AI-103 Developing AI Apps and Agents on Azure Questions and Answers

The correct observability signal is B. groundedness evaluation metrics . In a RAG solution, the key diagnostic question is whether the generated answer is supported by the retrieved context. Microsoft Foundry’s built-in evaluator reference defines Groundedness as the metric that measures how grounded the response is in the retrieved context, with scoring that indicates whether the model’s claims are supported by the provided source material.

This matches the issue after a content update. If retrieved chunks are stale, misleading, incomplete, or poorly aligned with the user query, groundedness results can show that generated responses are not reliably supported by the retrieved documentation. The RAG evaluator guidance explains that groundedness focuses on whether the response avoids content outside the grounding context, while other process metrics such as retrieval evaluate how relevant the retrieved chunks are. Latency traces are useful for performance troubleshooting, not response accuracy. Indexer status can reveal ingestion failures, but it does not show whether retrieved content is influencing generated answers negatively. Prediction drift is a model monitoring concept and is not the primary signal for RAG grounding quality. Reference topics: Microsoft Foundry observability, RAG evaluators, groundedness, retrieved context, and response quality evaluation.

The solution does not meet the goal. Protected material detection is intended to identify large language model output that matches known protected text or code, such as copyrighted text, selected web content, song lyrics, articles, recipes, or code. Microsoft describes protected material detection as a control for preventing AI-generated content from reproducing known protected material, not as a control for image safety or prompt injection.

The stated risk has two parts: users can upload unsafe images, and users can embed hidden instructions in images to manipulate the model. Unsafe image uploads require image moderation, because Azure AI Content Safety provides image APIs that detect harmful content across modalities and can support blocking decisions by harm category and severity. Hidden instructions extracted from images are indirect prompt injection or document attacks; Microsoft Prompt Shields are the capability designed to detect user prompt attacks and document attacks, including harmful instructions embedded in third-party content.

Therefore, protected material detection alone does not mitigate either primary risk. Reference topics: Azure AI Content Safety, image moderation, Prompt Shields, document attacks, indirect prompt injection, and protected material detection.

The solution requires a generative text model, so the Cognitive Services account must be created with the OpenAI kind from the values supplied in the exhibit.

The JSON object shown underneath contains Key Vault-based encryption settings such as the key source, key vault URI, key name, and key version. That JSON therefore belongs to the --encryption parameter.

Customer-managed key encryption lets the organization control the key used to protect supported stored data. This provides additional control over rotation, revocation, key governance, and separation of duties compared with relying solely on Microsoft-managed keys.

The Azure AI resource also needs permission to use the Key Vault key. In an operational deployment, this generally involves configuring a managed identity and granting the required Key Vault permissions.

--api-properties configures service-specific API properties rather than encryption. --assign-identity enables the account identity but does not itself supply the customer-managed encryption configuration shown in the question.

Therefore, the correct selections are OpenAI and --encryption .

Study Guide alignment: design Azure infrastructure, configure model resources, and secure AI systems through identity and resource-security controls . Microsoft Learn

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

To prevent access from the internet: Create a private endpoint

To limit access to query specific indexes: Use Azure roles

Create an Azure Private Link private endpoint for Search1 and disable public network access. The private endpoint assigns Search1 a private IP address within an Azure virtual network, allowing the hosted applications to reach the search service over the Microsoft backbone instead of through its public endpoint. Disabling public network access ensures that requests originating from the public internet are rejected. An IP firewall would still expose the public endpoint to approved public IP addresses and therefore would not provide complete internet isolation.

For index-level authorization, enable Microsoft Entra ID role-based access control and assign Azure roles to each application’s managed identity or service principal. The Search Index Data Reader role permits query and retrieval operations without allowing index modification. Azure AI Search supports scoping Search Index Data Reader or Search Index Data Contributor permissions to an individual index; custom role definitions can also be used when more precise permissions are required.

Query-key authentication is unsuitable because API keys normally provide service-level access and cannot reliably isolate applications to designated indexes.

