AI-103 Practice Test 32 — 15 Questions

Question 1

You are processing health-related clinical notes using the Azure AI Language service. You need to extract medical concepts and link them to standardized medical vocabularies such as ICD-10-CM and SNOMED-CT. Which specific feature and endpoint task should you use?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Healthcare text analytics (Healthcare-specific NER and relation extraction). Azure AI Language's healthcare text analytics feature is specifically designed to extract medical entities and relations from clinical text, and it can link entities to standardized vocabularies like ICD-10-CM, SNOMED-CT, and UMLS. It uses a dedicated endpoint task (e.g., /entities/health/jobs) and returns entities with categories, subcategories, and ontology links. This is the only option that provides medical concept extraction and vocabulary linking out of the box.

Answer

Named Entity Recognition (NER) with standard prebuilt-entity models

Answer

Key phrase extraction with medical domain parameters

Answer

Custom Named Entity Recognition with a trained ICD-10 dictionary

Question 2

You are preparing an ARM template to deploy an Azure AI multi-service account. Which TWO properties are mandatory within the ARM template resource definition? Each correct answer presents part of the solution.

Accepted Answer

sku. Option B (sku) is mandatory because an Azure AI multi-service (Cognitive Services) account resource requires a pricing/tier definition, e.g. "sku": {"name": "S0"}, to specify the billing tier and feature set of the account. Option E (location) is mandatory because every Azure resource, including a Cognitive Services multi-service account, must declare the Azure region where it is deployed, e.g. "location": "eastus". Options A (VirtualMachineSize), C (OperatingSystemType), and D (KernelVersion) do not belong because they are compute/VM-related properties relevant to virtual machines or VM scale sets, not to a Cognitive Services multi-service account, which is a PaaS resource with no OS or VM sizing attributes.

Answer

VirtualMachineSize

Answer

OperatingSystemType

Answer

KernelVersion

Question 3

You are configuring a code interpreter tool for an AI agent in Azure AI Agent Service. The agent needs to analyze a CSV dataset uploaded by a user. How does the agent execute this task securely?

Accepted Answer

The agent generates Python code to analyze the CSV and executes it inside a secure, isolated sandbox environment managed by the service.. The code interpreter tool in Azure AI Agent Service generates and runs Python code in a secure, isolated sandbox managed by the service. This allows the agent to analyze uploaded CSV data without exposing the host environment or production systems.

Answer

The agent compiles the CSV data into C# binaries and runs them on the host Azure App Service instance.

Answer

The agent sends the raw CSV data to an external public third-party execution API.

Answer

The agent translates the CSV data into SQL queries and executes them directly against the primary production database.

Question 4

You are designing a high-availability architecture for an Azure AI Speech service solution. Which THREE practices should you incorporate into your design? (Choose three.)

Accepted Answer

Configure Azure Front Door or Traffic Manager with health probes to route traffic away from unhealthy regions.. Option B is correct because Azure Front Door or Traffic Manager with health probes provides DNS-based global load balancing and automatically routes clients away from an unhealthy Speech region, which is essential for high availability. Option C is correct because Azure Monitor alert rules on metrics such as server error rates and latency enable operations teams to detect degradation quickly and trigger failover or remediation before users are broadly impacted. Option D is correct because deploying Speech resources in multiple Azure regions supports active-passive or active-active failover, ensuring service continuity if one region becomes unavailable. Option A is incorrect because hardcoding a primary region IP bypasses DNS-based failover and creates a single point of failure. Option E is incorrect because a single-region deployment without backups offers no redundancy and cannot meet high-availability requirements.

Answer

Hardcode the primary region IP address in all client applications without using a routing DNS layer.

Answer

Rely exclusively on a single-region deployment without backups to minimize operational complexity.

Question 5

You are designing a deployment strategy for an Azure AI Custom Vision project in an enterprise pipeline. Which THREE practices should you follow to ensure secure and repeatable deployments? Each correct answer presents part of the solution.

Accepted Answer

Store API keys and connection strings securely in Azure Key Vault rather than hardcoding them.. Option A is correct because storing API keys and connection strings in Azure Key Vault keeps secrets out of source code and pipeline logs, and Custom Vision training/prediction keys can be referenced at runtime via Key Vault references or managed identities, supporting secure deployments. Option C is correct because Infrastructure as Code with ARM templates or Bicep provisions Azure resources (Custom Vision accounts, storage, compute) idempotently and consistently across environments, which is essential for repeatable deployments. Option D is correct because automated integration and validation tests in CI/CD pipelines verify that the deployed Custom Vision endpoint, keys, and model behavior work before production release, catching regressions and misconfigurations early. Option B is incorrect because manual portal configuration is error-prone, undocumented, and not repeatable, undermining both security and consistency. Option E is incorrect because embedding production API keys as plaintext comments in Bicep files exposes secrets in source control and violates secure secret-management practices.

