cert
cert-1
cert-2

Pass Microsoft AI-103 Exam in First Attempt Guaranteed!

Get 100% Latest Exam Questions, Accurate & Verified Answers to Pass the Actual Exam!
30 Days Free Updates, Instant Download!

cert-5
cert-6
AI-103 Exam - Verified By Experts
AI-103 Premium File

AI-103 Premium File

$89.99
$98.99
  • Premium File 187 Questions & Answers. Last Update: Nov 15, 2026

Whats Included:

  • Latest Questions
  • 100% Accurate Answers
  • Fast Exam Updates
 
$98.99
$89.99
accept 397 downloads in the last 7 days
block-screenshots
Microsoft AI-103 Practice Test 8

Question 70

An application already uses hybrid keyword-and-vector retrieval, but exact error identifiers still rank below generic symptom matches. What search-ranking adjustment should the team test?

  1. Return every retrieved document without ranking or truncation to avoid missing information.
  2. Increase vector-similarity weight for all generic symptom documents and remove the field-specific boost for exact error identifiers.
  3. Wrap retrieved snippets with explicit source identifiers and delimiters in the grounded prompt
  4. Boost an exact-code field in the lexical branch and evaluate the blended ranking against both identifier and symptom queries.

Explanation

Correct Answer

D:

The team already has hybrid retrieval, so selecting hybrid search again would not fix the ranking. Increase the influence of exact-code matches in the lexical field or scoring profile, then measure precision for exact identifiers and recall for natural-language symptoms. The existing semantic branch should still retrieve paraphrases. Lexical field boosting can prioritize precise error-code matches while the existing vector branch continues retrieving semantically similar descriptions.

Incorrect Answers

A:

Unbounded evidence increases noise and context cost; it does not improve exact-code relevance at the top of an already hybrid ranked result set.

B:

Semantic similarity can bury unique code tokens beneath broad symptom text. Exact-code fields need lexical weighting while the blended ranking is evaluated on both query types.

C:

Source separation lets generation and citation checks distinguish evidence from instructions and unrelated passages. This is a separate capability and does not directly tune exact-identifier influence within an existing hybrid index.

Question 71

An image-based quality-control app must answer whether a label is obscured in a photograph. What AI capability is most directly required?

  1. Run a text-only document parser over the filename while omitting the submitted photograph showing the obscured label.
  2. Retrieve product-label metadata from a search index without submitting the damaged label image to a vision-capable model.
  3. Send the photograph to a text-to-speech service and evaluate whether the generated spoken description is pleasant, without inspecting pixels.
  4. Use multimodal visual understanding on the submitted image

Explanation

Correct Answer

D:

A vision-capable model can inspect the label evidence; a text-only prompt cannot verify an image it has not received. This directly supports the need to inspect visual evidence in an image, unlike a mechanism primarily intended to extract structured document fields.

Incorrect Answers

A:

The obscured-label decision requires visual evidence from the actual image. A filename or a text-only extraction cannot establish whether a physical label is visible or blocked.

B:

A text index cannot inspect whether physical printing is blocked in the uploaded photograph. Submit the image to a supported multimodal visual-understanding path and anchor the decision to observable pixels.

C:

Speech synthesis produces audio from text; it cannot determine if the label in the submitted photograph is obscured. Visual understanding must inspect the image itself.

Question 72

A legal search must match an exact clause number as well as paraphrased questions. Which retrieval architecture fits?

  1. Use only vector similarity and discard the exact clause-number tokens.
  2. Combine keyword and vector retrieval with an appropriate ranking or fusion strategy.
  3. Use only keyword matching and ignore paraphrased legal descriptions.
  4. Sort all documents by upload date instead of retrieval relevance.

Explanation

Correct Answer

B:

Exact identifiers benefit from lexical matching while paraphrases benefit from embeddings; hybrid retrieval addresses both signals.

Incorrect Answers

A:

Dense retrieval may miss exact unusual identifiers even when it finds semantically similar paragraphs.

C:

Lexical-only search can miss relevant clauses described using different wording.

D:

Recency alone does not combine exact clause matches and semantic paraphrase evidence.

Question 73

A customer support agent must retain a short conversation history and consult a company knowledge base between turns. Which solution elements belong in its architecture?