Study Guide alignment: secure Azure AI services by configuring private endpoints, network isolation, managed identities, Microsoft Entra authentication, and role-based access control.

Request approval to run the container.

Export model1 to Host1.

Run the container.

Approval is the first step because custom Speech container scenarios that operate with locally stored models, particularly disconnected deployments, are gated. Microsoft requires an access request to be submitted and approved before the applicable Speech container can be used in a disconnected environment. The Azure Speech resources used for the container must be associated with the approved subscription.

After approval, export or download the trained custom speech-to-text model to Host1. The model files must be available through a Docker volume mounted to the container’s model directory. The custom model determines the recognition locale and supplies the specialized acoustic or language behavior required by the container.

Finally, run the container and provide the model location or model identifier, volume mount, license and billing configuration, API key, and required compute allocation. Microsoft documents that the custom Speech container loads the model from the mounted volume and then exposes the speech-to-text service through its configured port.

Retraining is unnecessary because model1 already exists. Disk logging is optional operational configuration and is not required to deploy the model.

Study Guide alignment: implement custom speech solutions and deploy Azure AI services in containers .

Comprehensive and Detailed 150 to 250 words of Explanation From Azure AI Apps and Agents Developer/Course Guide/topics:

Few-shot learning , more precisely few-shot prompting, supplies a small number of representative input/output examples as part of the prompt. Those examples demonstrate the intended classification behavior before the model receives the new support ticket.

For example, the prompt might show one ticket classified as Billing, one as AccountAccess, and one as TechnicalIssue. The model can infer the expected labels, style, and output structure from the demonstrations and then apply the pattern to the incoming ticket.

The model ' s weights are not updated. The examples exist only in the prompt context for that invocation or conversation, so the approach satisfies both requirements: no retraining and no permanent knowledge storage.

Zero-shot prompting would supply only task instructions and no demonstrations. RAG retrieves external source information to ground model responses; it is not primarily a mechanism for showing examples of desired classification outputs. Chain-of-thought relates to reasoning behavior rather than labeled demonstration examples.

Few-shot prompting is therefore the direct method for increasing task accuracy by showing the model what correct results look like.

Study Guide alignment: Optimize generative AI systems through prompt engineering and adjust model behavior without retraining . Microsoft Learn

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

Comprehensive and Detailed 150 to 250 words of Explanation From Azure AI Apps and Agents Developer/Course Guide/topics:

Protected material detection addresses a different safety concern from standard objectionable-content moderation. It is intended to identify protected textual material or other supported protected-content classes rather than evaluate normal harm categories such as violence, sexual content, self-harm, and hate/fairness.

To optimize a harmful-content filter, representative prompts should be submitted through the Moderate text content interface. Microsoft documents that the moderation experience returns detected content categories and severity levels and lets the user adjust filter thresholds before rerunning the same content. Microsoft Learn

Because the objective is to tune the filter configuration applied to objectionable user input and generated output, the tests must produce the same harm-category signals used by those filters. Protected-material detection cannot provide those severity measurements and therefore cannot determine whether the configured harmful-content thresholds are correct.

Protected-material detection can still be an additional responsible-AI control when an application needs to identify protected or copyrighted-like material, but that is independent of the requirement in this question.

The AI-103 Study Guide distinguishes content moderation, safety filters, risk detection, safety evaluations, and related responsible-AI controls.

The correct answer is a workflow . Microsoft Foundry workflows are designed to orchestrate agents and business logic as declarative, predefined sequences of actions. The official workflow guidance states that workflows are ideal when you need to orchestrate multiple agents in a repeatable process, add branching logic such as if/else, and handle variables without writing application orchestration code. This directly matches the requirement for a deterministic, step-based process with conditional branching and shared state.

In this scenario, TriageAgent can classify the request first, the workflow can store the triage result, and conditional logic can determine whether to invoke PolicyAgent, ActionAgent, or both. The ticket action is optional, so it should be triggered through a workflow condition based on the triage output. This minimizes development effort because the branching, sequencing, and variable handling are managed in the Foundry workflow rather than being manually implemented across separate runs in application code.

