I85PREP
⌘ K
AI-200 study materials
AI-200 study resources

AI-200 practice questions

One question at a time. Build your AI-200 understanding.

First 50 of 200 questions. Each one lists the choices, the correct answer, and an explanation. The checker at the top of this page saves progress in this browser when it is available.

Continue in the practice tool above to load more questions, 50 at a time.

Question 1Plan and manage an Azure AI solution

A development team wants to use vision, language, and speech capabilities from one application and wants to manage a single key and a single endpoint. Which resource should they create?

  1. An Azure AI Search instance with a shared index
  2. A separate single-service resource for each AI service
  3. An Azure Machine Learning workspace with attached compute
  4. A single multi-service Azure AI services resource Correct answer.

Explanation

A multi-service Azure AI services resource exposes one endpoint and one key set for multiple services, so credentials, billing, and monitoring are managed in one place. Single-service resources are preferred only when you need per-service isolation of networking, billing, or limits.

Question 2Plan and manage an Azure AI solution

Which credential approach should an application use when calling an Azure AI services resource so that no long-lived secret must be stored or rotated manually?

  1. Microsoft Entra ID authentication with a managed identity Correct answer.
  2. A connection string stored in an environment variable
  3. A shared access signature generated for each request
  4. Embed the resource's subscription key in the application configuration file

Explanation

Microsoft Entra ID authentication with a managed identity lets the platform issue and rotate tokens automatically, so no secret is stored in code or configuration. Subscription keys are long-lived secrets that must be protected, rotated, and audited by the team.

Question 3Plan and manage an Azure AI solution

You must ensure that requests from an application in a virtual network to an Azure AI services resource never traverse the public internet. What should you configure?

  1. An IP firewall rule that allows the virtual network's outbound address
  2. A private endpoint connected to the Azure AI services resource Correct answer.
  3. A service endpoint enabled on the application subnet
  4. An Azure Firewall policy applied to the virtual network

Explanation

A private endpoint assigns a private IP address from the virtual network to the resource, so all traffic to it stays on the private network. Service endpoints keep traffic on the Azure backbone but the resource still exposes a public endpoint that remains reachable.

Question 4Plan and manage an Azure AI solution

You must deploy an identical Azure AI services resource to five regions in a repeatable, version-controlled way. Which approach should you use?

  1. Submit REST API calls to each regional endpoint and record the payloads
  2. Deploy a parameterized Bicep file that loops over the target regions Correct answer.
  3. Run Azure CLI commands manually in each region from an engineer's workstation
  4. Create the resource once in the portal and export the generated template for later use

Explanation

Infrastructure as code with parameters and loops produces the same result on every deployment and is stored in version control. Portal exports are one-time snapshots that drift from the desired state as soon as anything changes.

Question 5Plan and manage an Azure AI solution

You need to retain request logs from an Azure AI services resource for 90 days and query them with KQL. Where should the diagnostic setting send the logs?

  1. The resource's built-in metrics explorer
  2. An Azure Event Hubs namespace
  3. A Log Analytics workspace Correct answer.
  4. An Azure Storage account configured for blob storage

Explanation

A Log Analytics workspace stores diagnostic logs with a configurable retention period and supports KQL queries and workbooks. Storage and Event Hubs can receive the same logs but are not queried with KQL in place.

Question 6Plan and manage an Azure AI solution

An Azure AI Language resource handles a steady, high volume of transactions every day. Which pricing option lowers the effective per-transaction cost for this predictable workload?

  1. The pay-as-you-go standard tier
  2. Provisioned throughput units added to the resource
  3. A commitment tier configured for the resource Correct answer.
  4. The free (F0) tier

Explanation

A commitment tier bills a fixed amount for a defined level of usage, which reduces the unit price when consumption is steady and high. Pay-as-you-go pricing charges per transaction and costs more at sustained volume.

Question 7Plan and manage an Azure AI solution

An Azure OpenAI chat application must maintain consistent latency under peak load and avoid capacity-based throttling. Which deployment configuration should be used?