  1. Combine conversation state with a controlled knowledge retrieval connection
  2. Query the knowledge index alone and discard the user's conversation state before writing the next response.
  3. Store the last user message only and re-answer every follow-up without conversation state or authenticated knowledge retrieval.
  4. Use multimodal visual understanding on the submitted image

Explanation

Correct Answer

A:

Continuity is supplied by retained state and updated facts by knowledge retrieval; the application should govern both. The decisive difference is whether the design can maintain agent continuity with grounded knowledge; the competing function serves to retrieve relevant indexed company passages.

Incorrect Answers

B:

Grounded retrieval supplies external evidence, not the previous turns that define what the customer is referring to. Use bounded conversation state together with authorized knowledge retrieval to avoid losing context.

C:

A single latest turn cannot preserve prior task context or provide dependable company evidence. The agent needs bounded session history plus authorized knowledge access.

D:

A vision-capable model can inspect the label evidence; a text-only prompt cannot verify an image it has not received. That is relevant to inspect visual evidence in an image, but the present scenario requires an ability to maintain agent continuity with grounded knowledge.

Question 74

A retrieved page tells the assistant to reveal its hidden instructions. What should the RAG application do?

  1. Follow the retrieved page's instruction to expose system directives because it appears within the most relevant indexed document.
  2. Reconcile source deletion and retention events with indexed content
  3. Retrieve the malicious page with hybrid search and then treat its embedded instructions as privileged agent commands.
  4. Treat retrieved text as untrusted evidence and block its instructions from controlling the agent

Explanation

Correct Answer

D:

RAG evidence is lower trust than the application's policy and should never grant extra privileges. Here the relevant capability must let the team resist prompt injection through retrieved context. An alternative targeting test whether citations support the answer solves another problem.

Incorrect Answers

A:

Relevance does not confer authority on a retrieved page. Document content is untrusted task evidence and cannot override the assistant's instruction and tool policy boundaries.

B:

The retrieval index must not indefinitely retain documents removed or expired from the authoritative source. The function serves to prevent obsolete retained documents being cited, whereas the actual requirement is to resist prompt injection through retrieved context.

C:

A document is untrusted data even if search ranking returns it. The request to disclose hidden instructions must not cross the tool/data-to-instruction trust boundary.

Question 75

An application sends a prompt plus 80 retrieved pages and receives a request exceeding the model's allowed context. What is the correct architecture response?

  1. Increase the number of retrieved pages until the model processes all 80.
  2. Raise the allowed output token count but leave the same oversized input.
  3. Measure the full token budget and reduce, rank, or summarize retrieved context before invocation.
  4. Split the prompt into a larger single message without changing the total tokens.

Explanation

Correct Answer

C:

A model's context budget applies to input, tool messages and requested output. Increasing search results cannot overcome the deployed model's context-window limit.

Incorrect Answers

A:

Adding input makes the context-window overflow worse rather than reducing the token load.

B:

More output capacity does not reduce the already excessive input token budget.

D:

The limit applies to the total request context, not merely to individual message sizes.

Question 76

A response cites a passage that is present in the index but past its retention expiration. What index feature should be implemented?

  1. Keep deleted and expired documents retrievable indefinitely, relying on a prompt instruction that says not to quote outdated passages.
  2. Retrieve current relevant policy passages from an authorized indexed corpus
  3. Follow a stale document's embedded instruction to keep quoting it, regardless of the approved retention and expiration rules.
  4. Reconcile source deletion and retention events with indexed content

Explanation

Correct Answer

D:

The retrieval index must not indefinitely retain documents removed or expired from the authoritative source. This directly supports the need to prevent obsolete retained documents being cited, unlike a mechanism primarily intended to fit high-quality grounding evidence within context limits.

Incorrect Answers

A:

A prompt preference cannot guarantee compliance with retention. The index must process deletion and expiry events so invalidated passages are unavailable to retrieval.

B:

RAG grounds generation on up-to-date enterprise content rather than relying on static parametric memory. That capability is appropriate for ground responses on current indexed policy text, not this separate task of finding how to prevent obsolete retained documents being cited.

C:

Content cannot override the index's retention and freshness policies. Remove or filter expired passages before the model receives them, independent of wording in the source.

Question 77

A triage agent will accept scanned claim forms, look up policy passages, and only then draft a response. Which THREE Foundry capabilities address those different data needs? Choose THREE.