A group chat session is better for dynamic agent handoff, not a strict deterministic process. Threads and runs or separate app-coordinated calls require more custom orchestration. Reference topics: Microsoft Foundry workflows, multi-agent orchestration, conditional branching, variable handling, and agent-driven workflows.

Use image type detection to distinguish drawings from ordinary photographic content. Azure Vision’s ImageType visual feature returns information including lineDrawingType, which indicates whether the submitted image is a line drawing, and clipArtType, which estimates whether it is clip art. The application can use these pretrained results to implement the required photograph-versus-drawing categorization without collecting labeled images or training a custom classifier.

Use image descriptions to generate the caption. Azure Vision Image Analysis provides the Caption visual feature, which generates a concise, one-sentence description of the overall image. Dense Captions can additionally describe individual regions, but the standard Caption feature directly satisfies the stated requirement.

Object detection identifies and locates individual objects by using bounding boxes; it does not determine whether the entire image is a photograph or drawing. Content tags produce descriptive keywords rather than a natural-language caption. Azure Custom Vision classification could be trained for the image-type distinction, but it would require image collection, labeling, training, evaluation, and deployment, increasing development effort unnecessarily.

Study Guide alignment: configure applications to produce captions and implement solutions that identify visual characteristics within images .

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

Endpoint selection: assistants

Function declaration property: tools=[

Comprehensive and Detailed 150 to 250 words of Explanation From Azure AI Apps and Agents Developer/Course Guide/topics:

The command is creating/configuring an assistant that has access to a custom function named get_current_time, so the REST resource shown in the exhibit is assistants .

For the function definition, current Azure OpenAI/Foundry function-calling syntax uses the tools collection. A function tool is represented by an item with " type " : " function " and a nested " function " object containing its name, description, JSON parameter schema, and required arguments.

Microsoft explicitly documents that the older functions and function_call parameters have been deprecated and that tools and tool_choice should be used instead. Microsoft Learn

Therefore, although the embedded exhibit ' s answer image shows functions=[, the currently verified Microsoft syntax is tools=[ .

content represents message content rather than an agent tool definition. tool_resources is used for resources required by certain hosted tools and is not where a custom function schema is declared.

The AI-103 Study Guide requires candidates to define tool schemas and integrate agent tools, including APIs and custom functions . Microsoft Learn

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

Comprehensive and Detailed 150 to 250 words of Explanation From Azure AI Apps and Agents Developer/Course Guide/topics:

The system message is an example of priming because it establishes contextual expectations before the user supplies the actual mathematical problem.

The first sentence assigns the model a role: an assistant specialized in mathematical puzzles. The second establishes the expected communication style and audience level by requiring explanations suitable for a four-year-old. These instructions influence subsequent responses without modifying model weights.

It is not few-shot learning because no input/output examples are provided. Few-shot prompting would include several representative puzzles together with correctly formatted answers.

It is not chain-of-thought because the prompt does not explicitly provide or request a structured sequence of intermediate reasoning steps. An affordance is a cue that indicates a possible behavior or completion, but role and behavioral context established through a system message is most accurately described by the available option priming .

This technique is commonly used to establish persona, domain boundaries, tone, output style, and operating assumptions for a generative application.

The AI-103 Study Guide includes prompt engineering and adjustment of model behavior as part of optimizing generative AI systems. Microsoft Learn

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

Comprehensive and Detailed 150 to 250 words of Explanation From Azure AI Apps and Agents Developer/Course Guide/topics:

The Safety metaprompt is not the Content Safety feature intended to measure harm-category severity and tune moderation thresholds against representative sample text.

A safety metaprompt provides additional behavioral instructions intended to guide a generative model toward safer responses. It can complement other safety controls, but it does not perform the specific text moderation test required to determine whether configured severity thresholds accept or reject sample prompts.