  1. A provisioned deployment with reserved throughput Correct answer.
  2. A global batch deployment for asynchronous processing
  3. A standard deployment with a high tokens-per-minute quota
  4. A standard deployment with content filtering disabled

Explanation

Provisioned deployments reserve model capacity for the deployment, so throughput and latency stay predictable and requests are not competing for shared capacity. Standard deployments consume shared capacity and can return 429 responses when demand spikes.

Question 8Plan and manage an Azure AI solution

An application receives HTTP 429 responses from an Azure AI services endpoint during traffic bursts. What should the client implement?

  1. Enable customer-managed keys
  2. Retry with exponential backoff and jitter Correct answer.
  3. Increase the model temperature setting
  4. Move the resource to the free tier

Explanation

A 429 response means the request was rate limited, so the client should honor the Retry-After interval and retry with exponential backoff plus jitter to avoid synchronized retry storms. The other settings do not change the request rate against the service.

Question 9Plan and manage an Azure AI solution

Before applying an ARM template deployment, you want to see which resources will be created, modified, or deleted. Which capability should you use?

  1. The what-if operation of the deployment Correct answer.
  2. Template validation with a linked template
  3. The activity log of the subscription
  4. The deployment history of the resource group

Explanation

The what-if operation compares the template with the current state of the target scope and reports creates, modifications, and deletions without changing anything. Validation only checks template syntax and does not describe resulting changes.

Question 10Plan and manage an Azure AI solution

An App Service application must read an Azure AI services key from Azure Key Vault without storing credentials in code or configuration. What should you configure?

  1. A system-assigned managed identity granted access to the vault Correct answer.
  2. The vault token in App Service app settings
  3. A shared access signature scoped to the Key Vault secret
  4. A service principal client secret that is rotated every 90 days by a runbook

Explanation

A system-assigned managed identity gives the App Service an identity in Microsoft Entra ID whose token authenticates to Key Vault, so no secret is stored anywhere. Access is granted through a Key Vault access policy or an RBAC role such as Key Vault Secrets User.

Question 11Plan and manage an Azure AI solution

Where should subscription keys for an Azure AI services resource be stored so that access is audited and keys can be rotated without redeploying the application?

  1. In the resource's diagnostic settings
  2. In the application's appsettings.json file
  3. In the source code repository as build variables
  4. In Azure Key Vault as secrets Correct answer.

Explanation

Key Vault stores secrets encrypted, records every access in audit logs, and supports versioning so a rotated key can be referenced without changing code. Configuration files and repositories expose keys to anyone who can read them.

Question 12Plan and manage an Azure AI solution

You must ensure data at rest for an Azure AI services resource is encrypted with a key that your organization controls and can revoke. What should you configure?

  1. Azure Disk Encryption on the host virtual machines
  2. Transport Layer Security with a custom certificate
  3. Customer-managed keys stored in Azure Key Vault Correct answer.
  4. A service-managed key with automatic rotation

Explanation

Customer-managed keys let you supply a Key Vault key that wraps the account encryption key, so disabling or revoking your key removes access to the data. Service-managed keys are controlled by Microsoft rather than by your organization.

Question 13Plan and manage an Azure AI solution

Which settings must be supplied when running an Azure AI services container in connected mode so that it reports usage to Azure?

  1. SubscriptionId, TenantId, and ClientSecret
  2. ConnectionString, ContainerName, and AccountKey
  3. Endpoint, Region, and ResourceGroup
  4. Eula, Billing, and ApiKey Correct answer.

Explanation

Connected containers require Eula=accept to acknowledge the license, Billing set to the billing endpoint URI, and ApiKey set to the resource key so the container can send usage data. Without these values the container starts but refuses to serve requests.

Question 14Plan and manage an Azure AI solution

An organization wants to run an Azure AI services container on an isolated network with no outbound internet connectivity while remaining compliant with billing requirements. What should they do first?