Answer

Manually create and configure all production resources directly in the Azure portal without documentation.

Answer

Store production API keys as plaintext comments in Bicep template files.

Question 6

You are designing an advanced agentic solution in Azure AI Foundry that requires tool integration and state management. Which TWO practices should you implement to ensure robust agent reliability and security? (Choose two)

Accepted Answer

Implement human-in-the-loop confirmation steps before the agent executes high-impact or destructive transactional tool calls.. Option A is correct because human-in-the-loop confirmation is a key safety pattern in agentic systems: before the agent invokes high-impact or destructive transactional tool calls (e.g., deleting records, issuing payments, modifying production resources), a human approval gate prevents irreversible or costly actions caused by model hallucination or prompt injection. Option B is correct because model-generated parameters are untrusted input; validating and sanitizing them (type checks, allowlists, schema validation, escaping) before passing them to downstream enterprise tool APIs prevents injection attacks, malformed requests, and unintended data exposure. Option C is wrong because disabling logging and telemetry removes the observability needed to detect failures, audit tool calls, and debug agent loops, and it does not meaningfully improve reliability or security. Option D is wrong because granting unrestricted administrative permissions violates least privilege and dramatically increases blast radius if the agent is compromised or misbehaves. Option E is wrong because relying solely on the model's internal memory provides no durable, auditable multi-session state, leading to context loss, inconsistency, and inability to recover or trace conversations.

Answer

Disable all logging and telemetry to minimize latency during agent execution loops.

Answer

Grant the agent unrestricted administrative permissions to all corporate databases to maximize autonomous problem-solving.

Answer

Rely solely on the model's internal memory to maintain multi-session conversation state without external persistence.

Question 7

You are auditing your Azure AI services for cost optimization. You discover an Azure AI Search service that has been over-provisioned with too many search units and replica nodes. How can you right-size the service without losing your existing indexes?

Accepted Answer

Modify the scale settings in the Azure portal to reduce the number of replicas and partitions.. Azure AI Search lets you change scale settings — the number of replicas and partitions — directly in the portal without deleting the service, and existing indexes are preserved. Reducing replicas lowers query throughput capacity while reducing partitions lowers storage and indexing capacity, so you can right-size to match actual load. This is the non-destructive way to optimize cost.

Answer

Switch the pricing tier from Standard to Free.

Answer

Export all index definitions, delete the service, and re-import the indexes into a new Basic tier service.

Answer

Delete the search service and recreate it with fewer replicas and partitions.

Question 8

Your organization requires that all data sent to Azure OpenAI Service remains encrypted using a key that your security team fully controls and rotates. Which feature should you configure during resource creation?

Accepted Answer

Customer-managed keys (CMK) with Azure Key Vault. Customer-managed keys (CMK) with Azure Key Vault allow an organization to supply and control the encryption keys used to protect data at rest in Azure OpenAI Service. This satisfies the requirement for full key control and rotation by the security team, as opposed to Microsoft-managed keys.

Answer

Azure Information Protection client-side encryption

Answer

Service-managed keys (Default encryption)

Answer

Transparent Data Encryption with a SQL Managed Instance

Question 9

You are designing a customer feedback processing system. You need to extract key phrases from customer reviews and store them in an Azure Cosmos DB database. When you inspect the JSON response from the key phrase API, what data structure is returned for the key phrases?

Accepted Answer

An array of strings representing the key phrases found in the document.. The Azure AI Language key phrase extraction API returns a 'keyPhrases' field containing a JSON array of strings, where each string is a key phrase identified in the input document. This flat array structure makes it straightforward to iterate and persist phrases into Cosmos DB. It does not return weights, categories, or confidence scores for key phrases.

Answer

A single concatenated string with semicolons separating each phrase.

Answer

A nested dictionary mapping phrase weights to sentiment scores.

Answer

An array of objects containing 'text', 'category', and 'confidenceScore' properties.

Question 10

You are developing an application that needs to automatically extract key-value pairs and tabular data from standardized invoices without any prior labeled training data. Which Azure AI Document Intelligence model should you use?

Accepted Answer

Azure AI Document Intelligence Prebuilt Invoice model. The Prebuilt Invoice model is purpose-built for standardized invoice documents and extracts key-value pairs (vendor, invoice date, totals, line items) plus tabular data out of the box, with no labeled training data required. It is one of the Document Intelligence prebuilt models trained on millions of invoice documents, so it handles common invoice layouts immediately. The General Document model extracts text and generic key-value pairs but does not understand invoice-specific fields like InvoiceTotal or DueDate.

