AB-410 Exam Questions

Trigger phrases are the natural language examples used by the orchestrator to map user intent to a specific topic.

The Orchestrator coordinates the steps: taking the user input, querying the search index, and passing the context to the LLM.

Few-shot learning involves providing a small number of examples (shots) within the prompt to guide the model‘s output format or logic.

HITL ensures that critical or high-stakes decisions made by AI are verified by a person to ensure accuracy and safety.

Implicit sharing allows the app to perform actions on behalf of the user using the app creator‘s connection, subject to specific security boundaries.

Transparency ensures that users are aware of the AI‘s capabilities and limitations, including the fact that they are talking to a bot.

Temperature adjusts how “creative“ or “deterministic“ the model is; 0 is very literal, while 1.0 is very creative.

Top P (nucleus sampling) limits the model‘s choices to a cumulative probability percentage, helping balance diversity and focus.

PII Detection is a feature of Azure AI Language that identifies and can help mask sensitive info like names, emails, and SSNs.

Copilot Studio allows you to configure authentication providers (like Entra ID) to handle secure token exchange for API calls.

Fairness focuses on preventing bias and ensuring equitable treatment for all users.

The Prebuilt Receipt model is specifically trained to identify key-value pairs commonly found in financial receipts and invoices.

Using a read-only, verified knowledge source ensures the model‘s “truth“ comes from an authoritative database rather than untrusted user input.

Few-shot learning uses in-context examples to help the model understand the specific format or logic required for a task.

Slot filling automatically captures the entity value (seat type) from the user‘s natural language and saves it to a variable.

Knowledge Sources are the data repositories that a Copilot agent uses to retrieve information and answer user queries without manual topic creation.

Global variables allow the bot to remember information (like a user‘s ID or preference) as the conversation moves from one topic to another.

Presence Penalty encourages the model to talk about new topics by penalizing tokens that have already appeared in the output.

Object Detection not only identifies that an item is in an image but also provides the specific pixel coordinates (bounding box) for its location.

Few-shot prompting provides context and examples within the prompt, which helps the LLM understand complex patterns without needing a full fine-tuning process.

Semantic Kernel is the SDK that allows developers to orchestrate AI services with traditional programming languages.

Entity Extraction is designed to recognize and pull specific data points (like dates, names, or amounts) from unstructured text.

Lowering the temperature (closer to 0) makes the model more deterministic and focused on the most likely response.

The system role provides the foundational instructions that govern how the AI interacts and what rules it must follow.

Azure AI Search supports vector indexing and search, which is essential for matching user queries with the most relevant data chunks.

Grounding ensures the model‘s responses are based on trusted data (like your company‘s knowledge base) rather than its general training data.

Azure AI Document Intelligence offers prebuilt and custom models to extract specific fields and structures from documents.

Grounding involves providing the AI with a specific knowledge base (context) to ensure its responses are factually accurate and relevant to the organization.

RAG combines a search index (to find the specific paragraph) with an LLM (to draft the natural language answer based on that paragraph).

By integrating Azure AI Speech with CLU, the system can bypass the text-only phase and directly map spoken language to specific business intents.

Plugins allow Copilot agents to perform actions or fetch data from external systems via Connectors or Power Automate flows.

Intent Recognition (part of Azure AI Language) identifies the goal or purpose behind a user‘s input to trigger the correct workflow.

Brand Detection is a specialized feature within Azure AI Vision that can identify thousands of global logos and brands within images.

Embedding models convert text into high-dimensional vectors (embeddings) so they can be compared for semantic similarity.

Deflection Rate measures the percentage of sessions that are successfully handled by the AI without escalating to a human agent.

Sentiment Analysis can categorize feedback as positive, negative, or neutral, allowing the flow to flag “Negative“ feedback as urgent.

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

Hybrid search combines the strengths of traditional keyword matching with the semantic depth of vector search for the most accurate results.

Semantic Ranker uses Microsoft‘s language models to re-evaluate the top results of a search to ensure the most contextually relevant items are at the top.

The Unstructured Document Model (formerly General Document) uses layout-agnostic technology to find data even when document formats change.

System Topics are built-in templates that manage the basic conversational flow and error handling of a bot.

Tokens are the basic units of text that LLMs process; one token is roughly equivalent to 4 characters in English.

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

Microsoft Purview ensures that the data being fed into or generated by AI apps is governed, classified, and compliant with privacy regulations.

The Planner is the intelligence layer that determines which tools or agents to call in what order to solve a complex request.

Chain-of-Thought prompting encourages the model to “think step-by-step,“ improving its performance on complex logical or mathematical tasks.

The Fallback topic is triggered when the AI cannot match the user‘s input to any existing topic, allowing for a graceful error message or escalation.

Solutions are the transport mechanism in Power Platform for moving apps, flows, and Copilot agents across different environments.

Microsoft’s Phi-3 series are Small Language Models (SLMs) optimized for high performance on low-compute or edge environments.

Custom Connectors allow the Power Platform to talk to any service that has a public or private REST API.

Custom Text Classification allows developers to train a model on specific labels (like Claims or Billing) relevant to their business.

Dynamic Chaining allows the orchestrator to select the most relevant plugins and topics in real-time to fulfill a complex user request.

Azure APIM allows developers to publish, secure, and analyze APIs, making it the ideal “front door“ for AI agents to access backend data.

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

Providing specific formatting instructions (e.g., “Respond only in JSON format following this schema“) is essential for consistent programmatic output.

Semantic Kernel is the SDK designed to integrate LLMs with “skills“ and “connectors“ to interact with real-world data and applications.

A skillset defines a series of operations that extract information or create structure from raw data during the indexing process.

Speech-to-Text (part of Azure AI Speech) transcribes spoken audio into written text for further processing.

Generative Answers use an LLM to search specified data sources (like a website or SharePoint) to answer questions not explicitly programmed in the bot‘s topics.

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