  1. Disable telemetry on the connected container with an environment variable
  2. Use connected containers and cache billing responses locally on the host
  3. Request approval for disconnected containers and use a commitment tier Correct answer.
  4. Mirror the Azure billing endpoint inside the isolated network

Explanation

Disconnected containers are designed to run without outbound connectivity, but they require Microsoft approval and a commitment-tier plan, and they forward billing information when connectivity becomes available. Connected containers need outbound access to the billing endpoint and will not run indefinitely offline.

Question 15Plan and manage an Azure AI solution

Contoso must keep all customer data within Canada. What should you do when creating the Azure AI services resource?

  1. Enable customer-managed keys in a Canadian key vault
  2. Create the resource in a Canada region Correct answer.
  3. Enable diagnostic logging to a Canadian storage account
  4. Configure a private endpoint in a Canadian virtual network

Explanation

The region chosen at creation determines where the service stores and processes data, so the resource itself must be created in a Canadian region. Key storage, private endpoints, and log destinations do not change where inference data is processed.

Question 16Plan and manage an Azure AI solution

A workload is throttled because it exceeds the default transactions-per-second limit for its pricing tier. What is the appropriate first step?

  1. Delete and recreate the resource in another subscription
  2. Move the resource to the free tier to reset the limit
  3. Submit an Azure support request to raise the quota Correct answer.
  4. Reduce the application's retry count to lower load

Explanation

Quotas such as transactions per second are tied to the pricing tier and can be raised through a support request for justified workloads. Recreating the resource does not change the default limits for its tier.

Question 17Plan and manage an Azure AI solution

You need to notify the operations team whenever the number of failed calls to an Azure AI services resource exceeds a threshold. What should you create?

  1. An Azure Policy assignment with a deny effect
  2. A diagnostic setting that writes to an event hub
  3. An Application Insights availability test
  4. An Azure Monitor metric alert rule with an action group Correct answer.

Explanation

A metric alert rule evaluates a platform metric such as failed calls against a threshold and triggers an action group that sends notifications. Diagnostic settings route data to destinations but do not evaluate conditions or notify anyone.

Question 18Plan and manage an Azure AI solution

Several departments share one subscription that contains multiple Azure AI services resources. How can costs be attributed to each department?

  1. Use a separate resource group for each region
  2. Apply tags to resources and analyze cost by tag Correct answer.
  3. Enable diagnostic logging for each resource
  4. Place every resource in its own subscription

Explanation

Tags are carried into cost records, so cost analysis can group and filter spending by values such as department. Resource groups and subscriptions organize resources but do not by themselves attribute shared costs to a business unit.

Question 19Plan and manage an Azure AI solution

A developer must call an Azure AI services endpoint by using Microsoft Entra ID but must not be able to regenerate keys or change the resource configuration. Which role assignment meets the requirement?

  1. Owner on the resource group
  2. Cognitive Services User Correct answer.
  3. Cognitive Services Contributor
  4. Key Vault Secrets User

Explanation

The Cognitive Services User role permits calling the service with Entra ID credentials but does not include key management or resource configuration. Contributor and Owner both include key regeneration and configuration changes.

Question 20Plan and manage an Azure AI solution

Several teams need to share common connections and security settings while keeping their own work isolated in Azure AI Foundry. What should you create?

  1. A separate Azure AI services resource for each team
  2. One hub with a separate project for each team Correct answer.
  3. One hub per developer that shares a single storage account
  4. A single project shared by all teams with role assignments

Explanation

A hub provides shared connections, storage, and security configuration, while projects inside the hub isolate each team's assets and work. Multiple hubs or separate resources duplicate configuration and lose the shared governance benefit.

Question 21Plan and manage an Azure AI solution

An application uses an Azure OpenAI model and must block harmful outputs. Which feature should be configured?

