AI-901 Practice Exam — AI-901：Microsoft Azure AI Fundamentals

Option Analysis: A. This option is incorrect because AudioOutputConfig is used to configure the destination for speech synthesis output, not to configure the core connection to the Azure Speech service. B. This option is incorrect because SpeechSynthesizer is the top-level class used for text-to-speech operations, not for configuring service connection settings. C. This option is incorrect because AudioConfig is used to configure the audio input source for speech recognition, not the connection to the Azure Speech service itself. D. This option is correct because SpeechConfig is the Azure Speech SDK class that stores core connection configuration, including your speech resource subscription key and region, to authenticate and connect to the Azure Speech service. Key Concept: In the Azure Speech SDK, the SpeechConfig class is used to store core authentication and endpoint configuration required to connect to the Azure Speech service for all speech-enabled operations. References: Azure Speech SDK for Python documentation overview, https://learn.microsoft.com/en-us/python/api/azure-cognitiveservices-speech/azure.cognitiveservices.speech.speechconfig?view=azure-python Get started with the Speech SDK

Analysis: The correct answer is a base64-encoded image data. The other options are incorrect: a CSV file attachment stores only tabular text data, an MP4 video stream is for video content not static images, and an SAS token is an authentication credential for Azure storage resources, not a format for providing image data to the API. The OpenAI Responses API for vision-enabled models accepts images directly embedded as base64-encoded image data in the request payload, in addition to publicly accessible image URLs. Key Concept: When using vision-enabled models with the OpenAI Responses API, images are provided to the model as either base64-encoded raw image data or a publicly accessible image URL. References: OpenAI Responses API Vision Documentation, Microsoft Learn Azure OpenAI Vision quickstart

Option Analysis: A. This option is correct because system instructions in Microsoft Foundry define the specific role, behavior, and response guidelines that an AI agent follows when interacting with users. B. This option is incorrect because temperature controls the randomness and creativity of model outputs, not the agent's core role or defined behavior. C. This option is incorrect because tokens per minute (TPM) is a rate limiting setting that controls how many tokens can be processed per minute, it does not configure agent role or behavior. D. This option is incorrect because max completion tokens sets the maximum length of a single agent response, it does not define the agent's role or behavior. Key Concept: System instructions (system prompt) in Microsoft Foundry are used to define the custom role, behavior, and response constraints for a generative AI agent. References: Create and deploy an AI agent in Microsoft Foundry, Prompt basics in Azure OpenAI, https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/prompt-engineering

Analysis: Correct answers: Statement 1 "System prompts can be used to authorize users" = No; Statement 2 "A system prompt is used to reduce tokens per minute (TPM)" = No; Statement 3 "A system prompt guides the behavior of a generative AI model" = Yes. User authorization is handled by identity and access management systems, not system prompts for generative AI. System prompts are fixed context instructions that add token overhead to requests, they do not reduce TPM usage. The core purpose of a system prompt is to define desired behavior, output format, and constraints for a generative AI model. Key Concept: A system prompt is a top-level instruction given to a generative AI model to define its behavior, constraints, and output requirements, and it does not handle authentication or control token rate limits. References: Introduction to prompt engineering, Microsoft Learn, https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/prompt-engineering System messages, OpenAI Documentation

Option Analysis: A. This option is incorrect because an OCR-only pipeline only extracts raw text from documents and cannot output structured, schema-aligned data including nested extracted values. B. This option is incorrect because Azure Speech transcription is designed for converting audio speech to text, which is irrelevant to extracting structured data from visual invoice documents. C. This option is correct because Azure Content Understanding analyzers let you define a custom schema to extract structured fields including nested values from unstructured documents like invoices. D. This option is incorrect because Azure AI Search is a service for full-text search and indexing over custom content collections, not for schema-based extraction of structured fields from incoming invoices. Key Concept: Schema-based structured data extraction from unstructured documents in Microsoft Foundry Tools is handled by Azure Content Understanding custom analyzers. References: What is Azure Content Understanding?, https://learn.microsoft.com/en-us/azure/ai-services/content-understanding/overview Azure Content Understanding analyzers