AI-103 Timed Hard Questions — 25 Difficult Questions

Question 1

Your enterprise application uses Custom Vision. Due to compliance policies, you must ensure that your training data and model endpoints reside within a specific Azure region with strict data residency guarantees. Which resource configuration should you use?

15 minute time limit — choose an answer to begin.

Accepted Answer

Create a Custom Vision training and prediction resource in the targeted Azure region.. Custom Vision resources must be created in an Azure region that supports Custom Vision, and data remains within that region. Using a Cognitive Services multi-service resource or regional Custom Vision resource ensures regional residency.

Answer

Use the Global resource endpoint for Custom Vision.

Answer

Use Azure AI Vision multi-service account with global routing.

Answer

Store images in Azure Blob Storage with geo-replication.

Question 2

You are designing a multimodal solution that combines Azure AI Vision dense captions with Azure OpenAI GPT-4 Vision. You want to extract precise bounding box coordinates for multiple objects in complex scenes and pass them as structured context. Which API feature should you invoke?

Accepted Answer

features=denseCaptions. Dense captions return a set of regional bounding boxes and corresponding descriptive text snippets.

Answer

features=objects

Answer

features=smartCrops

Answer

features=tags

Question 3

You are using Document Intelligence Studio to build a custom extraction model. You need to verify the model quality before deploying. Which metric is most critical?

Accepted Answer

F1 Score. The F1 score provides a balance between precision and recall, serving as the primary indicator of model quality.

Answer

Recall only

Answer

Precision only

Answer

Model ID

Question 4

You are deploying a Custom Vision object detection model as a Docker container on an Azure Kubernetes Service (AKS) cluster. During load testing, you notice high memory consumption and container restarts. Which configuration setting in the deployment manifest should you review to ensure stability?

Accepted Answer

Kubernetes resource requests and limits for memory and CPU.. Custom Vision container deployments require proper CPU and memory limits/requests configured in the Kubernetes pod spec to prevent Out-Of-Memory (OOM) kills during heavy inference workloads.

Answer

Billing endpoint URL syntax.

Answer

Azure Blob Storage connection string persistence.

Answer

Custom Vision iteration publishing name environment variable.

Question 5

You are training a Custom Vision object detection model. You have 1,000 images, but only 200 of them contain the rare object you want to detect. The other 800 images are background images without the object. What impact will this class imbalance have on training, and how should you address it?

Accepted Answer

Ensure that you have tagged at least 15 bounding boxes per tag; having background images is beneficial for reducing false positives.. While having background (negative) images helps reduce false positives, an extreme imbalance or lack of positive bounding box instances per image can hinder learning. Ensuring sufficient positive instances (at least 15 per tag, preferably more for robust object detection) is essential.

Answer

Reduce training iteration time to 1 minute.

Answer

Convert the project to multilabel classification.

Answer

Delete all 800 background images immediately, as they will cause training to fail.

Question 6

You are building an agentic workflow in Microsoft Foundry using Semantic Kernel. The agent needs to call an external enterprise API to retrieve customer billing data. The API requires OAuth 2.0 user-delegated tokens. How should you configure the plugin registration to securely pass the user context?

Accepted Answer

Configure the Semantic Kernel plugin using an OpenApiFunctionDefinition that accepts a dynamic context parameter mapped to the incoming authorization header.. User-delegated authentication in Semantic Kernel plugins requires leveraging the active turn context to propagate the user's OAuth 2.0 token through the kernel invocation pipeline.

Answer

Disable authentication on the external API and rely solely on network-level Azure VNet integration.

Answer

Store the user's password in Azure Key Vault and have the plugin retrieve it on every tool call.

Answer

Hardcode the service account client secret inside the plugin configuration settings.

Question 7

You are implementing an agentic workflow where an agent needs to search a vector database, inspect the retrieved results, and if the results are insufficient, reformulate the query and search again before answering. What agent design pattern are you implementing?

Accepted Answer

Self-Correction / Reflection pattern. Self-Correction / Reflection pattern allows agents to evaluate their intermediate outputs, detect gaps or failures, and take corrective action (like query re-writing) before finalizing their response.