  1. A private endpoint on the resource
  2. Content filters on the model deployment Correct answer.
  3. A customer-managed key for the resource
  4. Semantic caching on the deployment

Explanation

Content filters evaluate prompts and completions and block or annotate harmful categories according to the configured severity thresholds. Networking, encryption, and caching settings do not affect the safety of generated output.

Question 22Plan and manage an Azure AI solution

A CI/CD pipeline deploys Azure AI resources and needs the Azure OpenAI key during integration tests. How should the pipeline obtain the key securely?

  1. Use a Key Vault-backed variable group Correct answer.
  2. Print the key to the deployment log for reference
  3. Commit the key to the pipeline's YAML definition file
  4. Pass the key as a plain-text pipeline parameter

Explanation

A variable group linked to Key Vault retrieves the secret at run time and masks it in logs, so the key never lives in the repository. Committing or logging the key exposes it permanently in history.

Question 23Plan and manage an Azure AI solution

Calls to an Azure AI services resource are slow, and you need to correlate the client request with downstream dependency latency. Which service should you use?

  1. The Azure Monitor activity log
  2. Azure Service Health
  3. Azure Advisor
  4. Application Insights Correct answer.

Explanation

Application Insights collects requests, dependencies, and timings so a single transaction can be traced end to end and the slow hop identified. Service Health and Advisor report platform events and recommendations rather than per-request telemetry.

Question 24Plan and manage an Azure AI solution

A solution must remain available if the Azure region hosting its Azure AI services resource becomes unavailable. What should be deployed?

  1. A single resource with customer-managed keys in two regions
  2. The resource in two regions behind a global load balancer Correct answer.
  3. A single resource with a higher transactions-per-second quota
  4. A single resource with zone-redundant diagnostic settings

Explanation

Deploying the resource in two regions and routing traffic through a global load balancer or traffic manager allows requests to fail over when one region is down. A single resource, regardless of quota or key placement, remains bound to one region.

Question 25Plan and manage an Azure AI solution

An administrator must prevent creation of Azure AI services resources outside a list of approved regions. What should be used?

  1. An Azure Policy assignment with a denied-region effect Correct answer.
  2. A resource lock placed on the subscription
  3. A budget alert configured on the subscription
  4. A role assignment that removes the Contributor role

Explanation

Azure Policy evaluates resource properties such as location during deployment and can deny non-compliant creates. Locks, role changes, and budgets do not restrict which regions may host a resource.

Question 26Plan and manage an Azure AI solution

You need to create a single Azure AI services resource that exposes vision, language, and speech capabilities by using one endpoint and one set of subscription keys. Which resource should you create?

  1. An Azure AI Search resource
  2. A multi-service Azure AI services resource Correct answer.
  3. An Azure Machine Learning workspace
  4. One resource for each of the three services

Explanation

A multi-service resource (kind CognitiveServices) bundles several AI capabilities behind a single endpoint and key pair, which simplifies client configuration and consolidated billing. Creating individual service resources yields separate endpoints and keys for each service and does not meet the single-endpoint requirement.

Question 27Plan and manage an Azure AI solution

An Azure Functions app must call an Azure AI services resource without storing any secret in code or application settings. What should you configure?

  1. Create a shared access signature for the AI services resource
  2. Store the resource key in the function app's application settings
  3. Enable a managed identity and grant it a role on the AI resource Correct answer.
  4. Upload a client certificate and use it for each request

Explanation

A managed identity gives the function app an automatically rotated Microsoft Entra ID identity, and role-based access control on the AI services resource authorizes the calls. No key or secret is ever stored in configuration, which is exactly what the requirement demands.

Question 28Plan and manage an Azure AI solution

Request and response logs from an Azure AI services resource must be retained and queried by using KQL. Where should you send the diagnostic setting?

  1. An Azure event hub namespace
  2. An Azure Storage account (tables)
  3. The resource's activity log
  4. A Log Analytics workspace Correct answer.

Explanation

Diagnostic settings can route resource logs to a Log Analytics workspace, where they are queried with KQL and retained according to the workspace retention policy. Storage accounts and event hubs can receive the same logs but do not provide KQL querying.

