AB-410 Exam Questions

Groundedness and Accuracy measure how closely the model‘s output matches the provided factual context or expert answers.

The Planner acts as the logical brain that determines which agents and tools are required to complete a complex request.

Document Translation is a specialized Translator service feature that preserves the file‘s physical layout.

Connecting Azure AI Speech with Conversational Language Understanding (CLU) enables the system to map audio directly to business intents.

Fairness focuses on treating all users equitably and reducing historical or systemic biases in data.

Microsoft Purview AI Hub provides centralized visibility into AI usage and data security.

Agentic “Computer Use“ enables the AI to understand the screen and interact with UI elements in the way a human would.

The Face API provides face detection and 1:1 verification matching.

Model Versioning lets developers control when they move to newer model releases, helping keep their apps stable.

Behavioral constraints, such as sentence length or persona, are defined in the System Message.

Regularly testing the model against a Golden Dataset of current data helps ensure accuracy stays high as trends change.

Using smaller chunks with overlap helps ensure detailed information is not lost at the boundaries between text segments during retrieval.

Document Intelligence, formerly Form Recognizer, is optimized for high-accuracy OCR, including challenging or low-quality images.

Ownership-based security provides detailed control based on the user who created or owns a specific record.

Dataverse Search, which is powered by Azure AI Search technology, provides a global and intelligent search experience across all relevant tables.

Accountability includes establishing ways for humans to oversee the system and provide feedback so the system functions as intended.

The Retrieval Engine queries the index to locate the most semantically similar context to add to the prompt.

SLMs provide high-quality reasoning in a smaller form, making them well suited for specific tasks or edge deployments.

Zero-shot prompting depends on the model‘s pre-trained knowledge to complete a request using only the instruction.

MCP is a standard used in the Copilot ecosystem to help agents safely connect AI models with local and remote data.

Entity Extraction is designed to extract structured data points, such as currency or dates, from natural language text.

RAG enables the agent to retrieve information from a dynamic external knowledge base so its responses are grounded.

Contextual chunking keeps related ideas, such as a complete paragraph, together, which improves the quality of retrieved context.

The context window sets the memory limit for a single session; after the conversation exceeds that limit, the model starts to “forget“ earlier content.

Rate limiting, or throttling, prevents API exhaustion and manages costs by limiting how many requests can be made per minute.

Application Insights monitors API failures and throttling, which is essential for maintaining service availability.

TTFT is an important performance metric for real-time applications because it measures how responsive the bot feels to the end user.

Azure AI Studio evaluation tools assess responses for safety risks, including hate speech or self-harm, and provide a measurable score.

“Computer Use“ enables agents to work with graphical user interfaces, including browsers and apps, in the same way a human operator would.

Content Safety includes Prompt Shields, which are designed specifically to block jailbreak and prompt injection attempts.

Purview provides data lineage, which is important for auditing the source and movement of data into AI models.

Global variables persist during the conversation, so data such as a User ID can be used across multiple topics.

Slot filling captures the entity value from natural language and stores it in a variable for logic processing.

Handoff makes sure complex or high-touch queries move smoothly to the right resource.

CoT prompting encourages the model to separate complex tasks into logical steps before reaching a conclusion.

User-scoped variables let the bot remember a specific person‘s preferences even when that person returns days later.

SLMs are efficient and fast, and they can often run locally or at the edge for specific reasoning tasks.

Kernel functions are the core building blocks, or skills and logic, that an agent uses to complete tasks or call LLMs.

Prompt Shields are a specialized defense in Content Safety that is designed to detect adversarial inputs.

Microsoft Purview AI Hub gives visibility into AI usage and helps support data security and compliance.

Inclusiveness ensures AI benefits everyone, including people with visual, auditory, or cognitive impairments.

Dynamic chaining lets an agent reason about which tools it needs and the order in which to use them to complete a multi-step task.

Azure OpenAI provides the same models along with Microsoft‘s layers of security, compliance, and private data residency.

Video Indexer identifies objects, people, and themes throughout a video‘s timeline, making visual content searchable.

Microsoft provides specialized dashboards for tracking the environmental impact of cloud and AI services, including energy and carbon.

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

Microsoft Purview‘s AI Hub gives centralized visibility into AI usage, risks, and compliance across the organization.

Transparency means making the AI‘s thought process and data sources clear to the end user.

Transparency makes the system‘s logic clear, often through “Thought“ traces or citations included in the response.

Metadata filtering lets the system exclude irrelevant data using structured attributes before semantic search is performed.

Entities enable the bot to recognize and “fill“ specific data slots, such as a Date or Name, from within a sentence.

Self-correction lets the agent refine its own reasoning so the response meets constraints or factual requirements.

Entities support “Slot Filling“ by helping the bot recognize specific data points in a natural language sentence.

Setting Temperature to 0 makes the model deterministic, which supports consistency and precision in technical scenarios.

The Face API in Azure AI Vision provides face detection, recognition, and verification through 1:1 matching.

Privacy and Security ensure that personal and sensitive data are managed in accordance with privacy laws and protected from leaks.

Managed Identities, which are part of Entra ID, provide applications with an identity they can use to connect to resources that support Entra authentication.

Purview tracks data lineage by showing how data moves from its source to its use in AI models.

Solutions are the ALM mechanism for bundling apps, flows, and bots so they can be deployed securely and consistently.

The Orchestrator or Planner controls the logical order of work and delegates tasks between specialized agents.