For threshold testing, Microsoft provides the Moderate text content experience. The user supplies sample text, selects Run test , and Content Safety returns detected categories and severity values. The Configure filters interface can then be used to change allowed or prohibited severity levels and retest the samples. Microsoft documents severity classifications including safe, low, medium, and high. Microsoft Learn

Therefore, the proposed Safety metaprompt solution does not meet the stated requirement.

This maps to the AI-103 objectives to configure safety filters, guardrails, risk detection, and content moderation , and to apply responsible-AI safety evaluations. Microsoft Learn

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

Azure Language in Foundry Tools provides sentiment analysis and opinion mining , which are the appropriate natural language processing capabilities for evaluating public perception. Sentiment analysis processes unstructured social-media text and assigns classifications such as positive, neutral, or negative, together with confidence scores. Opinion mining extends this analysis by associating expressed sentiment with particular aspects, attributes, products, or services mentioned in the text. This allows the application to determine not only the overall attitude toward the brand but also which specific brand characteristics are receiving favorable or unfavorable reactions.

Azure Document Intelligence is intended primarily for extracting text, fields, tables, and structures from documents. Content Safety detects potentially harmful material rather than measuring consumer opinion. Azure Vision processes visual content and is not the primary service for sentiment classification of written social-media posts.

Azure Language can be accessed through REST APIs, client libraries, or supported containers. The application submits the social-media text and receives document-level and sentence-level sentiment results that can be aggregated into brand-perception metrics, dashboards, or alerts.

Study Guide alignment: Implement text analysis solutions — perform sentiment analysis and opinion mining, process unstructured text, and interpret confidence scores and sentiment classifications.

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

Custom Vision model export is supported for compact domains . The existing project uses the ordinary General domain, so the first action is to change the domain to General (compact) .

A compact domain uses a model architecture specifically designed to support local/edge inference and export. Changing the domain alone does not transform the already-trained iteration into a compact model. Therefore, the second action is to retrain the model so that a new iteration is generated using the compact-domain architecture.

After the compact iteration has been trained, export the model in an appropriate supported format. The resulting model package can then be deployed within the disconnected environment and perform inference without contacting the Azure-hosted Custom Vision prediction service.

Creating a new classification model is incorrect because the existing scenario is object detection. Changing the classification type is also unnecessary and would not address exportability.

The required sequence is therefore:

General → General (compact) → retrain → export

This is a standard Custom Vision edge-deployment workflow and directly supports environments with restricted or unavailable internet connectivity.

Study Guide alignment: implement solutions that identify objects/components and deploy computer-vision functionality appropriate to the target environment .

The correct answer is B. Add a reflection pass before the responses are returned . A reflection pass is an application-orchestration step in which the generated summary is reviewed before final delivery, typically by asking the model or an evaluator step to check whether the answer covers the retrieved policy evidence and to revise the response when important details are missing. This directly addresses response completeness in application logic before the response is returned. The Microsoft Learn study guide explicitly includes Implement model reflection and Apply prompt engineering techniques to improve responses under optimization and operationalization of generative AI solutions.

This is also consistent with Microsoft Foundry agentic-loop guidance, which identifies reflection and planning cycles as patterns for multi-step reasoning in production agent systems. Completeness is a response-quality property: Azure AI evaluation defines completeness as whether a response contains all necessary and relevant information with respect to ground truth.

Option C is not correct because the scenario already says the agent generates summaries from retrieved policy documents, which is already a grounded retrieval pattern. Option A mainly reduces randomness, not missing content. Option D improves delivery experience, not answer completeness. Reference topics: model reflection, prompt engineering, agentic loops, response evaluation, and grounded generative AI solutions.

Because the existing classifier is an Azure AI Custom Vision project, the lowest-effort workflow is performed directly in the Custom Vision portal , not Azure Machine Learning studio or Vision Studio. Open the existing project so that the new product classes can be incorporated into its current training dataset.

Next, upload representative sample images of each new product. The images should include sufficient variation in viewing angle, lighting, background, size, and presentation to support reliable classification. Label each image with the appropriate new product tag. Custom Vision uses these tags as the target classes during supervised training. Microsoft’s portal workflow explicitly supports uploading and manually tagging images without requiring application code.

