Free Microsoft Developing AI Cloud Solutions on Azure AI-200 Exam Questions

Detailed If the priority is to minimize stale results, source-driven invalidation is the direct control: when the underlying record changes, delete the corresponding cache entry so the next request repopulates it from current data. Sliding expiration can keep a frequently accessed stale item alive, disabling expiration is worse, and an allkeys-LRU policy evicts according to memory pressure rather than data freshness. Invalidation therefore best addresses correctness without requiring constant cache-wide maintenance.

Study Guide Alignment: AI data-management workloads: Cosmos DB, PostgreSQL, caching, vector storage, vector retrieval, consistency, and connection optimization.

Official Microsoft Learn Reference: AI-200 Study Guide | Azure Managed Redis documentation

To reduce P95 latency for filtered vector similarity queries when experiencing sustained high CPU use, you should create a vector index on the embedding column. The pgvector extension in Azure Database for PostgreSQL supports creating indexes (such as HNSW or IVFFlat indexes) on vector columns. Since your queries filter by department before ranking by cosine distance, and you have high CPU utilization, the lack of proper indexing is causing the database to perform full table scans and compute distance calculations for all 5 million embeddings. Creating a vector index will significantly accelerate similarity searches by using approximate nearest neighbor algorithms, dramatically reducing both CPU usage and query latency. You may also create a composite index combining the department filter column with the vector column for further optimization.

To reduce false-positive intent predictions, add representative utterances to the None intent. The None intent exists specifically for user utterances that should not map to any defined business intent. Microsoft recommends adding examples that resemble potential false positives so that the model learns a stronger decision boundary between valid intent utterances and unrelated or ambiguous input.

For example, if the GetContactDetails intent contains requests such as ''give me the customer phone number,'' the None intent should contain similar-looking but semantically unrelated phrases that could otherwise be incorrectly classified as GetContactDetails. During prediction, an utterance can be classified as None when it resembles None-training examples or when the highest intent score falls below the configured None threshold. Microsoft specifically advises adding false-positive examples to the None intent to improve intent discrimination.

Adding more examples to GetContactDetails alone generally strengthens recognition of that intent but does not provide the model with enough negative examples. A machine-learned entity addresses entity extraction rather than intent classification. Active learning helps identify uncertain utterances for review, but it is not the direct corrective action requested.

Study Guide references: Azure AI Language Conversational Language Understanding; intents; None intent; intent classification; reducing false positives.

The solution statement is incorrect because the KQL query does not return every individual request. The summarize operator aggregates input records according to the grouping expression. In this query, count() calculates the number of request records in each group, while by resultCode creates one group for every distinct HTTP result code represented in the selected Application Insights time range. Microsoft documents that summarize produces one output row for each distinct combination of its by expressions.

The resulting dataset therefore contains columns equivalent to resultCode and request_count, not individual request rows. The subsequent order by request_count desc sorts those aggregate groups in descending order, placing the most frequent result code first. Microsoft provides essentially the same Application Insights pattern for displaying the distribution of requests by result code: summarize request counts by result code and order the resulting counts descending.

If the intent were to list individual requests, the query would omit summarize and typically use project to select fields such as timestamp, operation name, and result code.

Therefore, because the proposed solution incorrectly describes the query as listing each individual request, the answer is No.

Study Guide references: Azure Monitor Logs; Application Insights requests; KQL summarize; count(); grouping; descending sorting.

A retrieval-augmented generation (RAG) architecture is the correct mechanism when generated answers must be traceable to specific source documents. RAG first retrieves relevant document chunks from a search index or other knowledge source, supplies those chunks to the language model as grounding context, and then generates an answer based on that retrieved information. Microsoft specifically documents that RAG enables responses containing citations back to source content.

For reliable citation generation, each retrieved chunk should carry identifying metadata such as a source identifier, document title, section, page, or URL. Microsoft's Azure RAG guidance recommends labeling retrieved chunks and including source metadata with them, then explicitly instructing the model to cite the corresponding sources in its response. This allows users to verify individual claims against the underlying documents.

Modern Azure AI Search agentic retrieval can also return structured references and grounding metadata, specifically designed for citation linking.

Increasing top_p only alters token sampling. Content filtering addresses safety, not provenance. Using a base model without retrieval provides no authoritative document linkage.

Study Guide references: RAG architecture; grounding data; retrieval metadata; Azure AI Search references; prompt engineering for citations.

Use Azure AI Speech with Speech Translation. Microsoft documents Speech Translation as a real-time capability that accepts a live audio stream in a source language and returns translated text, synthesized speech, or both in one or more target languages. This matches the scenario precisely: the input is spoken English during a live support call, the required output is Spanish text, and latency must be minimal.

