Microsoft Developing AI Apps and Agents on Azure AI-103 Exam Questions

Answer : B

A vision-enabled Chat Completions request must represent the user message as a multimodal content array. One item contains the prompt with 'type': 'text', while another contains the image reference with 'type': 'image_url' and an image_url.url property:

'content': [

{'type': 'text', 'text': 'Analyze this image.'},

{'type': 'image_url', 'image_url': {'url': '<image-url>'}}

]

This structure explicitly identifies each modality, allowing the deployed model to route the text and image through the appropriate processing paths. A URL embedded in an ordinary string remains text and is not interpreted as visual input. Microsoft's documented Chat Completions schema shows the text and image objects inside the user message's content array.

Request metadata does not define model input, and a system message should contain behavioral instructions rather than the user's image attachment. Base64 images are supported, but the encoded value must be formatted as a data URL and supplied through the structured image item---not inserted into a plain content string.

Study Guide alignment: deploy and consume multimodal models, integrate generative workflows into applications, and design multimodal-understanding workflows.

====================

Answer : C

Azure Vision in Foundry Tools Spatial Analysis is designed to process real-time streaming video and analyze the presence, movement, and spatial relationships of people in physical environments. It provides predefined operations such as personcrossingpolygon, which can identify when a person enters or exits a configured zone. The resulting personZoneEnterExitEvent can be consumed by an application to trigger alerts when an employee enters an area designated as high risk.

Spatial Analysis minimizes development effort because it supplies pretrained people-detection and zone-monitoring capabilities. Developers configure camera streams, polygonal zones, and event parameters instead of collecting images, labeling objects, and training a custom model.

Image Analysis primarily evaluates individual images and does not maintain movement or zone state across a live stream. Azure AI Video Indexer extracts searchable insights from video or live-stream content, but it is not the specialized option for detecting physical zone-entry events. Custom Vision object detection would require training data, labeling, model training, deployment, and additional application logic to track movement.

The Study Guide explicitly associates Spatial Analysis with detecting the presence and movement of people in video under Implement computer vision solutions Analyze videos.

Answer : D

Set tool_choice to the named Grounding with Bing Search tool:

{'type': 'bing_grounding'}

A named tool choice directs the model to invoke a specific tool rather than allowing it to select from all tools attached to the agent. Microsoft's SDK specification describes AgentsNamedToolChoice as the mechanism used to force the model to call a particular tool and lists bing_grounding and azure_ai_search as supported named tool types.

Grounding with Bing Search retrieves current public information from the internet and supplies the results to the model for response generation. Selecting bing_grounding therefore directs the run to the public web source rather than the private Azure AI Search index.

Setting tool_choice to 'required' only establishes that a tool must be invoked; when multiple tools are available, it does not identify which one must be selected. 'auto' delegates the decision entirely to the model. Selecting {'type': 'azure_ai_search'} would force retrieval from the private product-documentation index, which is the opposite of the requirement.

Study Guide alignment: configure agent tools, control tool selection, integrate web grounding, and connect agents to enterprise knowledge sources.

====================

Answer : A, C

Select AI quality (AI assisted) metrics for groundedness and relevance. Groundedness evaluates whether the generated answer is supported by the retrieved RAG context, while relevance evaluates whether the answer appropriately addresses the user's query. These evaluations use an AI judge model to assess meaning and contextual alignment rather than relying only on exact token matches. Microsoft Foundry categorizes groundedness and relevance as quality evaluators and requires a compatible judge-model deployment for AI-assisted evaluation.

Select risk and safety metrics to evaluate harmful-content categories. These evaluators detect and score content involving violence, sexual material, self-harm, and hate or unfairness. They provide the safety dimension required alongside the RAG quality measurements.

AI quality NLP metrics include mathematically calculated measures such as F1, BLEU, GLEU, ROUGE, and METEOR. These primarily compare token or n-gram overlap with reference answers and do not provide the requested groundedness and content-harm assessments. A fluency evaluator measures language readability, while a similarity evaluator compares a response with ground truth; neither is a complete evaluation category covering the requirements.

Study Guide alignment: evaluate generative AI solutions by using quality, groundedness, relevance, risk, and safety metrics.

Answer : A

Use vector search because it performs similarity matching based on semantic meaning rather than requiring literal keyword overlap. During ingestion, the product manuals should be divided into appropriately sized chunks. An embedding model converts each chunk into a numerical vector, which Azure AI Search stores in a vector field. At query time, the user's question is also converted into an embedding, and the service retrieves chunks whose vectors are closest to the query vector. Consequently, wording such as ''reset the device to factory settings'' can match manual content such as ''restore the unit's original configuration,'' despite minimal lexical overlap. Azure AI Search explicitly defines vector fields as enabling similarity searches over mathematically represented content and supports this retrieval pattern for RAG applications.

Semantic ranking improves the ordering of an existing result set but is not the indexing mechanism that stores embeddings and performs nearest-neighbor similarity matching. Suggesters provide search-as-you-type functionality, while analyzers control tokenization and lexical processing for text fields.

This aligns with the AI-103 Study Guide objectives for implementing RAG, configuring Azure AI Search grounding, generating embeddings, and selecting vector retrieval techniques for agentic solutions.

====================

Answer : B

Configure hybrid search, which executes full-text and vector queries within the same Azure AI Search request. The full-text component searches the product-name and product-code fields through the lexical index, providing the precision required for exact or near-exact identifiers. Microsoft specifically identifies product codes and other specialized terms as scenarios that frequently perform better with keyword search.

The vector component compares the embedding of the user's natural-language query with the embeddings stored for product descriptions. This retrieves semantically similar products even when the query and indexed description do not share the same literal words. Azure AI Search runs the full-text and vector searches in parallel and combines their result sets by using Reciprocal Rank Fusion, returning one unified ranking to the Foundry agent.

Keyword-only search would preserve exact matching but perform poorly for conceptual or paraphrased descriptions. Vector-only search supports semantic similarity but can miss precise product codes and rare identifiers. Semantic search alone reranks text-search results using language understanding; it does not replace the vector query required to use the existing embedding field.

Study Guide alignment: configure semantic, hybrid, and vector search for grounding, choose an appropriate retrieval method, and connect retrieval pipelines to agent tools.

Answer : D

The temperature parameter controls randomness during token selection. A higher temperature broadens the probability distribution and permits less-likely tokens to be selected, producing more varied or creative responses. A lower temperature concentrates selection on the highest-probability tokens, resulting in responses that are more focused, concrete, and consistent across repeated requests. Microsoft's prompt-engineering guidance explicitly states that higher temperatures produce more divergent output, whereas lower values reduce randomness.

Therefore, decreasing the temperature---typically toward zero for automated validation scenarios---is the appropriate configuration change. It reduces wording and structural variation without changing the prompt or intentionally limiting the information available to the model. Although probabilistic model execution does not guarantee byte-for-byte identical output in every circumstance, this is the correct parameter for minimizing variability among the available choices.

Increasing max_tokens only raises the maximum response length and does not control sampling randomness. Stop sequences define where generation terminates; removing them can change response length but does not make responses deterministic. Increasing temperature would directly increase variability.

Study Guide alignment: Optimize and operationalize generative AI systems---tune generation behavior through prompt engineering and model parameters, and evaluate agent behavior for production reliability.

====================