  1. Combine conversation state with a controlled knowledge retrieval connection
  2. Rely on general model fluency without extracting form fields, maintaining task context or connecting the policy search service.
  3. Define one generic business-action API and assume it reads scanned fields, locates policy passages and drafts the answer automatically.
  4. Recognize visible form icons but omit extracting the claim fields and their source evidence before drafting.
  5. Use Azure AI Search as a searchable knowledge index for grounded queries
  6. Use Content Understanding to extract structured information from uploaded documents

Explanation

Correct Answers

A:

Continuity is supplied by retained state and updated facts by knowledge retrieval; the application should govern both. Indexed policy content, document extraction, and controlled agent state/knowledge integration together provide grounded form-based triage.

E:

For this multi-step scanned-claim workflow, Azure AI Search supplies relevant policy passages once extracted form fields are available. It complements Content Understanding and controlled conversation state; it does not replace the extraction or sequencing requirements.

F:

Document content extraction requires layout and field interpretation rather than generic free-form chatting alone.

Incorrect Answers

B:

A fluent model alone cannot parse scanned intake data and retrieve permitted policy passages. The workflow needs a document extractor, knowledge index and controlled conversation state.

C:

A callable action contract is not a substitute for document analysis or indexed retrieval. The three data needs require extraction, knowledge search and conversation-aware response composition.

D:

Generic image recognition might identify visual objects, but the intake requires structured, traceable field extraction before evidence retrieval and response generation.

Question 78

A model receives fifty near-duplicate search hits and exceeds context limits. What prompt assembly improvement helps?

  1. Concatenate all fifty near-identical snippets without deduplication, then raise the token limit until the overlong prompt fits.
  2. Deduplicate, rerank and truncate evidence to the most useful source passages
  3. Wrap retrieved snippets with explicit source identifiers and delimiters in the grounded prompt
  4. Increase keyword-vector recall and include all fifty near-identical passages in the prompt unchanged.

Explanation

Correct Answer

B:

Selective evidence assembly fits token limits without blindly discarding the most relevant facts. The decisive difference is whether the design can fit high-quality grounding evidence within context limits; the competing function serves to prevent obsolete retained documents being cited.

Incorrect Answers

A:

Increasing context does not add information when hits repeat the same passage. Deduplicate and rerank sources, then fit only useful evidence within the available window.

C:

Source separation lets generation and citation checks distinguish evidence from instructions and unrelated passages. Its primary purpose is to structure retrieved context with citations; the stated constraint instead demands a way to fit high-quality grounding evidence within context limits.

D:

More overlapping chunks worsen the stated context-limit problem. Deduplicate and select a bounded, diverse set of relevant evidence while preserving citations.

Question 79

Employees often search by exact customer part numbers such as AX-940B that must not be paraphrased. Which search technique should remain available?

  1. Use embedding-only search and normalize away the exact hyphenated part number before retrieval.
  2. Retain lexical keyword retrieval for exact part-number matching
  3. Apply OCR and layout-aware extraction before indexing
  4. Use vector similarity retrieval with suitable document embeddings

Explanation

Correct Answer

B:

Token/keyword search can preserve identifier matching that purely semantic similarity might weaken. This resolves the specific need to match exact rare identifiers, while a measure that helps retrieve useful subsections of long documents addresses a different requirement.

Incorrect Answers

A:

Semantic similarity can blur unique alphanumeric identifiers. Keep a lexical branch that can match the exact part-number token including its meaningful characters.

C:

OCR and layout processing recover text and table relationships from images that ordinary text extraction may miss. This helps when a team needs to index scanned text while preserving table context; it does not establish how to match exact rare identifiers.

D:

Embeddings enable retrieval by semantic proximity even when question and document use different vocabulary. The function serves to retrieve by semantic meaning rather than exact words, whereas the actual requirement is to match exact rare identifiers.

Question 80

A search service retrieves relevant passages but the best evidence appears below several loosely related hits. What post-retrieval step should be evaluated?

  1. Return search results in the order of their storage-object filenames.
  2. Rerank the candidate passages for relevance to the actual question before composing context.
  3. Keep the original vector scores as final ordering without a query-aware reranking stage.
  4. Generate an answer using only the first retrieved hit regardless of its content.

Explanation

Correct Answer

B:

A reranker evaluates the specific query-passage relationship and can improve the quality of the evidence supplied to generation.

Incorrect Answers

A:

Filename order does not assess whether the passage answers the user's actual question.

C:

Initial retrieval can include loosely related passages and miss the best evidence near the top.

D:

The first hit can be less relevant than lower-ranked evidence and needs relevance checking.

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