Retrain the classifier after the labeled images have been added. Training creates a new iteration based on all current training images and tags, including the newly introduced product categories.

Finally, publish the trained iteration to the prediction resource. Publishing makes that iteration available through the Custom Vision Prediction API used by the application.

Study Guide alignment: create, train, evaluate, retrain, publish, and consume custom image-classification models .

Extract the text: Azure Vision in Foundry Tools

Identify well-known locations: Azure Language in Foundry Tools

Azure Vision in Foundry Tools is the appropriate service for extracting text from road signs. Its pretrained Optical Character Recognition capability detects printed or handwritten text in images and returns recognized words, lines, and their coordinates. Microsoft specifically identifies street signs as an applicable OCR scenario. Because the OCR model is already trained, it avoids the data labeling and custom-model training that would increase development effort.

After OCR produces the text, Azure Language in Foundry Tools can analyze it through entity linking . Entity linking recognizes and disambiguates well-known entities and connects them to entries in a recognized knowledge base. This is suitable for determining that text on a sign represents a known geographic location rather than merely returning an unclassified text string. Named entity recognition can also classify extracted entities under categories such as Location, but entity linking provides the stronger interpretation implied by “well-known locations.”

Azure Document Intelligence is optimized for structured and text-heavy business documents, while Azure AI Search provides indexing and retrieval rather than primary OCR and linguistic analysis.

Study Guide alignment: implement OCR-based ingestion and use Foundry Tools to extract entities from text .

The correct answer is B. Use real-time speech to text to process streaming audio input . The scenario requires live transcription from a continuous telephony stream, with transcript text appearing within a few seconds while the call is still in progress. Azure Speech in Foundry Tools real-time speech recognition is specifically intended for immediate transcription scenarios such as call center assistance, dictation, and live meeting captioning. Microsoft’s Speech guidance describes real-time speech to text as processing audio input and returning transcriptions in real time, which matches the supervisor monitoring requirement.

Batch transcription is inappropriate because it processes stored audio files after recording, not an active live stream. Speech translation is used when the primary goal is translating speech into another language, not simply producing live same-language call transcripts. Text to speech performs the reverse operation by generating spoken audio from text and does not transcribe inbound calls. Real-time speech to text provides the low-latency streaming recognition path required for live operational monitoring. Reference topics: Azure Speech in Foundry Tools, real-time speech recognition, streaming audio input, call center transcription, and live captions.

Configure a multi-file task in pro mode . At analysis time, provide the invoice and its related purchase order as the input documents. During analyzer creation, add the static vendor contracts as reference data . Pro mode can reason across multiple input documents and use reference documents as contextual knowledge for validation, enrichment, and discrepancy detection. Microsoft specifically documents the scenario of supplying an invoice and purchase order as inputs while using contract files as reference data to identify inconsistencies.

Define a field schema containing the required invoice information and generated discrepancy findings. Building the task creates an analyzer ID that applications invoke through one analyzer API endpoint, producing a unified structured result governed by that schema.

Standard mode is optimized for straightforward processing of individual files. It does not support cross-file analysis, reference-dataset integration, or the multi-step reasoning required to compare invoices, purchase orders, and contractual conditions. A “multi-file task in standard mode” is therefore not a supported configuration. Confidence scores do not address document comparison and, notably, are not available in pro mode.

Study Guide alignment: configure Content Understanding analyzers, design structured schemas, process multiple documents, integrate reference data, and implement document validation workflows .

The correct answer is C because the requirement is a localized image edit: remove only the competitor logo while preserving the rest of the already generated image. Azure OpenAI image editing is designed for modifying existing images based on a text instruction, rather than regenerating the entire image from scratch. Microsoft’s Azure OpenAI image guidance states that the Image Edit API modifies existing images and requires an input image as part of the request. In a mask-based inpainting workflow, the mask identifies the exact region to change, allowing the model to replace only the logo area while retaining surrounding background, composition, lighting, and product content.

