only when you want more creative variety.

          
You must measure whether app answers are fabricated or off-topic before shipping. What do you use?
Answer
Foundry evaluators / evaluations — run quality and safety evaluations (e.g., groundedness, relevance, coherence, fluency) to detect fabrications and measure relevance, quality, and safety.

          
An agent needs real-time data from your systems. You describe a function's name, parameters, and purpose. What is this?
Answer
Function calling
  • you define tool schemas
  • the model requests a call, your app executes the function and returns the output, then the agent uses the result to complete its response.

          
Agent A asks Agent B a question and then continues talking to the user itself. Which connection pattern is this?
Answer
The A2A tool — Agent A calls Agent B, B's answer returns to A, and A stays in control of the conversation. (In a multi-agent workflow, B would take over the conversation instead.)

          
You're migrating from the classic "Connected Agents" tool, which isn't in the new Agent Service. What replaces it?
Answer
Use a multi-agent workflow or the A2A tool to call one Foundry agent from another — the classic Connected Agents tool is not available in the new Foundry Agent Service.

        

      

      
      

        
Hands-on labs

        
        

          
Order the RAG Pipeline
Click & Order

          
Arrange the retrieval-augmented generation steps in the correct order.

          

        

        

          
Python: Call a Foundry Model
Code Lab

          
Run each cell to build a grounded chat completion request in Python (simulated, no SDK needed).

          

        

        

          
Match Agent Tools & Concepts
Click & Match

          
Match each agent capability to the correct tool or concept.

          

        

      

      

        
Key terms

        
Review key terms from each Learn module. Select a card to view the definition, implementation notes, and code sample.

        

          
          1 / 12
          
        

        

          

            

              
Learn module

              
Develop a generative AI chat app with Microsoft Foundry

              
Responses API chat

            

            

              
Definition, notes, and code

              

              
The Responses API is the unified entry point for generative apps — prefer it over legacy Chat Completions for new work.

              

              
              

            

          

        

      

      

        
Knowledge check

        
After each attempt you can review recommended lessons below your score and retake the assessment.

        

        

      

      

        
Notes

        
        
 

      

      

        
      

    


    
    

      

        
Module 3

        
Implement Computer Vision Solutions

        
Implement computer vision solutions: generate and edit images and video, build multimodal understanding workflows, and apply responsible AI to visual content.

      

      

        
Learning objectives

        
    
          
  • 3.1 — Generate images and videos from text prompts and reference media; edit with inpainting and mask-based edits
    • 
          
  • 3.2 — Analyze visual context with multimodal models; produce captions and answer questions grounded in visual evidence
    • 
          
  • 3.2 — Generate alt-text/extended descriptions; use Content Understanding to extract visual characteristics and analyze video
    • 
          
  • 3.2 — Identify objects, components, or regions within images or video
    • 
          
  • 3.3 — Classify unsafe visual content, mitigate indirect prompt injection from text in images, enforce visual policy (watermarks, prohibited symbols)
    • 
        

      

      

        

          
Lessons

          Lesson 1 of 6
        

        
Each lesson aligns to a Microsoft Learn module and includes official terminology, worked examples, and Python samples.

        

        
Loading lessons…

        

          0 of 6 lessons completed
          

            

          

        

        

          
          
        

      

      

        
Scenario exercises

        
Select a scenario to view the recommended approach.

        

          
You must replace just one object in a generated image while keeping the rest intact. Which technique?
Answer
Inpainting with a mask
  • mask the region to change and provide a prompt
  • the model regenerates only the masked area (mask-based / prompt-driven editing).

          
An accessibility requirement needs concise alt-text plus a longer description for each product image. Which capability?
Answer
Multimodal captioning — generate concise captions and extended image descriptions aligned to accessibility guidelines (alt-text + detailed descriptions).

          
You want a one-paragraph description of a standalone image with no embedded text. Which Content Understanding analyzer?
Answer
The prebuilt-imageSearch analyzer — returns a one-paragraph description of image content. For images that contain text or handwriting, use prebuilt-documentSearch instead.