Speech Translation can be implemented through the Speech SDK, which returns intermediate recognition and translation results as speech is detected. That streaming behavior is materially better suited to live conversations than a workflow that first transcribes audio and then separately calls a text-only translation API. Microsoft specifically positions Speech Translation for low-latency, real-time speech-to-text and speech-to-speech translation scenarios.

Azure AI Translator translates text and therefore would require a separate speech-to-text stage before translation. Azure AI Language provides NLP capabilities such as classification, sentiment, and entity extraction rather than live speech translation. Azure AI Document Intelligence extracts structured information from documents and forms and is unrelated to streaming conversational audio.

Therefore, Azure AI Speech (Speech Translation) is the correct choice.

Study Guide references: Azure AI Speech Speech Translation; real-time transcription and translation; Speech SDK; streaming audio translation; target-language text output.

Azure Container Apps session affinity, also called sticky sessions, routes requests from the same client to the same container-app replica. Microsoft documents that this functionality is implemented using HTTP cookies, so it requires HTTP ingress rather than raw TCP ingress.

The second requirement is single revision mode. Microsoft explicitly states that session affinity is supported only when the container app is in single revision mode and HTTP ingress is enabled. In single revision mode, all traffic is directed to the latest active revision, allowing the ingress layer to consistently associate a client with a particular replica within that revision.

Multiple revision mode is designed for scenarios such as traffic splitting, blue-green deployments, and A/B testing, where inbound traffic can be distributed among multiple active revisions. That model is incompatible with the documented session-affinity restriction. TCP ingress is also incorrect because sticky-session behavior relies on HTTP cookie affinity rather than generic TCP connection routing.

Therefore, the required configuration is HTTP ingress with single revision mode.

Study Guide references: Azure Container Apps ingress; sticky sessions/session affinity; revision modes; HTTP routing.

Azure Container Registry Tasks provides Azure-hosted container build capabilities, eliminating the need to maintain external Docker build servers or dedicated CI build infrastructure. A Quick task uses commands such as az acr build to send the build context to ACR-managed compute, where Azure builds the container image and, by default, pushes the successfully built image directly into the registry. Microsoft describes quick tasks as cloud-based, on-demand build-and-push operations that do not require a local Docker Engine.

For automatic execution after source-code changes, use a Source-triggered task. ACR Tasks can associate a task with a GitHub or Azure Repos repository and automatically execute the image build whenever code is committed to the configured branch. Microsoft documents that when a commit occurs, the source-control trigger causes ACR Tasks to build and push the resulting image.

A Webhook is not the required standalone ACR build component here. ACR Tasks creates and manages the necessary source-repository webhook as part of configuring the source trigger. An Artifact Cache rule caches upstream container artifacts and does not build application images from Git commits.

Study Guide references: Azure Container Registry Tasks Quick tasks; Automatically triggered tasks; Source-code commit triggers; cloud-based image builds.

Detailed The API should use a system-assigned managed identity so no application credential is stored or rotated by the development team. That identity should receive only the permission needed to read secret values: Key Vault Secrets User at the vault scope is materially narrower than Key Vault Administrator at subscription scope. Storing a secret value in App Configuration would violate the requirement to protect secrets and increases exposure. The corrected pair therefore implements both credential-free authentication and least-privilege authorization with a limited blast radius.

Study Guide Alignment: Security and operations: Key Vault, App Configuration, managed identity, OpenTelemetry, Azure Monitor, and KQL-based troubleshooting.

Official Microsoft Learn Reference: AI-200 Study Guide | Azure Key Vault RBAC guide | Managed identities for Azure resources

Use the Azure AI Document Intelligence prebuilt invoice model. The prebuilt invoice model is specifically designed to analyze invoices and extract structured fields and line items without requiring an organization to train a custom model. Microsoft documents support for fields including InvoiceId, VendorName, InvoiceTotal, InvoiceDate, CustomerName, AmountDue, tax information, addresses, and invoice line items.

Because the model is pretrained to recognize common invoice structures, it can process invoices that use different layouts and formats while returning normalized structured output. This directly satisfies the requirement to extract invoice number, total, and vendor information from scanned documents while minimizing development and training effort.

The Azure AI Vision Read API performs OCR and can extract printed or handwritten text, but it does not provide invoice-specific semantic field extraction. Custom named entity recognition would require training and is designed for text entity extraction rather than document-layout understanding. A Document Intelligence custom neural model can handle organization-specific forms and layouts, but it requires custom model training, directly violating the requirement.

Therefore, the prebuilt invoice model provides the required combination of OCR, document understanding, structured extraction, and zero custom training.

Study Guide references: Azure AI Document Intelligence prebuilt invoice model; structured field extraction; OCR; invoice fields and line items.