Question 29Plan and manage an Azure AI solution

You must configure a new client with the endpoint URL and a subscription key for an existing Azure AI services resource. Where do you obtain these values in the Azure portal?

  1. The Identity blade of the resource
  2. Access control (IAM) for the resource
  3. The Networking blade of the resource
  4. The Keys and Endpoint blade of the resource Correct answer.

Explanation

The Keys and Endpoint page exposes the resource endpoint and two independently regenerable subscription keys. The IAM, Identity, and Networking blades control authorization, managed identity, and connectivity respectively, and do not display credentials.

Question 30Plan and manage an Azure AI solution

Clients must call an Azure AI services resource by presenting a Microsoft Entra ID bearer token instead of a subscription key. What must the resource have to support token authentication?

  1. A system-assigned managed identity enabled on the resource
  2. A private endpoint connected to the resource
  3. A custom subdomain assigned to the resource Correct answer.
  4. A second subscription key generated for the resource

Explanation

Token authentication requires the resource to be addressed through a custom subdomain endpoint such as https://<name>.cognitiveservices.azure.com, because the shared regional endpoint cannot resolve the resource from the token. Enabling a managed identity or adding a private endpoint is unrelated to accepting Entra ID tokens.

Question 31Plan and manage an Azure AI solution

An Azure AI services resource must be reachable only from inside a virtual network, with no access from the public internet. Which combination should you implement?

  1. A private endpoint and public network access set to Disabled Correct answer.
  2. An IP firewall rule that permits only the application's outbound address
  3. Azure Front Door with a web application firewall in front of the resource
  4. A service endpoint on the application subnet and a matching virtual network rule

Explanation

A private endpoint places the resource in the virtual network with a private IP, and disabling public network access removes the internet-facing entry point entirely. Service endpoints and IP rules restrict the public endpoint rather than eliminating it.

Question 32Plan and manage an Azure AI solution

A data scientist must send inference requests to an Azure OpenAI deployment by using a Microsoft Entra ID identity only. Which built-in role provides the least privilege needed?

  1. Owner on the subscription
  2. Reader on the resource
  3. Cognitive Services Contributor
  4. Cognitive Services OpenAI User Correct answer.

Explanation

Cognitive Services OpenAI User grants data-plane access for inference such as chat completions and embeddings without permitting key or resource management. Contributor and Owner include management rights far beyond inference, which violates least privilege.

Question 33Plan and manage an Azure AI solution

A security policy requires that clients authenticate to an Azure OpenAI resource only with Microsoft Entra ID, and that subscription keys stop working. Which setting should you change on the resource?

  1. Enable a system-assigned managed identity
  2. Add an IP firewall rule for the client subnet
  3. Rotate both subscription keys
  4. Set local authentication to Disabled Correct answer.

Explanation

Disabling local authentication turns off key-based access so that every caller must present an Entra ID token authorized by role assignment. Rotating keys or adding network rules leaves the key credential usable.

Question 34Plan and manage an Azure AI solution

An administrator must ensure the production Azure AI services resource cannot be deleted, including by subscription owners, while normal configuration changes remain possible. What should be applied?

  1. A deny assignment on the resource group
  2. A ReadOnly lock on the resource
  3. Removing the Owner role from all administrators
  4. A CanNotDelete lock on the resource Correct answer.

Explanation

A CanNotDelete lock blocks deletion of the resource while still allowing read and write operations, which is precisely the required behavior. A ReadOnly lock over-restricts by also blocking configuration changes.

Question 35Plan and manage an Azure AI solution

Several applications call multiple Azure AI services resources. The security team wants one place to enforce rate limits, log requests, and swap backend keys without changing client code. What should you deploy?