          
A processed image hides text that says "ignore your instructions." What is the risk and mitigation?
Answer
Indirect prompt injection via text embedded in an image — detect and mitigate it (e.g., Prompt Shields / image-text screening) so the model doesn't follow hidden instructions.

          
Policy requires flagging prohibited symbols and applying a watermark to generated images. What do you enforce?
Answer
Visual policy rules — apply watermarks/content credentials, flag prohibited symbols, uphold brand requirements, and detect potentially inappropriate content.

        

      

      
      

        
Hands-on labs

        

          
Official Microsoft Learn hands-on exercises
Azure labs

          
Do these on the official Microsoft Learn lab site (opens in a new tab). They use a real Azure subscription and the Microsoft Foundry portal.

          
        

        

          
Match Vision Tasks to Capabilities
Click & Match

          
Match each vision scenario to the capability that handles it.

          

        

        

          
Image Generation & Editing Terms
Fill in the Blank

          
Type the correct image generation/editing term for each clue.

          

        

      

      

        
Key terms

        
Review key terms from each Learn module. Select a card to view the definition, implementation notes, and code sample.

        

          
          1 / 6
          
        

        

          

            

              
Learn module

              
Develop a vision-enabled generative AI application

              
Vision chat with image input

            

            

              
Definition, notes, and code

              

              
Pass image URLs or base64 content alongside text to a multimodal deployment.

              

              
              

            

          

        

      

      

        
Knowledge check

        
After each attempt you can review recommended lessons below your score and retake the assessment.

        

        

      

      

        
Notes

        
        
 

      

      

        
      

    


    
    

      

        
Module 4

        
Implement Text Analysis Solutions

        
Implement text and speech analysis solutions: entity extraction, sentiment, translation, Azure AI Language, and speech-to-text and text-to-speech for agentic applications.

      

      

        
Learning objectives

        
    
          
  • 4.1 — Extract entities, topics, summaries, and structured JSON via prompting and Foundry Tools
    • 
          
  • 4.1 — Detect sentiment, tone, safety, and sensitive content; customize for domain tasks
    • 
          
  • 4.1 — Translate text with Azure Translator or LLM-powered flows
    • 
          
  • 4.2 — Convert speech to text and text to speech; add custom speech and audio reasoning
    • 
          
  • 4.2 — Translate speech into other languages as an agent modality
    • 
        

      

      

        

          
Lessons

          Lesson 1 of 7
        

        
Each lesson aligns to a Microsoft Learn module and includes official terminology, worked examples, and Python samples.

        

        
Loading lessons…

        

          0 of 7 lessons completed
          

            

          

        

        

          
          
        

      

      

        
Scenario exercises

        
Select a scenario to view the recommended approach.

        

          
You need to pull people, places, and organizations out of free text. Which capability?
Answer
Entity extraction (Named Entity Recognition)
  • via Azure AI Language or generative prompting with Foundry Tools
  • you can also request structured JSON output for downstream use.

          
An agent needs spoken interaction: hear the user and speak back. Which service and capabilities?
Answer
Azure AI Speech — speech-to-text and text-to-speech (with optional custom speech models), enabling speech as an agent modality and speech translation across languages.

        

      

      
      

        
Hands-on labs

        

          
Official Microsoft Learn hands-on exercises
Azure labs

          
Do these on the official Microsoft Learn lab site (opens in a new tab). They use a real Azure subscription and the Microsoft Foundry portal.

          
        

        

          
Match Search & Extraction Services
Click & Match

          
Match each task to the correct language, speech, or search capability.

          

        

      

      

        
Key terms

        
Review key terms from each Learn module. Select a card to view the definition, implementation notes, and code sample.

        

          
          1 / 7
          
        

        

          

            

              
Learn module

              
Analyze text with Azure Language in Foundry Tools

              
Detect language and entities

            

            

              
Definition, notes, and code

              

              
Azure AI Language provides detect-language, NER, and PII extraction APIs in Foundry Tools.

              

              
              

            

          

        

      

      

        
Knowledge check

        
After each attempt you can review recommended lessons below your score and retake the assessment.