Answer

Static zero-shot generation pattern

Answer

Direct REST API proxy pattern

Answer

Database indexing batch pattern

Question 8

Your enterprise requires strict security and governance over your Azure AI Vision resources. Which THREE security configurations should you implement? (Choose three.)

Accepted Answer

Configure Azure Private Endpoints to restrict access to your Virtual Network.. Enterprise security for Azure AI Services includes Private Endpoints, disabling local authentication in favor of Microsoft Entra ID (RBAC), and Key Vault integration for secrets.

Answer

Store API keys in plain text configuration files on public GitHub repositories.

Answer

Disable TLS encryption to improve API response latency.

Question 9

You are integrating Azure AI Vision into a serverless Azure Functions architecture. A batch of 500 high-resolution images needs to be processed. To prevent function timeouts and optimize throughput, how should you architect the function triggers?

Accepted Answer

Use a Blob Storage trigger to enqueue image messages onto an Azure Storage Queue, and configure Queue-triggered functions to process image analysis asynchronously.. Using Azure Blob Storage triggers combined with Azure Queue Storage and queue-triggered functions allows decoupled, scalable, asynchronous processing without function timeout bottlenecks.

Answer

Use Azure Logic Apps with a synchronous batch loop.

Answer

Trigger a single synchronous Azure Function that loops through all 500 images sequentially within a single HTTP request.

Answer

Configure an Azure Event Grid subscription to directly invoke Azure AI Vision REST endpoints without compute logic.

Question 10

You are optimizing an application that calls Azure AI Vision Read API. Which THREE techniques help reduce latency and improve overall throughput? (Choose three.)

Accepted Answer

Resize extremely large images to optimal dimensions while maintaining readability.. Throughput and latency improvements include resizing excessively large images before upload, using asynchronous polling efficiently without tight busy-wait loops, and co-locating resources.

Answer

Send uncompressed RAW bitmap data to bypass image decoding overhead.

Answer

Disable SSL/TLS verification to speed up TCP handshakes.

Question 11

Your enterprise application uses Azure AI Vision. Security policies require that all API keys are rotated every 30 days without causing application downtime. How should you manage the key rotation?

Accepted Answer

Switch client applications to Key 2, regenerate Key 1 in the Azure portal, switch clients to Key 1, and then regenerate Key 2.. Azure AI Services resources provide two keys (Key 1 and Key 2). To rotate keys without downtime, update applications to use Key 2, regenerate Key 1, update applications to use Key 1, and then regenerate Key 2.

Answer

Restart the Azure App Service hosting the client application.

Answer

Delete the Azure AI service resource and create a new one every 30 days.

Answer

Disable authentication on the resource during the rotation window.

Question 12

You are designing an agentic workflow in Azure AI Agent Service where multiple specialized agents need to collaborate to solve complex, multi-step financial analysis tasks. One agent manages the workflow state and decides which sub-agent to invoke next. What architectural pattern is best suited for this implementation?

Accepted Answer

Orchestrator-workers pattern, where a central agent dynamically assigns sub-tasks to specialized worker agents based on intermediate results.. The orchestrator-workers (or router) pattern uses a central orchestrator agent to manage state, evaluate progress, and delegate tasks to specialized worker agents.

Answer

Single monolithic prompt pattern, where all instructions are combined into one massive system prompt without multi-agent handoffs.

Answer

Map-reduce batch processing pattern, where all sub-agents execute independently in parallel without communication.

Answer

Static sequential pipeline, where agents execute strictly in a predefined linear order without conditional branching.

Question 13

Your team is implementing 'Human-in-the-loop' (HITL) for an agent that performs account updates. Which mechanism should you use to pause the agent and wait for approval?

Accepted Answer

Use a Prompt Flow 'Pause' node or custom state management logic.. In agentic workflows, an interruption or 'callback' mechanism allows the system to pause execution until human input is received.

Answer

Set a long timeout in the API call.

Answer

Increase the model's 'stop_sequences'.

Answer

Use an infinite loop in Python.

Question 14

You are designing a RAG solution in Azure AI Foundry where sensitive enterprise documents must be retrieved. Users should only see search results corresponding to their security clearance level. How should you implement document-level access control in Azure AI Search?