  1. Azure Application Gateway with a web application firewall
  2. Azure API Management with the AI services as backends Correct answer.
  3. Azure Front Door with response caching enabled
  4. A separate private endpoint for each application

Explanation

API Management acts as a gateway that centralizes authentication, rate limiting, logging, and backend key substitution, so clients call a stable URL while the platform manages credentials. Application Gateway and Front Door provide web traffic protection or acceleration, not API policy management.

Question 36Plan and manage an Azure AI solution

An Azure OpenAI model deployment must remain on its current model version and must never be upgraded automatically. Which deployment setting should you configure?

  1. The model version upgrade policy Correct answer.
  2. The content filter configuration
  3. The diagnostic settings
  4. The tokens-per-minute quota

Explanation

Each deployment carries a model version update policy that controls whether it is upgraded automatically when a new default version is released or stays on the selected version until an administrator upgrades it. Content filters and quotas govern output safety and throughput, not versioning.

Question 37Plan and manage an Azure AI solution

Applications authenticate to an Azure AI services resource with Key1 only. You must rotate credentials with no downtime. Which sequence should you follow?

  1. Regenerate Key1 immediately, then update every application to use Key1
  2. Switch applications to Key2, regenerate Key1, then switch back to Key1 Correct answer.
  3. Regenerate both Key1 and Key2 at the same time, then update the applications
  4. Delete and recreate the resource so that new keys are issued

Explanation

Because two independent keys exist, you move clients to the key that will be kept, regenerate the one in use, and then migrate clients back. This keeps at least one valid credential available at every moment, which is the point of the dual-key design.

Question 38Plan and manage an Azure AI solution

Every new Azure AI services resource in a subscription must send its logs to a central Log Analytics workspace without relying on engineers to configure it. Which approach requires the least operational effort?

  1. A CanNotDelete lock and a cost-center tag on each resource
  2. An Azure Monitor alert rule created for each new resource
  3. A built-in DeployIfNotExists policy with a remediation identity Correct answer.
  4. A nightly script that exports activity logs to storage

Explanation

A policy with the DeployIfNotExists effect detects resources that lack the required diagnostic setting and creates it automatically through a remediation managed identity. This enforces the configuration at scale with no manual work, whereas the other options either only detect problems or apply no diagnostic configuration at all.

Question 39Plan and manage an Azure AI solution

Clients in a virtual network reach an Azure AI services resource through a private endpoint, but name resolution still returns the public IP address. What should you configure?

  1. DNS forwarding rules configured on the virtual network's DHCP options
  2. An A record added to the public cognitiveservices.azure.com zone
  3. A service endpoint enabled on the client subnet
  4. A linked private DNS zone for privatelink.cognitiveservices.azure.com Correct answer.

Explanation

Private endpoints require a private DNS zone whose name matches the privatelink subdomain of the service, linked to the client virtual network, so that the resource FQDN resolves to the private IP. Without that zone the public DNS answer continues to be returned.

Question 40Plan and manage an Azure AI solution

You create two Azure OpenAI deployments of the same model in the same region and split traffic between them to double throughput, but clients still receive HTTP 429 responses at the previous load level. What is the most likely cause?

  1. Both deployments draw from one quota allocation for that model in the region Correct answer.
  2. Content filters are rejecting the additional requests
  3. Requests must be sent from the REST API rather than an SDK
  4. The two deployments must use different model versions

Explanation

Tokens-per-minute quota is allocated per subscription, per region, per model, so multiple deployments of the same model in the same region divide a single allocation rather than multiplying it. Adding deployments therefore redistributes capacity instead of increasing it.

Question 41Implement generative AI solutions

You are designing a chat solution that uses the Azure OpenAI chat completions API. What is the primary purpose of the system message?

  1. To replay earlier assistant replies to the model
  2. To carry the end user's credentials for the request
  3. To set the assistant's persona and behavioral constraints Correct answer.
  4. To configure sampling parameters such as temperature

Explanation

The system message is the first message in the conversation and establishes role, tone, boundaries, and rules the assistant should follow. Credentials and sampling settings are handled outside the message content, through authentication and request parameters.