        

        

      

      

        
Notes

        
        
 

      

      

        
      

    


    
    

      

        
Module 5

        
Implement Information Extraction Solutions

        
Implement information extraction and grounding pipelines: hybrid and vector search, OCR-based ingestion, and structured extraction with Azure AI Content Understanding and Document Intelligence.

      

      

        
Learning objectives

        
    
          
  • 5.1 — Ingest and index documents, images, audio, and video for grounding
    • 
          
  • 5.1 — Configure semantic, hybrid, and vector search; enrich with built-in or custom skills
    • 
          
  • 5.1 — Configure RAG ingestion with OCR; connect retrieval pipelines to agent tools
    • 
          
  • 5.2 — Extract with multimodal pipelines combining OCR, layout, and field extraction
    • 
          
  • 5.2 — Produce structured/Markdown output via Content Understanding for RAG and agents
    • 
        

      

      

        

          
Lessons

          Lesson 1 of 2
        

        
Each lesson aligns to a Microsoft Learn module and includes official terminology, worked examples, and Python samples.

        

        
Loading lessons…

        

          0 of 2 lessons completed
          

            

          

        

        

          
          
        

      

      

        
Scenario exercises

        
Select a scenario to view the recommended approach.

        

          
A search index must return verbatim captions and rerank by language understanding. Which feature?
Answer
Semantic ranker (semantic ranking) — reranks the top results using machine comprehension and returns semantic captions/answers from verbatim source text.

          
You must ingest scanned PDFs into a search index for RAG, including the text inside images. What do you configure?
Answer
OCR-based ingestion — use OCR / document extraction skills (or Content Understanding) to extract text and layout, chunk and vectorize it (integrated vectorization), then index for retrieval.

          
You need clean Markdown plus schema-aligned key-value fields from invoices for an agent. Which service?
Answer
Azure AI Content Understanding
  • content extraction produces structured Markdown
  • field extraction returns schema-aligned key-value fields with confidence scores and grounding (e.g., the prebuilt-invoice analyzer).

        

      

      
      

        
Hands-on labs

        

          
Official Microsoft Learn hands-on exercises
Azure labs

          
Do these on the official Microsoft Learn lab site (opens in a new tab). They use a real Azure subscription and the Microsoft Foundry portal.

          
        

        

          
Python: Parse Extracted Fields
Code Lab

          
Parse structured fields from a Content Understanding-style JSON result.

          

        

      

      

        
Key terms

        
Review key terms from each Learn module. Select a card to view the definition, implementation notes, and code sample.

        

          
          1 / 2
          
        

        

          

            

              
Learn module

              
Extract data with Azure Document Intelligence

              
Analyze documents

            

            

              
Definition, notes, and code

              

              
Document Intelligence extracts text, tables, and key-values via prebuilt or custom models; test in Document Intelligence Studio.

              

              
              

            

          

        

      

      

        
Knowledge check

        
After each attempt you can review recommended lessons below your score and retake the assessment.

        

        

      

      

        
Notes

        
        
 

      

      

        
      

    


  

    

      

        
📦 Question Bank

        
Instructor question management & import pipeline

      

      

        
        
      

    


    

      
—Published

      
—Drafts

      
—Pending Review

      
—Failed Jobs

    


    
Recent Import Jobs

    

      
Loading…

    

  


  
  

    

      

        
⬆ Import Questions

        
Upload a PDF, Word document, or scanned exam sheet

      

      
    


    

      
📄

      
Drag & drop a file here, or

      
      
Accepted: PDF, DOCX, PNG, JPG, TIFF · Max 50 MB

    


    
  


  
  

    

      

        
🔬 Review Studio

        
Reviewing import job

      

      

        
        
      

    


    

      
      

        
No items loaded.

      

      
      

        
Select a question from the list to edit it.

      

    

  


  
  

    

      

        
Draft Queue

        
Questions pending review across all import jobs

      

      
    

    

      
Loading…

    

  


  
  

    

      

        
🔍 Browse Question Bank

        
Search, filter, edit, and manage published questions

      

      

        
        
      

    


    

      
      
      
    


    

      
Loading…