AB-410 Exam Questions

Embeddings models convert natural language into mathematical representations that support semantic matching in vector databases.

Video Indexer automatically extracts metadata, speech, and visual insights to produce summaries and structured breakdowns of video content.

Power Automate flows provide the bridge between Copilot Studio and external systems by using standard or custom connectors.

Accountability includes creating a feedback loop so humans can supervise and correct system outputs.

Power Automate can evaluate the current time and use conditional logic to determine whether the agent’s action should continue.

Prompt injection is a security threat in which users attempt to trick the AI into bypassing safety filters or exposing sensitive data.

The search engine, such as Azure AI Search, queries the vector index to locate the data most semantically related to the prompt.

Rollup columns summarize data from related records, giving the AI pre-calculated metrics directly from the table.

Model versioning helps prevent application breakage by letting developers test and migrate to newer model releases at their own pace.

Transparency makes sure the processes and data used to make AI decisions are clear and explainable to users.

Field-level security lets administrators restrict access to specific high-sensitivity columns in a table.

Groundedness verifies whether the answer is factually based on the provided context rather than fabricated by the model.

Azure AI Studio provides the interface used to configure and monitor content filtering policies for deployed models.

Power Automate serves as the logic engine that connects to external systems and returns data to the bot through connectors.

Speech-to-Text is the core service used to convert spoken audio into machine-readable text formats.

Grounding ties the model’s output to relevant, verifiable context, such as internal PDFs, to help prevent hallucinations.

Semantic Kernel is the orchestration SDK used to build AI agents that can “think” with an LLM and “do” through Plugins or Functions.

Prediction models are intended to forecast binary or multiple outcomes by using historical data patterns.

Managed Identities give applications an identity that can be used to connect to resources supporting Microsoft Entra authentication.

Semantic Ranker uses deep learning to re-evaluate the top search results so the most contextually relevant items are prioritized.

Max Tokens controls the length of the generated output, so for long summaries this limit needs to be high enough to allow the complete text.

Ownership-based security roles in Dataverse provide granular control over who can view, edit, or delete specific records.

Trigger phrases are examples supplied to the bot so it can identify the user’s intent and start the appropriate logic.

Azure OpenAI provides enterprise-grade security and compliance capabilities for using large language models in a managed environment.

Image Classification categorizes an entire image into specific labels, such as “Damaged” or “Intact.”

Prompt Shields, which are part of Content Safety, detect and block malicious instructions hidden in retrieved data or user prompts.

Azure AI Search is a specialized search engine that supports vector storage and fast retrieval for AI applications.

Prompt Flow enables developers to build a DAG (Directed Acyclic Graph) for orchestrating and evaluating AI workflows.

Top P restricts the model’s word choices to the most likely candidates until their combined probability reaches P, which supports high-quality variety.

Application Insights provides real-time telemetry and error tracking to help maintain the application’s operational health.

Global variables, also called Bot variables, persist for the full conversation, allowing information to move between different intents.

The Orchestrator, such as Semantic Kernel, coordinates multiple agents so they can solve complex, multi-step problems.

Language Detection identifies the source language, enabling the bot to route the interaction to the appropriate translation or response logic.

Validation and sanitization should be performed in the logic layer, such as Power Automate or APIM, before data reaches the backend database.

The Unstructured Document model uses layout-agnostic AI to locate and extract data from free-flowing text documents.

The system message contains the “core instructions“ that determine how the AI behaves, such as “You are a professional legal assistant.“

Sentiment Analysis reviews text to identify emotional tone, which is important for escalation and customer satisfaction logic.

Stop sequences are important for helping the model produce concise answers or follow a strict formatting template.

Prompt Shields are included in Azure AI Content Safety and are designed specifically to protect against jailbreaking and prompt injection.

Privacy and Security focuses on protecting personal data and making sure the system is resilient against unauthorized access.

Speech-to-Text converts the audio into text, and Azure OpenAI (LLM) summarizes the transcribed text.

Azure Cost Management + Billing provides detailed reports for token consumption and related spending.

Grounding keeps the AI’s responses anchored to your specific business data instead of relying only on general training knowledge.

Object Detection can find and label multiple specific items, such as a vest or helmet, within one image.

Few-shot prompting uses in-context examples to “teach“ the model the expected pattern or logic for a task.

Category Classification is the appropriate model for tagging and routing text-based documents into defined business categories.

The Prediction model examines historical trends to forecast binary outcomes, such as Will return or Will not return.

Inclusiveness helps ensure AI applications empower everyone and meet the needs of diverse populations regardless of ability.

Implicit sharing lets users use AI capabilities without requiring each user to receive separate permissions for the backend service.

Azure AI Vision extracts visual features and tags, and Azure OpenAI (LLM) uses those tags to generate a natural language description.

Zero-shot prompting relies on the model’s internal general knowledge to complete a request without specific guidance examples.

Document Intelligence provides specialized models for accurate OCR and field extraction from both handwritten and printed forms.

RBAC in Dataverse helps ensure that only authorized users can run sensitive AI processes or access protected data.

Coherence assesses the logical and structural quality of the AI output to confirm that the response makes sense overall.

Zero-shot prompting depends entirely on the model’s pre-trained knowledge to understand and follow an instruction.

Sentiment Analysis examines text to determine the dominant emotional tone, which is essential for customer service routing.

Max Tokens places a hard limit on the number of tokens the model can generate, directly influencing both cost and response length.

Accountability establishes clear responsibility for how an AI system functions and for its ethical outcomes.

Slot filling is the process in which the bot identifies and collects the specific information, or entities, required to complete a task.

The System Message specifies how the model should behave, such as “You are a helpful assistant that only answers based on PDF context.”