Practice Exam AI-103: Azure AI Apps and Agents Developer Associate

Question bank: 100 · timed mock exam: 120 min

Official exam page on the vendor's site (opens in a new tab) Official exam guide (opens in a new tab) Official prep course (opens in a new tab)

Exam guide: format, domains and how to prepare

Compare: AI-901 vs AI-103

  • Implement generative AI and agentic solutions 34 questions
  • Plan and manage an Azure AI solution 30 questions
  • Implement computer vision solutions 12 questions
  • Implement text analysis solutions 12 questions
  • Implement information extraction solutions 12 questions
Study mode (free) Exam mode (subscribers)

Free: 25 questions in study mode, all domains, no account.

Subscription: the 100 questions of this certification, the timed exam mode, and all other certifications too.

Subscribe · €9.99 / month

Secure card payment (Stripe) · No commitment · Cancel online

How to prepare for Azure AI Apps and Agents Developer Associate with InfraPrep

The 100 questions cover 5 domains: Implement generative AI and agentic solutions, Plan and manage an Azure AI solution, Implement computer vision solutions, Implement text analysis solutions, Implement information extraction solutions. 25 of them are free, without an account.

Study mode

Answer at your pace: each answer is corrected straight away, with an explanation and the link to the documentation it relies on.

Exam mode

A timed exam of 120 minutes, with a question map to flag and come back to questions before you submit.

Results by domain

Your score is broken down by domain, so you know what to review first.

Original questions

Every question is written from the public documentation, never copied from the exam. You can contest an answer from study mode.

10 free sample questions

Question 1 Plan and manage an Azure AI solution

A retailer is building a shopping assistant with Foundry Agent Service. Returning customers complain that they must repeat their clothing sizes and preferred brands every time they open a new session. Which capability should the developer add so the agent keeps this user-specific context from one session to the next?

  1. The code interpreter tool with a stored preferences script
  2. A Foundry IQ knowledge base built from the product catalog
  3. The file search tool pointed at uploaded customer documents
  4. Memory in Foundry Agent Service attached to the agent
Show the answer and the explanation

Correct answer: D

The documentation positions memory as the capability for user-specific context that is learned and persists through interactions over time, across sessions.

Source: official documentation

Question 2 Implement generative AI and agentic solutions

An application extracts invoice fields with an Azure OpenAI model and feeds them into an ERP import. The team currently uses JSON mode, but some responses are valid JSON that miss required fields or add unexpected keys. Which feature should they switch to?

  1. Prompt caching of the invoice extraction prompt
  2. Predicted outputs with the expected invoice template
  3. A lower temperature combined with JSON mode
  4. Structured outputs with a strict JSON schema
Show the answer and the explanation

Correct answer: D

Structured outputs make the model follow the supplied JSON schema, whereas the older JSON mode only guaranteed valid JSON without strict adherence to the schema.

Source: official documentation

Question 3 Implement computer vision solutions

A marketing app uses the image edit API with a gpt-image series model to replace only the sky in an existing product photo, leaving the rest untouched. How should the developer prepare the mask?

  1. A PNG mask of the same size as the photo, with only the product left transparent
  2. A JPEG mask at half the photo's size, with the sky area painted white
  3. A PNG mask of the same size as the photo, with the sky area fully transparent
  4. A PNG mask of the same size as the photo, with the sky made fully opaque black
Show the answer and the explanation

Correct answer: C

The mask marks the area to edit with fully transparent pixels (alpha of zero), and it must be a PNG file with the same dimensions as the input image.

Source: official documentation

Question 4 Implement text analysis solutions

A legal team wants to translate one contract at a time with Azure Translator, keep the original layout, get the translated file back in the same response, and avoid provisioning any storage account. Which option fits?

  1. Custom Translator model training
  2. Single file document translation (synchronous)
  3. Batch document translation (asynchronous)
  4. Text translation of the extracted contract text
Show the answer and the explanation

Correct answer: B

Synchronous single file translation keeps the structure and format of the document, returns the translated document directly to the caller, and doesn't require an Azure Blob Storage account.

Source: official documentation

Question 5 Implement information extraction solutions

A parts distributor grounds its agent on an Azure AI Search index. Users sometimes type exact part numbers and sometimes describe a part in their own words ("the rubber seal for the rear door"). The team wants one query request that handles both well. Which retrieval approach should it configure?

