Microsoft AI-103: Developing AI Apps and Agents on Azure

Last checked October 22, 2026 against the official Microsoft exam page.

1. Question 1Build agents and tool use

   An agent must look up order status by calling an internal REST API. Which mechanism lets the model decide when to call that API and with which arguments?

   1. ALengthening the system prompt with the full REST API documentation and example payloads
   2. BStoring the order table as embeddings in a vector index for lookups
   3. CTool or function calling with a JSON schema that describes the API parameters
   4. DFine-tuning the model on past order records and support transcripts
   Show answer and explanationHide answer

   Correct answer: C

   Function calling exposes a named tool with a typed parameter schema, and the model returns structured arguments that the application then executes. Pasting documentation into the prompt or fine-tuning does not give the runtime a reliable, callable interface.

2. Question 2Retrieval-augmented generation

   A RAG app retrieves documents by keyword and returns exact product codes, but misses answers that paraphrase the question. The team wants one query that combines keyword matching with semantic similarity. Which Azure AI Search capability fits?

   1. AFaceted navigation over non-searchable field
   2. BAutocomplete with a suggester on the title field
   3. CHybrid search combining keyword and vector queries
   4. DA wildcard filter on the content field that matches prefixes
   Show answer and explanationHide answer

   Correct answer: C

   Hybrid search runs a full-text query and a vector query together and merges the ranked results, so exact codes and paraphrased questions both match. Facets and suggesters do not perform semantic retrieval, and a wildcard filter does not rank by meaning.

3. Question 3Model integration and deployment

   A team wants the lowest-latency responses for simple classification at high volume, and quality needs are modest. Which deployment choice fits best?

   1. AA nightly batch job that processes requests once a day
   2. BA smaller, faster model deployment sized for the workload
   3. CThe largest reasoning model with the maximum output token limit
   4. DFine-tuning the largest frontier model for each request
   Show answer and explanationHide answer

   Correct answer: B

   A smaller model answers faster and costs less per call, which suits a narrow classification task. Large reasoning models add latency, and a nightly batch cannot serve interactive requests.

4. Question 4Evaluation and quality

   Which metric in a RAG evaluation checks whether the generated answer is supported by the retrieved context?

   1. APerplexity of the generated text
   2. BThroughput in requests per second
   3. CToken count of the prompt
   4. DGroundedness
   Show answer and explanationHide answer

   Correct answer: D

   Groundedness measures whether each claim in the answer can be traced to the retrieved passages. Perplexity measures language-model fluency, and throughput and token counts measure performance and cost.

5. Question 5Plan and manage Azure AI solutions

   A team is building a generative AI app on Azure. They want to keep API keys out of application code and let the app call Azure OpenAI using its Azure identity. Which approach should they use?

   1. AAssign a managed identity to the app and grant it an Azure OpenAI data-plane role
   2. BCreate a long-lived shared access signature for the model endpoint, valid for twelve months and reused
   3. CHard-code the key in plain text inside a configuration file committed to source control
   4. DShare the storage account key through a pipeline variable group that every developer can read
   Show answer and explanationHide answer

   Correct answer: A

   A managed identity lets the app obtain Microsoft Entra ID tokens without storing a secret, and a role assignment limits what it can do. Hard-coded or shared keys leak easily and cannot be scoped to a single workload.