AI-901 Sample Questions & Answers
Implementing with Microsoft Foundry, from generative apps and agents to text, speech and image-generation capabilities, carries well over half the weight, with the rest on responsible AI principles, model components and common workload types.
Launch the full AI-901 simulator →Showing 6 of 12 free samples.
- Question 1IntermediateSelect 2
Identify AI concepts and capabilities · Identify appropriate model deployment options and configuration parameters
You are deploying a large language model via the Microsoft Foundry portal to draft marketing copy. The marketing team complains that the model's responses are too repetitive and lack creative variance. Furthermore, the responses occasionally cut off mid-sentence.
Which TWO configuration parameters should you adjust to resolve these specific issues? (Select TWO)
Show answer & explanation
Correct answers: B, D
Temperature controls the randomness and creativity of the model's output. A higher temperature (closer to 1.0) will make the responses more creative and varied, resolving the complaint about repetitive copy. Max tokens controls the length limit of the response; increasing it prevents the output from cutting off mid-sentence.
Max tokens determines the maximum number of tokens (words or word pieces) the model can generate in a single response. If responses are cutting off mid-sentence, the max tokens limit is set too low and needs to be increased. Temperature is adjusted to fix the lack of creativity.
- Question 2Beginner
Identify AI concepts and capabilities · Identify appropriate model deployment options and configuration parameters
True or False: Before you can use the Microsoft Foundry SDK to build a generative AI application, you must first deploy a model from the model catalog to an endpoint.
Show answer & explanation
Correct answer: A
True. In Microsoft Foundry, models exist in a catalog as artifacts. To interact with a model programmatically via the Foundry SDK (or via the portal playground), you must first provision hosting resources by deploying the model to an endpoint. The SDK uses this endpoint URL and an API key to communicate with the model.
- Question 3Intermediate
Identify AI concepts and capabilities · Describe common text analysis techniques
A legal firm wants to automatically scan thousands of contracts to locate and highlight specific names of companies, dates of signing, and monetary values. Which text analysis technique is best suited for this workload?
Show answer & explanation
Correct answer: B
Entity detection (or Named Entity Recognition) is the technique used to identify and categorize specific entities in text, such as people, organizations, dates, locations, and monetary values. Keyword extraction identifies the main talking points but does not categorize them as specific entity types. Sentiment analysis measures the positive/negative tone of the text.
- Question 4Advanced
Identify AI concepts and capabilities · Identify features and capabilities of computer vision and image-generation models
A manufacturing plant is implementing a quality control system on its assembly line. The system uses cameras to evaluate parts as they pass by. The application needs to identify if a microscopic crack is present on a part, and if so, return the exact bounding box coordinates of the crack so a robotic arm can remove the part.
Which computer vision capability is required for this workload?
Show answer & explanation
Correct answer: A
Object detection is required because it not only identifies what is in the image (a crack) but also provides the spatial location of the object via bounding box coordinates. Image classification only provides a label for the entire image (e.g., 'defective' or 'normal') without pinpointing where the defect is located. Image generation creates new images, which is not relevant here.
- Question 5Intermediate
Identify AI concepts and capabilities · Identify scenarios for common AI workloads
You are evaluating AI workloads for a customer service department. They want a solution that can receive a customer's email, determine if the customer is asking for a refund, query the internal billing SQL database to check eligibility, and automatically process the refund via an API if approved.
Which type of AI workload best describes this scenario?
Show answer & explanation
Correct answer: B
Agentic AI (or AI agents) refers to AI systems that go beyond merely generating text; they can autonomously plan tasks, make decisions, and invoke external tools or APIs (like querying a SQL database and triggering a refund API) to achieve a goal. Standard Generative AI would only be able to draft a response email, not take independent actions in external systems.
- Question 6Advanced
Identify AI concepts and capabilities · Identify techniques to extract information from text, images, audio, and videos
A media broadcasting company has a vast archive of unstructured data, including:
- Scanned historical news articles (PDFs with text and images)
- Audio recordings of radio broadcasts (MP3)
- Video footage of news events (MP4)
The company wants to build a unified search index. They need to extract key entities, dates, and summaries from all three data types using a single, unified service architecture.
flowchart TD A[PDF Documents] --> D[Unified Extraction Service] B[Audio Files] --> D C[Video Footage] --> D D --> E[Structured JSON Output] E --> F[Search Index]Based on the architecture above, which Azure AI capability is designed to handle this multimodal extraction workload?
Show answer & explanation
Correct answer: C
Azure Content Understanding is a multimodal service designed to extract structured information, entities, and summaries from a variety of unstructured formats, including documents, images, audio, and video. While Azure AI Language handles text and Azure AI Vision handles images, Content Understanding unifies these capabilities into a single extraction pipeline.
Ready for the real thing?
The full AI-901 simulator has every exam-style question, timed mode, and instant scoring.