  1. Full-text search with autocomplete suggestions
  2. Pure vector search on the embeddings field
  3. Hybrid search combining full-text and vector queries
  4. Vector search with a stricter nearest neighbors k
Show the answer and the explanation

Correct answer: C

Hybrid search combines the strengths of vector and keyword search in a single request; keyword search catches exact matches such as product codes, while vector search finds conceptually similar content.

Source: official documentation

Question 6 Plan and manage an Azure AI solution

An insurance company wants a chatbot that answers questions using its internal policy documents, which are revised every week. According to the Microsoft Foundry guidance on choosing an approach for private data, which approach fits this need?

  1. Fine-tune the model every week on the latest policy documents
  2. Raise the model temperature so it improvises on policy topics
  3. Deploy a larger model that already knows more public content
  4. Use retrieval augmented generation (RAG) over the policy documents
Show the answer and the explanation

Correct answer: D

Foundry recommends RAG when answers must be grounded in private or frequently changing data, which matches weekly revised internal documents.

Source: official documentation

Question 7 Implement generative AI and agentic solutions

A developer enables structured outputs on requests that also provide several function tools to an Azure OpenAI deployment. The requests do not behave as expected when the model tries to call more than one tool in the same turn. Which setting should the developer change?

  1. Allow additional properties in every schema object
  2. Switch the request back to JSON mode for tool calls
  3. Set parallel_tool_calls to false on the request
  4. Remove the strict property from the JSON schema
Show the answer and the explanation

Correct answer: C

The documentation states that structured outputs are not supported with parallel function calls and that parallel_tool_calls must be set to false.

Source: official documentation

Question 8 Implement computer vision solutions

A broadcaster uses Azure Content Understanding to analyze video and wants custom fields such as the brands that appear in each segment. In the two-stage video pipeline, which stage produces these custom fields?

  1. Shot detection, which returns shot boundary timestamps
  2. Content extraction, which captures transcripts and shots
  3. Field extraction, which uses a generative model
  4. Key frame extraction, which picks thumbnails per shot
Show the answer and the explanation

Correct answer: C

The second stage, field extraction, uses a generative model to produce custom fields and perform segmentation.

Source: official documentation

Question 9 Implement text analysis solutions

A contact center wants to redact names and phone numbers from chat logs that are stored as turn-by-turn conversations before they are used for analytics. Which PII detection feature type in Azure Language in Foundry Tools is optimized for this input?

  1. Conversation PII
  2. Text PII
  3. Document-based PII
  4. Custom named entity recognition
Show the answer and the explanation

Correct answer: A

Conversation PII is optimized for turn-based transcript and chat structures, which matches turn-by-turn chat logs.

Source: official documentation

Question 10 Implement information extraction solutions

A team wants Azure AI Search itself to chunk its source documents and call an Azure OpenAI embedding model during indexing (integrated vectorization), rather than pushing precomputed vectors. What does this internal processing approach require?

  1. A push API client, an external embedding job, and a vector profile per field
  2. An indexer, a supported data source, and a skillset that drives chunking and embedding
  3. A semantic ranker configuration, a scoring profile, and a synonym map on the index
  4. A knowledge store, a Foundry agent, and a vector store for the file search tool
Show the answer and the explanation

Correct answer: B

For internal processing, Azure AI Search offers integrated chunking and vectorization in an indexer pipeline, which requires an indexer, a supported data source, and a skillset that drives chunking and embedding.

Source: official documentation

Frequently asked questions

Are there free questions, and do I need an account?

Yes. 25 questions of this certification are free in study mode, across all its domains, without a subscription and without registration. The account is created when you subscribe, at checkout.

How does the subscription work, and can I cancel?

One monthly subscription covers all certifications and the timed practice exams. There is no minimum period. You cancel online, at any time, with the "Cancel your contract" function in My account; cancellation takes effect at the end of the month already paid, and no further payment is taken.

Where do the questions come from?

Every question is original, written from the vendor's public documentation. Each answer comes with an explanation and a link to the documentation page it relies on. InfraPrep is not affiliated with the vendors, and certification names belong to their owners.

In which language are the questions?

The questions are written in English, also on the French version of the site.

Is the catalog updated?

Yes, the catalog changes over time: new certifications and new questions are added.