Answer

Azure AI Document Intelligence General Document model

Answer

Azure AI Document Intelligence Custom Extraction model

Answer

Azure AI Vision OCR feature

Question 11

You are using Azure AI Foundry to deploy an agent. You want to ensure that the agent only uses approved data sources. Which feature should you configure?

Accepted Answer

Grounding data configuration in the Agent settings.. Grounding data configuration in the Agent settings is where you specify the approved data sources (e.g., Azure AI Search indexes, uploaded files, or connected knowledge bases) that the agent can use. By configuring only approved sources, you restrict the agent's knowledge to vetted content, ensuring it does not pull from unapproved external information. This is the correct control for data source governance in Azure AI Foundry agents.

Answer

Prompt engineering for 'Do not use external info'.

Answer

Model temperature control.

Answer

Content Safety policy filters.

Question 12

You are building an agent in Azure AI Foundry using a prompt flow. You need to ensure that the agent only accesses specific internal documents and avoids hallucinating outside the provided context. Which grounding technique should you implement?

Accepted Answer

Implement grounding with data using Azure AI Search. Grounding with data using Azure AI Search connects the agent to a specific indexed document set, so responses are generated only from retrieved content. This directly prevents hallucination by constraining the model's context to the internal documents. System messages alone cannot guarantee the model will ignore its pretrained knowledge.

Answer

Use a system message to forbid outside information

Answer

Enable prompt caching

Answer

Increase the temperature parameter to 1.0

Question 13

You are building an AI agent in Microsoft Foundry that must answer questions grounded exclusively in a set of internal PDF policy documents. The agent must cite the exact page of the source document in every response, and the documents are updated weekly. You want the least operational overhead while keeping citations accurate. What should you do?

Accepted Answer

Upload the PDFs to the agent's file store and enable the built-in file search tool, which chunks, embeds, and indexes the documents in a managed vector store.. The managed file search tool handles chunking, embedding, and vector indexing automatically and returns citation metadata that includes file and page references. Because the index is rebuilt from uploaded files, weekly document refreshes need only file updates rather than custom reindexing pipelines, which satisfies both the accurate-citation requirement and the low-overhead requirement.

Answer

Create an Azure AI Search index manually, generate embeddings with a deployment, and wire the index to the agent through a custom function tool.

Answer

Enable the code interpreter tool and have the agent parse the PDFs at request time to extract text and page numbers.

Answer

Paste the full text of every PDF into the agent's system instructions so the model can quote the pages directly.

Question 14

Your team runs a multilingual complaint-triage pipeline using the Azure AI Language service. Documents arrive with a language field already set to 'en', but a subset actually contains Spanish text because upstream systems mislabel them. Sentiment results for those documents come back with low confidence and occasionally fail. You want the service to determine the language itself for every document rather than trusting the supplied value. What should you configure in the request?

Accepted Answer

Call the Language Detection task (kind 'LanguageDetection') first, then pass the returned ISO 639-1 code into the language property of the sentiment request.. Language Detection is a distinct prebuilt task that returns each document's detected language and confidence. By running it first and feeding the resulting ISO 639-1 code into the language property of the sentiment request, the correct language-specific model is applied and the mislabeled Spanish documents are analyzed accurately.

Answer

Set the model-version parameter to a multilingual model and continue sending the supplied 'en' value.

Answer

Set language to 'auto' in the documents array so the service performs language detection before analysis.

Answer

Remove the language property from each document object in the documents array.

Question 15

You are building an agentic solution in Microsoft Foundry where a planner agent delegates subtasks to several specialist agents. During testing, the planner repeatedly loses track of completed subtasks and re-delegates work, causing duplicate API calls and inflated costs. You need the planner to maintain reliable state across turns without exceeding the context window. What should you implement?

Accepted Answer

Implement a shared thread or state store where each subtask's status is recorded, and give the planner a tool to read and update that state between delegations.. Persisting subtask status in a shared store and exposing it to the planner through read and update tools provides durable, queryable state that survives across turns without consuming context. The planner checks status before delegating, so completed work is not repeated, and prompts stay small because only the relevant state is retrieved.

Answer

Convert the planner into a single monolithic agent that performs all subtasks sequentially in one prompt.

Answer

Raise the planner's temperature so it explores alternative delegation orders and avoids repeating identical calls.

Answer

Increase the model's context window by switching to a larger deployment and append the full conversation history to every planner request.