McAfee-Secured Website

Exam Code: AI-103

Exam Name: Developing AI Apps and Agents on Azure

Certification Provider: Microsoft

Corresponding Certification: Microsoft Certified: Azure AI Apps and Agents Developer Associate

100% Microsoft AI-103 Guaranteed Success With Testking Training Materials

Get Real AI-103 Exam Questions & Expert Verified Answers as Experienced on the Real Exam!

187 Questions & Answers with Testing Engine

"Developing AI Apps and Agents on Azure Exam", also known as AI-103 exam, is a Microsoft certification exam.

Pass your tests with the always up-to-date AI-103 Exam Engine. Your AI-103 training materials keep you at the head of the pack!

Satisfaction Guaranteed

Money Back Guarantee

Testking provides hassle-free money back guarantee with our products. That is because we have 100% trust in the abilities of our professional and experience product team, and our record is a proof of that.

99.6% PASS RATE
Was: $137.49
Now: $124.99

Microsoft AI-103 Practice Test 22

Question 210

A maintenance assistant must answer a question about an attached equipment photo and a text instruction. A text-only model cannot inspect the photo. Which change enables the task?

  1. Choose an LLM with image input support for mixed image and text prompts.
  2. Extract metadata such as width and file type, then supply only those values to a text model.
  3. Describe the likely visual scene from text instructions without supplying the actual uploaded picture.
  4. Pass a file path as ordinary text to a deployment with no supported image input representation.

Explanation

Correct Answer

A:

Joint modalities provide the visual evidence needed to interpret the written question.

Incorrect Answers

B:

Dimensions and format do not communicate the actual photo content required to answer the visual question.

C:

A model needs the real image input to answer about its observed contents.

D:

A filename string does not provide the pixels the model must interpret.

Question 211

An app has a vision-capable model but sends an image by embedding its local filesystem path inside plain text. The model says it cannot see it. Which request construction matters?

  1. Attach the image bytes to the search-index name field and omit the inference image content.
  2. Write the path in larger letters inside the user prompt.
  3. Include the image using the API-supported multimodal input format.
  4. Change the image extension without changing the request body.

Explanation

Correct Answer

C:

A valid image input block conveys the actual visual data rather than an inaccessible local path.

Incorrect Answers

A:

Search configuration cannot substitute for the model's required multimodal request representation.

B:

Text formatting does not upload the referenced file to the inference service.

D:

A renamed path is still not a supported image-input payload.

Question 212

An analyst asks why a chart's two series cross at a particular point. OCR extracts visible labels but not the plotted relationship. What additional capability is needed?

  1. Use multimodal visual reasoning over the original chart together with the text question.
  2. Rebuild an approximate chart from OCR labels alone without validating the plotted marks and scale.
  3. Use OCR text alone and assume the series crossing from axis labels, ignoring plotted marks.
  4. Use only a string list of axis labels and infer the lines from their names.

Explanation

Correct Answer

A:

Interpreting trends and spatial intersections requires more than isolated OCR tokens.

Incorrect Answers

B:

Recognized labels alone cannot establish where the real values are plotted.

C:

OCR can recover labels but cannot establish the geometric crossing without interpreting the plotted chart.

D:

Labels do not encode actual plotted curves or intersection positions.

Question 213

A repo review assistant should explain whether a patch changes API behavior and test coverage, not merely summarize issue comments. Which model should be evaluated?

  1. Sort changed files by size and treat the largest as the root cause.
  2. Evaluate a code-capable model on actual diffs, affected call sites and associated test changes.
  3. Use an invoice OCR extractor on the repository's README.
  4. Generate a hypothetical patch explanation without reading the diff.

Explanation

Correct Answer

B:

Code-oriented evaluation tests whether the model understands actual changes across modules and tests.

Incorrect Answers

A:

File length is not a dependency or behavior analysis.

C:

Document OCR does not analyze code semantics or patch effects.

D:

A guess cannot substantiate repository-specific impact.

Question 214

A multilingual HR RAG assistant works in English but fails on French requests despite French documents in the corpus. What should tests include?

  1. Retrieve only English documents and translate the final answer without validating multilingual retrieval coverage.
  2. Judge multilingual readiness from one English-only benchmark.
  3. Test multilingual RAG queries against document-language coverage.
  4. Measure only model output latency as evidence of French correctness.

Explanation

Correct Answer

C:

Multilingual behavior depends on retrieval and content coverage, not simply the final model's language fluency.