Increasing prompt guidance strength would affect adherence during generation, but it would not safely remove a specific logo from a completed image. Modifying the original prompt and regenerating may create a different image and does not guarantee preservation of the approved visual content. Rerunning with a different random seed also changes the image unpredictably and may introduce new brand or legal issues. Mask-based inpainting is the minimal-change remediation method for post-generation brand cleanup. Reference topics: Azure OpenAI image editing, inpainting, mask-guided edits, image generation governance, and computer vision solutions.

The correct built-in skills are Azure OpenAI Embedding and Text Split. The case study requires an indexing pipeline that enables semantic and vector search over the product sheets stored in Azure Blob Storage, so Agent1 can retrieve relevant product information for natural language customer questions. For a RAG pipeline, long PDF content must first be broken into retrievable chunks, and each chunk must then be vectorized for semantic similarity retrieval.

Microsoft’s Azure AI Search integrated vectorization guidance states that you create a skillset that calls the Text Split skill for chunking and the Azure OpenAI Embedding skill to vectorize the chunks. The Text Split skill breaks text into chunks and provides positional metadata, making it suitable when downstream embedding skills have input-length limits. The Azure OpenAI Embedding skill connects to an embedding model deployed in Azure OpenAI or a Microsoft Foundry project and generates embeddings during indexing.

Language Detection, Entity Recognition, and key phrase extraction can enrich text, but they do not create vector embeddings. Merge is useful for combining OCR text with document text, but it does not satisfy the core vector-search requirement. Reference topics: Azure AI Search skillsets, Text Split skill, Azure OpenAI Embedding skill, integrated vectorization, and RAG indexing.

credential = DefaultAzureCredential()

agent = project_client.agents.get(agent_name=myAgent)

The correct authentication option is DefaultAzureCredential() because the case study states that API keys must not be used to access Foundry-deployed models and that Contoso developers must authenticate to Microsoft Foundry resources by using Microsoft Entra authentication. It also states that access to Project1 must be assigned to Agent1Dev Team by using the security group SC_Agent1_Dev . Microsoft Foundry authentication guidance recommends Microsoft Entra ID for production workloads because it supports least-privilege RBAC, per-principal auditing, and keyless authentication. AzureKeyCredential() would violate the no-API-key requirement, and None would not provide a valid credential.

The correct agent operation is get because the task is to access an existing agent named Agent1, not create a new version or retrieve a specific published version. Microsoft Foundry SDK examples show AIProjectClient created with DefaultAzureCredential() and then using project agent operations to create, retrieve, or interact with agents by name. To meet the compliance requirement, the group SC_Agent1_Dev must also be granted the appropriate project-scoped Foundry role, such as Foundry User, for Project1. Reference topics: Microsoft Entra authentication, Foundry RBAC, AIProjectClient, and project agent access.

Deployment type: Standard

Version update policy: Once the current version expires

The correct deployment type is Standard . The case study specifies that Project1 is deployed in an EU Azure region and that model-processed data must remain within the EU. It also requires scalable, high-throughput generative AI workloads that dynamically handle variable customer support traffic without reserved throughput capacity. In Microsoft Foundry Models, Standard is a pay-per-token deployment type that processes data in a single Azure region, while Global Standard can process requests across regions and Global Provisioned uses reserved provisioned throughput. Microsoft’s deployment-type guidance identifies Standard as single-region, pay-per-token, whereas Global Provisioned is cross-region with reserved capacity.

The correct version update policy is Once the current version expires . This keeps Agent1 on the selected model version during its supported lifecycle, which supports stable and consistent responses, but still preserves continuity by automatically moving to a supported replacement when the current version is retired. Microsoft’s model versioning guidance states that this policy updates only when the current model version expires, while upgrading when a new default is available changes the deployment sooner and opting out can cause the deployment to stop working after retirement. Reference topics: deployment types, regional data processing, model versioning, throughput capacity, and stable production deployments.

