AI-102 Practice Questions
Prepare for AI-102 with more than an answer.
Unlock the full exam and previous versions
- v1Version 1 251 questions Current
- AI-100Legacy Designing and Implementing an Azure AI Solution 50 questions Locked
- Exam fee
- $165 USD
- Level
- Associate
- Valid for
- 1 year
Domains covered on the exam 6
- Plan and manage an Azure AI solution22.5%
- Implement generative AI solutions17.5%
- Implement an agentic solution7.5%
- Implement computer vision solutions12.5%
- Implement natural language processing solutions17.5%
- Implement knowledge mining and information extraction solutions17.5%
- 1
True or False: When using the Semantic Kernel SDK, a 'Plugin' is strictly limited to native code functions (C#/Python) and cannot include prompts defined in text files.
Show answer details
Correct answer: B
False. In Semantic Kernel, a Plugin can contain both 'native functions' (code) and 'semantic functions' (prompts defined in text/configuration files). Both are treated as functions that the kernel can orchestrate.
- 2
You are building a custom neural voice for a brand mascot using Azure AI Speech. You have recorded 500 utterances with a professional voice actor. Before you can train the model, what mandatory file must you upload to the Speech Studio to satisfy Microsoft's responsible AI requirements?
Show answer details
Correct answer: D
To prevent deepfakes and unauthorized voice cloning, Microsoft requires a 'Voice Talent Verbal Consent' recording. This is a specific audio file where the voice actor reads a predefined statement acknowledging that their voice will be used to create a synthetic voice model.
- 3
You need to export data from an Azure AI Search skillset to be visualized in Power BI. You want to store complex hierarchical JSON data extracted from documents into a format that Power BI can easily ingest as related tables. Which type of Knowledge Store projection should you configure?
Show answer details
Correct answer: C
Table projections in Knowledge Store flatten the enriched document hierarchy into Azure Table Storage tables. This relational format allows Power BI to easily import the data and establish relationships between the tables using the generated partition and row keys.
- 4
You are fine-tuning a GPT-4 model using Azure OpenAI. You have prepared your training data in the JSONL format. Which of the following is a strict requirement for the 'messages' array in each training example?
Show answer details
Correct answer: C
For chat completion fine-tuning, the model learns to generate the response. Therefore, the final message in every training example MUST be the target completion, which corresponds to the 'assistant' role. If the last message is from the 'user', there is nothing for the model to learn to generate.
- 5
You are implementing a Conversational Language Understanding (CLU) solution. You have a requirement to extract a list of product codes that follow a specific pattern (e.g., 'PROD-1234') but the prefix might vary (e.g., 'ITEM-5678', 'SKU-9999'). You do not have a finite list of all possible codes. Which entity component is best suited for this task?
Show answer details
Correct answer: A
In CLU, the 'Learned' component is the primary way to extract entities based on context. By tagging examples of 'PROD-1234', 'ITEM-5678', etc., the model learns the context and structure of the product codes to extract new, unseen ones.
- 6
You need to build a chatbot that meets the following requirements:✑ Supports chit-chat, knowledge base, and multilingual models✑ Performs sentiment analysis on user messages✑ Selects the best language model automaticallyWhat should you integrate into the chatbot?
Show answer details
Correct answer: B
- 7
Your company wants to reduce how long it takes for employees to log receipts in expense reports. All the receipts are in English. You need to extract top-level information from the receipts, such as the vendor and the transaction total. The solution must minimize development effort. Which Azure service should you use?
Show answer details
Correct answer: A
Form Recognizer, now documented as Azure AI Document Intelligence (its old receipt documentation URL redirects there), has a prebuilt receipt model that uses OCR and deep learning to extract key fields from sales receipts, such as merchant name, transaction date, tax and transaction total, and returns structured JSON. Because the model is prebuilt, no training or custom parsing code is needed, which meets the minimal-development requirement. Computer Vision OCR (Read) returns raw text only, so you would still have to write code to find the vendor and total; Custom Vision trains image classification and object detection models, not field extraction; and Personalizer chooses actions with reinforcement learning and is being retired. Sources: Microsoft Learn, 'Document Intelligence receipt model'; 'What is Azure AI Vision?'; 'What is Personalizer?'.
- 8
You are developing a new sales system that will process the video and text from a public-facing website.You plan to monitor the sales system to ensure that it provides equitable results regardless of the user's location or background.Which two responsible AI principles provide guidance to meet the monitoring requirements? Each correct answer presents part of the solution.NOTE: Each correct selection is worth one point.
Show answer details
Correct answer: A, E