Accepted Answer

Store user security groups in document metadata and apply OData filter expressions during search queries. Security trimming in Azure AI Search is implemented by adding user security identifiers (IDs or group memberships) as metadata fields on each document and applying OData filter expressions at query time.

Answer

Configure Microsoft Entra ID authentication on the Azure OpenAI endpoint

Answer

Encrypt the search index using customer-managed keys (CMK) in Azure Key Vault

Answer

Use Azure AI Content Safety to redact sensitive keywords before indexing

Question 15

You are implementing a disaster recovery strategy for a critical image processing application that relies on Azure AI Vision. If the primary Azure region experiences an outage, how should you configure failover?

Accepted Answer

Deploy a secondary Azure AI Vision resource in a paired Azure region and configure your application to fail over to the secondary endpoint if errors occur.. Deploying a secondary Azure AI Vision resource in a paired region and updating the application configuration (or using Azure Front Door / Traffic Manager) enables regional failover.

Answer

Use geo-replication on the Azure Blob Storage account to automatically fail over the AI service endpoints.

Answer

Configure Azure Site Recovery to replicate the Azure AI Vision service backend VMs.

Answer

Rely on automatic synchronous replication of Custom Vision models across all global Azure datacenters without application changes.

Question 16

You are implementing indirect prompt injection defenses in a RAG application where documents are ingested from public web pages. Which THREE mitigation strategies should you deploy? (Choose THREE)

Accepted Answer

Use strict structural delimiters (like XML tags) to isolate retrieved context from system instructions. Prompt shields in Content Safety, strict prompt delimiters separating data from instructions, and content validation filters protect against indirect prompt injection.

Answer

Trust all incoming web page text unconditionally without inspection

Answer

Disable all content moderation filters to maximize processing speed

Question 17

You are orchestrating a multi-agent system in Microsoft Foundry where Agent A generates code and Agent B reviews it. Agent B frequently gets stuck in a loop trying to optimize minor syntax formatting. How should you design the agentic workflow to prevent this infinite refinement loop?

Accepted Answer

Configure a maximum turn counter and state transition condition in the orchestrator workflow to terminate the loop after a set threshold.. Establishing a strict iteration limit or state machine check within the orchestrator prevents agents from looping indefinitely.

Answer

Disable function calling for Agent B so it can only output text responses.

Answer

Increase the model's presence penalty so Agent B is forced to generate novel review comments.

Answer

Use a smaller model for Agent B to reduce the quality of its reviews.

Question 18

You are optimizing an agentic workflow in Azure AI Foundry. The agent needs to solve complex reasoning problems by breaking them down into distinct sub-tasks, executing them iteratively, and reflecting on intermediate results. Which THREE architectural frameworks or patterns support this behavior? (Choose three.)

Accepted Answer

Reflection pattern where the agent critiques its own intermediate outputs and refines them.. Chain-of-thought, ReAct (Reason+Act), and plan-and-solve are foundational agentic reasoning patterns.

Answer

Static hardcoded IF-ELSE decision trees without model inference.

Answer

Single-turn stateless completion calls with temperature set to zero.

Question 19

You are designing a multi-agent solution in Azure AI Foundry where agents collaborate asynchronously. One agent produces an intermediate JSON payload that must be strictly validated before being consumed by the next agent. What feature should you implement?

Accepted Answer

Structured Outputs with strict JSON Schema enforcement. Structured Outputs (JSON mode with strict schema enforcement) ensures that model completions conform precisely to a defined JSON Schema.

Answer

Model temperature set to 1.0

Answer

Azure API Management XML transformation policy

Answer

Azure AI Content Safety hate speech detection

Question 20

You are tuning the prompt for a generative AI application in Azure AI Foundry. The model frequently hallucinates when asked complex multi-part questions. You decide to use chain-of-thought (CoT) prompting. How should you structure the prompt to implement CoT effectively?

Accepted Answer

Instruct the model to think step-by-step and show its intermediate reasoning before providing the final answer. Chain-of-thought prompting involves guiding the model to articulate intermediate reasoning steps before arriving at the final conclusion.

Answer

Set the model seed parameter to a fixed integer value for reproducibility

Answer

Reduce the maximum completion token limit to 150 tokens

Answer

Instruct the model to output its response exclusively in JSON format with strict schema validation