Call client.videos.create() to submit the prompt and start a video-generation job against the model deployment. Video generation is asynchronous: the initial response represents a job and contains properties such as its unique ID and current status rather than immediately returning the completed video. Microsoft’s official Python example creates the job by using client.videos.create(model=..., prompt=...).

The application must then poll the job. After waiting between requests, client.videos.retrieve(video.id) obtains the latest state of that specific job. Polling continues until the status becomes completed, failed, or cancelled. list would enumerate jobs rather than refresh the selected job, while download_content is used only after successful completion to retrieve the generated video file.

To comply with Contoso’s prohibition on API keys, the client should use a Microsoft Entra bearer-token provider. The OpenAI Python client accepts this provider through its api_key constructor parameter even though no static API key is used.

Study Guide alignment: design and implement video-generation solutions and integrate generative workflows through Foundry SDKs.

Agent1 requires memory because the business requirement explicitly states that the agent must retain conversation context and recall relevant information during future interactions. Microsoft Foundry Agent Service memory provides persistent knowledge that can be retained and retrieved across separate conversations, sessions, devices, and workflows. It can store meaningful information extracted from prior interactions, including user preferences, relevant facts, and summarized context, and then retrieve that information when personalizing later responses.

Short-term memory maintains context within the current conversation. Long-term memory supports continuity across future sessions, which directly addresses Contoso’s requirement. Memory scopes should be associated with individual user identities so that one customer’s stored information cannot be retrieved during another customer’s interaction.

Guardrails enforce safety or behavioral restrictions but do not provide conversation recall. Tools enable the agent to access external systems, while instructions define the agent’s role and operating boundaries. Neither tools nor instructions persist user-specific information across sessions.

The AI-103 Study Guide maps this requirement to choose appropriate memory, tool, and knowledge integration services and build agents that integrate retrieval, function calling, and conversation memory .

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

The correct recommendation is Azure Content Understanding in Foundry Tools . The case study states that Contoso’s finance department must manually review vendor invoices to verify that invoice details match vendor contract terms, and that the invoices contain tables, logos, and varied layouts that make consistent processing difficult. It also states that the planned solution must evaluate both the visual layout and textual content of the invoices.

Azure Content Understanding is designed for this type of multimodal document-processing workload. Microsoft describes Content Understanding as a Foundry Tool that processes unstructured and multimodal content, including documents and images, and transforms it into structured output for AI applications. It can use document analyzers to extract text, layout, tables, fields, and relationships from diverse document types.

Chat completions alone would not reliably extract structured invoice fields from complex layouts. Azure Document Intelligence can extract OCR, layout, and tables, but Content Understanding is the better end-to-end Foundry capability for combining visual and textual understanding with structured extraction for downstream verification. Image Analysis focuses on image-level visual features and is insufficient for invoice field and table review. Reference topics: Content Understanding, document analyzers, multimodal extraction, invoice processing, tables, layout, and structured JSON output.

The correct answer is B. Modify the system message instructions . The case study states that Agent1 answers general questions about Contoso products and that a business requirement is for Agent1 to answer questions only about products sold by Contoso. This requirement defines the agent’s allowed domain and refusal boundary, so it must be expressed in the agent’s system-level instructions. Microsoft Foundry guidance states that system messages steer Azure OpenAI chat model behavior and are used to define the assistant’s role, boundaries, output format, and safety or quality constraints.

The system message should instruct Agent1 to answer only Contoso-product questions, use Contoso product documentation when available, and decline questions about non-Contoso products. This directly enforces the intended business scope at the highest instruction level. Few-shot examples can reinforce desired behavior but are not the primary control for defining mandatory operating boundaries. Top-p sampling and temperature are decoding controls; they influence randomness and diversity, not whether the agent restricts answers to a specific product domain. Increasing temperature would likely reduce consistency. Reference topics: Microsoft Foundry agent instructions, system message design, prompt engineering, response boundaries, and grounded generative AI behavior.