Question 42Implement generative AI solutions

A team wants more deterministic, repeatable completions from a chat model. Which parameter change best supports this goal?

  1. Enable streaming
  2. Increase max_tokens
  3. Set temperature closer to 0 Correct answer.
  4. Set top_p to 1.0

Explanation

Lowering temperature reduces the randomness of token selection, so the same prompt tends to produce similar output. Determinism is never absolute, but a temperature near zero is the standard way to make responses more consistent.

Question 43Implement generative AI solutions

Which task are text embeddings primarily suited for?

  1. Parsing model replies into JSON
  2. Generating photorealistic images from captions
  3. Measuring semantic similarity between documents Correct answer.
  4. Translating speech to text in real time

Explanation

Embeddings map text to vectors so that semantically similar content lands close together in vector space, which enables similarity search and clustering. They do not generate images, transcribe audio, or enforce output formatting.

Question 44Implement generative AI solutions

What is the main benefit of retrieval-augmented generation (RAG)?

  1. It eliminates the need for a system message
  2. It grounds answers in retrieved content at query time Correct answer.
  3. It removes all unsupported statements from outputs
  4. It permanently fine-tunes the model on private documents

Explanation

RAG retrieves relevant content from an index and injects it into the prompt so responses can reference current, organization-specific information without changing model weights. It reduces but does not guarantee elimination of unsupported statements.

Question 45Implement generative AI solutions

Azure OpenAI applies content filtering to prompts and completions. What does the default configuration do?

  1. It flags or blocks content across harm categories such as hate, violence, and self-harm Correct answer.
  2. It translates prompts into English before processing
  3. It rejects prompts longer than 1,000 characters
  4. It encrypts prompts before they reach the model

Explanation

The built-in content filter evaluates both input and output text against harm categories and applies severity thresholds that can block or annotate content. Length limits, encryption, and translation are unrelated to content filtering.

Question 46Implement generative AI solutions

In a chat completions request, which role represents the end user's input?

  1. system
  2. assistant
  3. user Correct answer.
  4. tool

Explanation

Messages with the user role carry input supplied by the person or calling application. The assistant role holds model-generated turns, and the tool role returns results from tool invocations back to the model.

Question 47Implement generative AI solutions

Which request parameter caps the length of the model's generated response?

  1. temperature
  2. max_tokens Correct answer.
  3. presence_penalty
  4. n

Explanation

max_tokens sets an upper bound on the number of tokens produced in the completion, so responses stop when the limit is reached. The other parameters influence randomness, repetition, and the number of returned choices.

Question 48Implement generative AI solutions

Before an application can call a model hosted in an Azure OpenAI resource, what must exist in that resource?

  1. A storage account with blob containers
  2. A logic app workflow
  3. A deployment that maps a model to a callable name Correct answer.
  4. A virtual network gateway

Explanation

Calls target a deployment name rather than the model name directly, so the model must first be deployed. Network gateways, storage accounts, and workflows are optional supporting resources, not prerequisites for inference calls.

Question 49Implement generative AI solutions

Your source material consists of long policy PDFs, and answers must cite the relevant passage. Which approach best prepares the content for retrieval?

  1. Replace each document with a single embedding of its summary
  2. Split documents into overlapping chunks and index each chunk Correct answer.
  3. Index each document as one large chunk to preserve context
  4. Store only the document titles as searchable fields

Explanation

Chunking to a size that fits the retrieval and prompt budget lets the system return focused passages that can be cited directly. One giant chunk per document dilutes relevance and risks exceeding the model's context window.

Question 50Implement generative AI solutions

A hybrid query in Azure AI Search combines which two techniques?

  1. Caching and sharding
  2. Image classification and OCR
  3. Keyword search and vector search Correct answer.
  4. Speech recognition and machine translation

Explanation

Hybrid search runs a keyword (full-text) query and a vector query in parallel and merges the results using reciprocal rank fusion. This captures exact term matches while also matching on semantic similarity.