Free AI-103 sample questions with answers
Try these questions before you look at the answers. Each one maps to an official AI-103 skill area.
Question 1. Your agent in Microsoft Foundry must call an Azure AI Search index. Company policy forbids storing keys in application code or configuration. What should you use to authenticate?
- A. An admin key stored in an environment variable
- B. A managed identity with a role assignment on the search service
- C. A query key hard-coded in the agent definition
- D. A shared access signature (SAS) token
Show answer
Answer: B
A managed identity lets the Azure resource authenticate to other Azure services through Microsoft Entra ID without any secret in code or configuration. Grant it the appropriate role on the search service. Every other option relies on a key or token that has to be stored somewhere.
Question 2. Users complain that your RAG chatbot gives confident answers that do not appear in your documents. Which evaluation metric best measures this problem?
- A. Fluency
- B. Relevance
- C. Groundedness
- D. Coherence
Show answer
Answer: C
Groundedness measures whether the answer is supported by the retrieved source content. Relevance measures whether the answer addresses the question, and fluency and coherence measure language quality, not factual support.
Question 3. You need to turn thousands of scanned invoices into structured JSON with fields such as invoice number, date and total, so an agent can reason over them. What is the most suitable approach?
- A. An information extraction pipeline that combines OCR, layout analysis and field extraction
- B. Plain OCR, then asking users to copy the fields
- C. Image captioning on each invoice
- D. Sentiment analysis on the invoice text
Show answer
Answer: A
An information extraction service that combines OCR, layout analysis and field extraction produces structured output directly. Plain OCR gives only text without fields, and image captioning describes pictures rather than extracting data.
Question 4. Your agent must answer questions about internal HR policies that change monthly. Retraining a model every month is not an option. What should you implement?
- A. Fine-tune the model every month
- B. Increase the model’s temperature
- C. Put all policies in the system prompt permanently
- D. Retrieval-augmented generation (RAG) over an index of the policy documents
Show answer
Answer: D
Retrieval-augmented generation retrieves the current policy documents at query time and passes them to the model, so answers stay current without retraining. Fine-tuning bakes knowledge into the model and would need repeating after every change.
Question 5. A voice-enabled agent must understand spoken questions and reply out loud. Which two capabilities do you need?
- A. Translation and speaker recognition
- B. Speech to text and text to speech
- C. Text to speech and translation
- D. Speaker recognition and speech to text
Show answer
Answer: B
The agent needs speech to text to turn the spoken question into text for the model, and text to speech to speak the reply. Translation and speaker recognition are only needed for other requirements.
How did you do?
Five questions only show a direction. The real exam mixes domains and question types under time pressure, so test yourself with a full-length exam before you book.