The correct answer is B. Personally Identifiable Information (PII) Detection . The case study states that Agent1 must never reveal customer information , even if a document containing customer data is added accidentally to the product sheet repository in storage1. This is a privacy and compliance control requirement, so the appropriate capability is PII Detection.

Azure Language PII Detection is a Foundry Tools capability that identifies, classifies, and redacts sensitive information across text, conversations, and native documents. Microsoft states that PII Detection can be used to implement privacy controls, reduce sensitive data exposure, and support compliance requirements. In this scenario, PII Detection should be applied to retrieved product-sheet content and generated responses so customer names, contact details, identifiers, and other sensitive values are not exposed to users.

Prompt Shields are important for a separate requirement: protecting Agent1 from malicious instructions hidden in documents or embedded text. Microsoft describes Prompt Shields for documents as protection against hidden instructions embedded in external content. However, the option that directly satisfies the requirement to prevent disclosure of customer information is PII Detection. Self-harm and violence filters address harmful-content categories, not privacy leakage.

The correct answer is B. Modify the system message instructions . The case study states that Agent1 answers general questions about Contoso products and that the business requirement is for Agent1 to answer questions only about the products sold by Contoso . This is a behavioral boundary for the agent, so it should be implemented in the highest-priority instructions that define the agent’s role, allowed scope, and refusal behavior.

Microsoft Foundry guidance states that a system message is used to steer model behavior, define the assistant’s role and boundaries, and add safety or quality constraints for the scenario. The system message should instruct Agent1 to answer only when the question concerns Contoso products, use the configured Contoso product documentation as grounding, and politely refuse or redirect questions about non-Contoso products.

Top-p sampling and temperature control randomness, not business-domain scope. Increasing temperature would make responses less deterministic. Few-shot examples can support desired behavior, but examples alone are weaker than explicit system-level instructions for defining operating boundaries. Reference topics: system message design, prompt engineering, agent instructions, response constraints, and grounded generative AI behavior.

A Retrieval Augmented Generation evaluator is required because Agent1 will retrieve product-sheet content and use it as grounding context when generating answers. Foundry RAG evaluators assess both retrieval quality and the quality of the final grounded response. This directly supports Contoso’s requirement that answers based on the product sheets be relevant, complete, and accurate.

The appropriate evaluation suite can include groundedness , which determines whether the answer is supported by the retrieved product information without fabrication; relevance , which evaluates whether the answer accurately and completely addresses the customer’s question; and response completeness , which measures whether critical expected information is missing. Retrieval evaluators can additionally determine whether the indexing and search pipeline returned context chunks relevant to the query.

A coherence evaluator assesses readability and logical organization but does not establish that claims are supported by storage1. A custom guardrail controls or blocks behavior rather than measuring RAG quality. Fine-tuning changes model behavior and does not evaluate whether retrieved documentation was selected and used correctly.

The AI-103 Study Guide includes evaluating models and applications for fabrications, relevance, quality, and safety , as well as monitoring grounding quality and retrieval relevance.

The correct recommendation is Azure AI Search. The case study states that the product detail sheets are stored as PDFs in storage1, and that Agent1 must be enabled to retrieve and use detailed product information from those sheets. It also specifies that the indexing pipeline must enable semantic and vector search, and that Agent1 must answer natural language questions about product details by using the product sheet information. Azure AI Search is the Azure service designed to ingest content from sources such as Azure Blob Storage, create searchable indexes, and support keyword, semantic, hybrid, and vector retrieval for Retrieval Augmented Generation (RAG) solutions.

Microsoft’s Azure AI Search guidance states that integrated vectorization can chunk content and generate embeddings during indexing, enabling vector search over source documents. It also states that Azure AI Search supports text and vector queries and can improve raw content for search-related scenarios through enrichment pipelines. Azure Translator is unrelated to retrieval. Document Intelligence can extract document structure, but it is not the retrieval index for Agent1. Grounding with Bing Search retrieves public web content, not Contoso’s private PDFs in storage1. Reference topics: Azure AI Search, RAG, semantic search, vector search, Azure Blob Storage indexing, and agent grounding.