Free Developing AI Apps and Agents on Azure Practice Questions

Sample Practice Questions

Try these 10 sample questions from the Developing AI Apps and Agents on Azure practice bank. Each includes the answer and explanation. Continue with the interactive practice above to review the full question set.

1. 1An application only needs to send a prompt to a model and display its response. It requires no tools or orchestration. Which Foundry starting point matches this requirement?

   1. A. A hosted agent
   2. B. A single model call
   3. C. A prompt agent
   4. D. A voice-based prompt agent

   Answer: B

   Explanation: Microsoft states, "If you only need to send prompts to a model with no tools or orchestration, start with a single model call." This application has exactly those requirements; adding an agent introduces unneeded orchestration.

   Reference: What is Microsoft Foundry? — Start by building an agent

2. 2A Python app constructs AIProjectClient for a Foundry project. Which endpoint path identifies the project rather than only the resource?

   1. A. https://<resource>.openai.azure.com/openai/v1
   2. B. https://<resource>.services.ai.azure.com/anthropic
   3. C. https://<resource>.services.ai.azure.com/api/projects/<project>
   4. D. https://<resource>.services.ai.azure.com

   Answer: C

   Explanation: The documented project endpoint is "https://<resource-name>.services.ai.azure.com/api/projects/<project-name>". AIProjectClient uses that project scope; the other URLs represent resource-level or model-provider API surfaces.

   Reference: Microsoft Foundry SDKs and endpoints — Foundry SDK — project endpoint

3. 3An app edits a selected region with the GPT-image Image Edit API. How should it identify the editable pixels in a valid same-size PNG mask?

   1. A. Set their alpha channel to 255
   2. B. Set their alpha channel to zero
   3. C. Set their alpha channel to 128
   4. D. Set their RGB channels to red

   Answer: B

   Explanation: The mask guidance states, "It defines the area of the image that you want the model to edit, using fully transparent pixels (alpha of zero) in those areas." Alpha zero marks edit regions; alpha 255 is opaque, alpha 128 is partial transparency, and RGB red alone does not set the required alpha.

   Reference: Azure OpenAI image generation models — Specify image edit API options — Mask

4. 4A text-extraction app needs output that conforms to a specified JSON Schema, not merely syntactically valid JSON. Which Azure OpenAI feature targets this requirement?

   1. A. JSON mode without a supplied schema
   2. B. Structured outputs with the supplied schema
   3. C. A larger model context window alone
   4. D. A lower sampling temperature alone

   Answer: B

   Explanation: The guide states, "Structured outputs make a model follow a JSON Schema definition that you provide as part of your inference API call." Older JSON mode guarantees JSON syntax rather than strict schema conformance; context size or temperature does not supply a schema.

   Reference: Structured outputs with Azure OpenAI — Overview — structured outputs versus JSON mode

5. 5A grounding pipeline needs exact keyword matching and vector similarity in one Azure AI Search request, with one combined result set. Which query approach supplies this?

   1. A. A metadata-only filter
   2. B. A semantic caption request without a search query
   3. C. A vector-only query with no text component
   4. D. A hybrid query with search and vectorQueries

   Answer: D

   Explanation: The guide states, "Hybrid search is a single query request configured for both full-text and vector queries." Metadata filtering, captions alone, or vector-only retrieval do not execute both requested search components.

   Reference: Hybrid search using vectors and full-text search in Azure AI Search — Hybrid search overview

6. 6A team wants Foundry to run an agent defined by instructions, a model, and attached tools. It wants no application code or containers to maintain. Which agent type fits?

   1. A. A hosted agent running a custom framework
   2. B. A containerized agent on managed compute
   3. C. A prompt agent configured declaratively
   4. D. A locally hosted agent using a custom runtime

   Answer: C

   Explanation: The guide says, "Foundry hosts and runs the agent for you, with no application code or containers to maintain." It identifies this declarative option as a prompt agent. Custom hosted or local runtimes require the code or container ownership excluded by the requirement.

   Reference: What is Microsoft Foundry? — Start by building an agent — Declarative

7. 7A Foundry app must list project connections, then invoke the Responses API. How should it use the two SDK client types?

   1. A. Project client for connections; OpenAI-compatible client for Responses
   2. B. OpenAI-compatible client for connections; project client for Responses
   3. C. Project client for both connection listing and Responses execution
   4. D. OpenAI-compatible client for both connection listing and Responses

   Answer: A

   Explanation: Microsoft states, "Use the project client for setup and configuration, then use the OpenAI-compatible client for running agents, evaluations, and calling models." Connections are project-native configuration, while Responses uses the compatible runtime client. Reversing or assigning both operations to one surface confuses these API roles.

   Reference: Microsoft Foundry SDKs and endpoints — Make your first Foundry SDK call

8. 8A GPT-image-1 generation response contains data[0].b64_json. What must the application do to obtain image bytes for a file?

   1. A. Treat b64_json as an HTTPS URL and download it
   2. B. Save the UTF-8 characters of b64_json as the image
   3. C. Parse b64_json as a JSON object with an image URL
   4. D. Base64-decode b64_json and save the decoded bytes

   Answer: D

   Explanation: Microsoft states, "The b64_json field contains the base64-encoded output image data." The application must decode that representation to image bytes; treating it as a URL, writing the encoded text, or parsing it as an embedded JSON object does not recover those bytes.

   Reference: Azure OpenAI image generation models — Output; Python SDK — decode image response

9. 9A strict structured-output schema includes a field whose value may be absent from the source text. How does the documented schema subset emulate an optional value?

   1. A. Remove every field from required
   2. B. Set additionalProperties to true
   3. C. Replace the whole root object with anyOf
   4. D. Keep the field required and allow its type to include null

   Answer: D

   Explanation: The guide says, "If needed, you can emulate an optional parameter by using a union type with null." Fields remain required under this schema subset, while nullable values represent missing information; permissive extra properties or a root anyOf do not implement that pattern.

   Reference: Structured outputs with Azure OpenAI — Supported schemas and limitations — All fields must be required

10. 10Which algorithm merges full-text and vector result rankings in Azure AI Search hybrid search?

    1. A. Reciprocal Rank Fusion (RRF)
    2. B. BM25 alone
    3. C. HNSW alone
    4. D. Exhaustive k-nearest neighbors alone

    Answer: A

    Explanation: Microsoft states, "An RRF algorithm merges the results." BM25 ranks text retrieval, while HNSW and exhaustive kNN serve vector retrieval; those component algorithms do not perform the hybrid rank fusion.

    Reference: Hybrid search using vectors and full-text search in Azure AI Search — How does hybrid search work?