AB-410 Exam Questions

Azure AI Content Safety includes built-in filters that detect and block hate, violence, self-harm, and sexual content in AI-generated text and images.

Administrators use the Power Platform Admin Center to configure DLP policies that manage how data is shared through connectors and AI agents.

Temperature adjusts the probabilities of the next token; values closer to 0 make the model more predictable and focused.

Speech-to-Text, part of Azure AI Speech, converts spoken audio into written text for additional processing.

A prompt is the text or instruction provided to an LLM that guides what it should generate or perform.

Azure AI Personalizer is a cloud-based service that uses reinforcement learning to help applications select the best content or item to show users.

Groundedness assesses whether the model’s response is factually based on the supplied reference context instead of being fabricated.

The Vector Store contains numerical representations, or embeddings, of your data so the system can quickly locate relevant context for a prompt.

A Confusion Matrix is a table that describes how a classification model performs on data where the true values are known.

The “Adult“ capability in Azure AI Vision Image Analysis is specifically designed to detect mature, racy, or violent content.

SLMs, such as the Phi-3 family, use fewer parameters, which makes them faster and less expensive to run on edge devices than massive LLMs.

Sentiment Analysis is a core Azure AI Language capability used to identify positive, negative, or neutral tone in text.

Grounding gives the AI a specific source of “truth,“ such as a product list, so the generated answer is factually accurate.

Dynamic Chaining enables the orchestrator to choose the most relevant plugins and topics in real time to complete a complex user request.

Clear formatting instructions, such as telling the model to respond only in JSON using a specified schema, are required for consistent programmatic output.

Although many settings exist, ensuring that the bot can handle rich media such as Adaptive Cards is essential for interactive intelligent applications.

Language Detection identifies the source language, which can then be used to route the conversation or translate the output.

Azure AI Speech includes Speech-to-Text and Speech Translation capabilities that support real-time translation across multiple languages.

Azure APIM lets developers publish, secure, and analyze APIs, making it the appropriate “front door” for AI agents that need access to backend data.

The max_tokens parameter places a hard limit on generated response length, although tokens do not correspond perfectly to words.

Azure AI Studio includes evaluation tools that calculate metrics such as groundedness, relevance, and coherence against a test dataset.

The Copilot pane in Power Pages Design Studio lets users describe the form they need, then automatically creates the Dataverse table and form layout.

The Unstructured Document Model, formerly General Document, uses layout-agnostic technology to find data even when document formats differ.

Power Automate provides connectivity to external messaging services for transactional tasks such as sending SMS codes.

Azure AI Document Intelligence includes advanced OCR capabilities designed for printed and handwritten text across different document formats.

The Fairness principle requires AI systems to treat people equally and avoid biased outcomes.

Coherence evaluates the logical consistency of AI-generated text and how smoothly its ideas transition.

HITL makes sure that AI-generated decisions in critical or high-stakes situations are checked by a person for accuracy and safety.

A skillset specifies a sequence of operations that extract information or add structure to raw data during indexing.

Implicit sharing lets the app perform actions on behalf of the user through the app creator’s connection, within defined security boundaries.

Object Detection can be trained to identify and locate specific items, such as logos or products, inside an image.

The Copilot Studio Canvas is the visual designer where developers organize the logic, triggers, and actions of intelligent agents.

Fine-tuning retrains the model on specific data to modify its behavior or knowledge, while RAG supplies external context to a pre-trained model at query time.

Intent Recognition, part of Azure AI Language, detects the goal or purpose of a user’s input so the correct workflow can be triggered.

Few-shot learning means including a small number of examples, or shots, in the prompt to guide the model’s output format or logic.

The System Message tells the model how to behave, what tone to use, and which constraints to follow during the session.

The Planner is the intelligence layer that decides which tools or agents to call, and in what sequence, to solve a complex request.

Input variables make the prompt dynamic by pulling in data from Dataverse or a Power Automate flow to provide specific context.

Hallucination happens when an LLM produces output that is not supported by its training data or by the provided context.

System Topics are built-in templates that handle a bot’s basic conversational flow and error handling.

Microsoft is introducing a native Vector data type in Dataverse to support efficient storage and retrieval of embeddings for RAG-based applications.

Adaptive Cards provide a platform-agnostic way to display UI snippets, such as buttons, images, and text blocks, generated by an AI bot.

Hybrid search brings together traditional keyword matching and the semantic depth of vector search to return more accurate results.

Prompt Flow in Azure AI Studio supports creating, evaluating, and iteratively testing complex prompt logic and AI workflows.

Copilot Studio lets you configure authentication providers, such as Entra ID, to support secure token exchange for API calls.

Microsoft’s Phi-3 series consists of Small Language Models (SLMs) designed for strong performance in low-compute or edge environments.

The Indexer is the crawler that connects to a data source, reads the data, and serializes it into the search index.

Custom Text Classification enables developers to train a model using business-specific labels, such as Claims or Billing.

Agentic Workflow means the AI agent makes logical decisions and calls external actions, such as Plugins or Flows, to complete a complex business process.

Vector search uses embeddings to return results that are conceptually related to the query, even when the exact words are different.

AI Builder offers a no-code, user-friendly interface in Power Apps and Power Automate for training and using custom document processing models.

The Fallback topic runs when the AI cannot match the user’s input to an existing topic, enabling a graceful error message or escalation.

PII Detection is an Azure AI Language feature that identifies and can help mask sensitive information such as names, emails, and SSNs.

Dataverse supports geospatial data types and functions that let apps calculate distances and find records based on geographic coordinates.

The Text Summarization pre-built model is intended to create concise versions of long-form text documents, including handbooks.

Text Recognition (OCR) extracts the text, Translator converts it to another language, and Text-to-Speech (TTS) produces the audio output.

Top P, or nucleus sampling, restricts model choices to a cumulative probability percentage, helping balance diversity and focus.

Semantic Ranker uses Microsoft’s language models to reassess the top search results so the most contextually relevant items appear first.

Lowering Max Tokens reduces the allowed output length, which forces the model to produce a shorter response.

Models such as GPT-4o-mini or GPT-3.5 Turbo are optimized for speed and cost efficiency compared with larger models such as GPT-4.