Incorrect Answers

A:

Answer translation cannot recover non-English source evidence omitted by indexing or search.

B:

English success does not establish French retrieval recall.

D:

Speed says nothing about whether appropriate passages were found.

Question 215

A chatbot records explicit thumbs-down feedback and also computes automatic relevance scores. Analysts currently mix them into one column. What should be stored?

  1. Average both values without documenting differing scales or context.
  2. Rename all automatic scores as user satisfaction without recording their origin.
  3. Delete actual user feedback and retain only model self-evaluation.
  4. Keep direct user feedback distinct from evaluator-generated labels and preserve their provenance.

Explanation

Correct Answer

D:

Human judgments and model metrics have different meaning, bias and reliability characteristics.

Incorrect Answers

A:

Unexplained mixing can create misleading trends and thresholds.

B:

A synthetic evaluation score is not evidence a user expressed that opinion.

C:

Automated metrics do not fully represent the user's task outcome.

Question 216

A finance assistant must answer numerical policy questions exactly. The evaluation set contains only broad conversational prompts and misses arithmetic mistakes. What should the benchmark add?

  1. Evaluate only whether responses sound professional and concise.
  2. Measure the number of retrieved documents without testing final totals.
  3. Use labeled numerical cases with correct calculations and cited evidence.
  4. Remove cases involving explicit numbers because they are difficult.

Explanation

Correct Answer

C:

Task-specific gold cases expose arithmetic and citation defects that generic fluency tests miss.

Incorrect Answers

A:

Polished prose can contain incorrect monetary calculations.

B:

Candidate retrieval volume does not verify numerical answer correctness.

D:

Avoiding risky tasks conceals the application's actual performance requirements.

Question 217

An answer cites a safety clause but adds a fabricated exception. Which evaluator should compare the actual claim against the retrieved passage?

  1. Measure output-token count and treat shorter claims as true.
  2. Run groundedness/faithfulness evaluation on the cited context and generated statement.
  3. Check HTTP response success and assume evidence alignment.
  4. Use only a readability score for the final prose.

Explanation

Correct Answer

B:

Groundedness detects factual claims not entailed by source evidence.

Incorrect Answers

A:

Length cannot establish whether an exception exists in the source.

C:

Transport success does not verify factual entailment.

D:

Readable text may still contain invented conditions.

Question 218

A search stage returns passages about the wrong product family, although the model writes coherent answers. Which metric tests the passage selection?

  1. Evaluate retrieved-passage relevance against query intent.
  2. Measure how many completion tokens were used.
  3. Score only the grammatical quality of the final answer.
  4. Count server restarts instead of retrieved passage relevance.

Explanation

Correct Answer

A:

Retrieval relevance evaluates whether the evidence set actually matches user intent.

Incorrect Answers

B:

Token usage does not indicate correct evidence matching.

C:

Fluent output can be based on irrelevant retrieved material.

D:

Process health does not establish search relevance.

Question 219

A model answer says an employee gets 20 days of leave, but the attached policy passage states 15. What should quality checks flag?

  1. Accept the model's number if the answer sounds more confident.
  2. Detect answer fabrication when retrieved context conflicts with model reply.
  3. Replace the document with a generated citation matching the model's claim.
  4. Raise generation temperature so future answers vary more.

Explanation

Correct Answer

B:

A contradiction between output and source is a groundedness/factuality failure.

Incorrect Answers

A:

Confidence wording cannot supersede the policy source.

C:

Fabricated evidence would conceal the actual error.

D:

Sampling variance is not a reliable conflict resolution strategy.

Question 220

A JSON extraction prompt produces inconsistent field names across runs. Developers want stable structure without creating a custom model. What prompt technique may help?

  1. Use task-specific few-shot examples paired with a strict output schema and validation.
  2. Allow the model to invent new fields freely for each record.
  3. Set the maximum response length to one token and omit examples.
  4. Use broad generic prompt examples instead of examples that demonstrate the target JSON structure.

Explanation

Correct Answer

A:

Relevant demonstrations clarify desired extraction patterns while schemas enforce the receiving contract.

Incorrect Answers

B:

Unbounded keys do not meet downstream parser requirements.

C:

A tiny output budget cannot express the required structured record.

D:

Unrelated demonstrations do not teach the required field names and typing contract.

Developing AI Apps and Agents on Azure (AI-103) Practice Tests