AI-103 · Sample questions

AI-103 sample questions with answers and explanations

9 real questions from the CertifyDeck AI-103 bank, spread across the domains of the official Microsoft Certified: Azure AI Apps and Agents Developer Associate skills outline, with the correct answer and a full explanation of why the other options are wrong. They are original questions written from the public syllabus, not exam dumps, and they are the same style you will find in the practice tests and mock exams.

1. Question 1Domain: Implement generative AI and agentic solutionsMedium

   A RAG app retrieves passages, but the model ignores source identifiers. What should context assembly include?

   • A.Explicit source labels and instructions to cite supplied evidence
   • B.Higher temperature without sources
   • C.The entire index without ranking
   • D.A random citation after generation
   Show answer and explanation

   Correct answer: A. Explicit source labels and instructions to cite supplied evidence

   Explanation

   Source labels and grounding instructions associate claims with retrieved evidence; post-processing cannot invent valid citations.

   Official references

   • AI-103 study guide
2. Question 2Domain: Implement computer vision solutionsMedium

   A generated clip has the right subject but an incorrect final segment. What is the efficient editing approach?

   • A.Regenerate unrelated audio only
   • B.Convert the clip to a search index
   • C.Discard all source controls
   • D.Edit or regenerate the affected segment while preserving the rest
   Show answer and explanation

   Correct answer: D. Edit or regenerate the affected segment while preserving the rest

   Explanation

   Segment-level editing preserves accepted content and limits cost and unintended variation.

   Official references

   • AI-103 study guide
3. Question 3Domain: Implement computer vision solutionsMedium

   An inspection app must reason about a photo and a technician's question together. Which model input is required?

   • A.A keyword index only
   • B.A multimodal request containing both image and text
   • C.A text-only request without the photo
   • D.Speech output only
   Show answer and explanation

   Correct answer: B. A multimodal request containing both image and text

   Explanation

   A multimodal model can jointly interpret visual evidence and the natural-language question.

   Official references

   • AI-103 study guide
4. Question 4Domain: Plan and manage an Azure AI solutionMedium

   A support app classifies short requests on edge devices with strict latency and cost limits. Which model type should be evaluated first?

   • A.The largest multimodal model
   • B.An image generator
   • C.A speech synthesis model
   • D.A small language model optimized for the task
   Show answer and explanation

   Correct answer: D. A small language model optimized for the task

   Explanation

   A small language model fits a narrow task with edge, latency, and cost constraints; the other model types add unnecessary capabilities.

   Official references

   • AI-103 study guide
5. Question 5Domain: Plan and manage an Azure AI solutionMedium

   An order agent must retrieve policy documents and query a live transactional API. How should the two capabilities be represented?

   • A.A knowledge source for policies and a tool for the API
   • B.Both as permanent model weights
   • C.Both only as conversation history
   • D.A content filter for the API
   Show answer and explanation

   Correct answer: A. A knowledge source for policies and a tool for the API

   Explanation

   Knowledge sources ground informational answers; tools perform actions and obtain current transactional data.

   Official references

   • AI-103 study guide
6. Question 6Domain: Plan and manage an Azure AI solutionMedium

   A Foundry app must access Search and Storage without secrets, and traffic must remain private. Which design meets both requirements?

   • A.Embedded access keys and public endpoints
   • B.Anonymous access with a header
   • C.A public IP protected by prompts
   • D.Managed identities, private endpoints, and private DNS
   Show answer and explanation

   Correct answer: D. Managed identities, private endpoints, and private DNS

   Explanation

   Managed identities provide keyless authentication; private endpoints and private DNS keep service traffic on private address space.

   Official references

   • AI-103 study guide
7. Question 7Domain: Plan and manage an Azure AI solutionMedium

   An expense agent can draft reimbursements, but payments above EUR 1,000 require a manager. Which control best enforces this?

   • A.Mention the threshold in the welcome message
   • B.Give unrestricted credentials
   • C.Raise temperature for costly requests
   • D.Restrict the payment tool and require approval above the threshold
   Show answer and explanation

   Correct answer: D. Restrict the payment tool and require approval above the threshold

   Explanation

   Tool authorization plus a deterministic approval gate enforces the threshold even if generated text is incorrect.

   Official references

   • AI-103 study guide
8. Question 8Domain: Implement information extraction solutionsMedium

   A RAG corpus contains PDFs, scans, images, and recorded interviews. What ingestion design is appropriate?

   • A.Treat every binary file as plain text
   • B.Index filenames only
   • C.Discard metadata
   • D.Route each modality through suitable extraction before chunking and indexing normalized content
   Show answer and explanation

   Correct answer: D. Route each modality through suitable extraction before chunking and indexing normalized content

   Explanation

   Modality-specific extraction produces usable normalized content while preserving metadata for retrieval.

   Official references

   • AI-103 study guide
9. Question 9Domain: Implement information extraction solutionsMedium

   A RAG ingestion flow receives scanned reports with complex tables. What should happen before embeddings are created?

   • A.Generate citations without extraction
   • B.Apply OCR and layout extraction, then chunk the structured content with metadata
   • C.Embed the encrypted binary directly
   • D.Remove all table relationships
   Show answer and explanation

   Correct answer: B. Apply OCR and layout extraction, then chunk the structured content with metadata

   Explanation

   OCR and layout processing recover meaningful structure that should be chunked before vectorization.

   Official references

   • AI-103 study guide