AI-103 Practice Test

Answer choices

1. A. VisualFeatures.CAPTION, VisualFeatures.OBJECTS, and VisualFeatures.COLOR
2. B. VisualFeatures.ALL
3. C. VisualFeatures.TEXT, VisualFeatures.TAGS, and VisualFeatures.PEOPLE
4. D. VisualFeatures.OCR, VisualFeatures.DESCRIPTIONS, and VisualFeatures.PALETTE

Wrong-answer review

• B. VisualFeatures.ALL: VisualFeatures.ALL is incorrect because requesting all features runs unnecessary classification models (like Face or OCR), which increases API latency and latency costs.
• C. VisualFeatures.TEXT, VisualFeatures.TAGS, and VisualFeatures.PEOPLE: VisualFeatures.TEXT, VisualFeatures.TAGS, and VisualFeatures.PEOPLE is incorrect because this requests OCR, tags, and people detection, failing to retrieve object bounding boxes, dominant colors, or image captions.
• D. VisualFeatures.OCR, VisualFeatures.DESCRIPTIONS, and VisualFeatures.PALETTE: VisualFeatures.OCR, VisualFeatures.DESCRIPTIONS, and VisualFeatures.PALETTE is incorrect because it contains obsolete or hallucinated enum values (in v4.0, descriptions is replaced by caption, and palette is color).

Answer choices

1. A. Prebuilt Layout model (`prebuilt-layout`), because it classifies identity documents.
2. B. Prebuilt ID document model (`prebuilt-idDocument`), because it is pre-trained to extract identity fields from passports and driver's licenses.
3. C. Prebuilt Read model (`prebuilt-read`), because passport fonts are standardized for OCR.
4. D. Custom Template model, trained on five sample passports.

Wrong-answer review

• A. Prebuilt Layout model (`prebuilt-layout`), because it classifies identity documents.: Prebuilt Layout model (`prebuilt-layout`), because it classifies identity documents. is incorrect because the Layout model does not perform semantic classification of identity fields; it only outputs raw layout coordinates and text blocks.
• C. Prebuilt Read model (`prebuilt-read`), because passport fonts are standardized for OCR.: Prebuilt Read model (`prebuilt-read`), because passport fonts are standardized for OCR. is incorrect because the Read model only extracts raw text strings and coordinates. It does not map them to identity fields (e.g. FirstName, PassportNumber) natively.
• D. Custom Template model, trained on five sample passports.: Custom Template model, trained on five sample passports. is incorrect because training a custom model is redundant since the prebuilt ID document model already handles passports and driver's licenses natively.

Answer choices

1. A. The API will ignore the unsupported parameters and process the request successfully using the model's default settings.
2. B. The API will return a 400 Validation Error, because reasoning models require temperature, top_p, and presence_penalty to remain at their default values (typically 1.0 or 0.0) and setting them otherwise is unsupported.
3. C. The API will process the parameters normally, dynamically scaling the reasoning token length based on the temperature.
4. D. The API will return a 401 Unauthorized Error, since custom parameters are only allowed on dedicated provisioned throughput units (PTU) deployments.

Wrong-answer review

• A. The API will ignore the unsupported parameters and process the request successfully using the model's default settings.: The API will ignore the unsupported parameters and process the request successfully using the model's default settings. is incorrect because the API does not silently ignore these parameters; it strictly validates the request payload and returns a validation error.
• C. The API will process the parameters normally, dynamically scaling the reasoning token length based on the temperature.: The API will process the parameters normally, dynamically scaling the reasoning token length based on the temperature. is incorrect because these parameters are not processed; reasoning models do not support sampling temperature variations or penalty controls in this manner.
• D. The API will return a 401 Unauthorized Error, since custom parameters are only allowed on dedicated provisioned throughput units (PTU) deployments.: The API will return a 401 Unauthorized Error, since custom parameters are only allowed on dedicated provisioned throughput units (PTU) deployments. is incorrect because unauthorized errors are caused by missing or incorrect credentials (keys/tokens), not by passing invalid query parameters.

Answer choices

1. A. ```python def get_order_status(order_id: str) -> str: # No authentication or payload details provided return f"Order status for {order_id} retrieved."
2. B. ```python def get_order_status(order_id: str) -> str: import requests headers = {"Authorization": "Bearer YOUR_API_KEY"} payload = {"orderId": order_id} response = requests.post("https://legacy.example.com/orderstatus", headers=headers, json=payload) return response.text ```
3. C. ```python def get_order_status(order_id: str) -> str: import requests headers = {"X-Auth-Token": "YOUR_TOKEN"} payload = {"order_id": order_id} response = requests.get("https://legacy.example.com/orderstatus", headers=headers, params=payload) return response.text ```
4. D. ```python def get_order_status(order_id: str) -> str: import requests headers = {"Authentication": "Basic YOUR_CREDENTIALS"} payload = {"orderId": order_id} response = requests.post("https://legacy.example.com/orderstatus", headers=headers, data=payload) return response.text ```

Wrong-answer review

• A. ```python def get_order_status(order_id: str) -> str: # No authentication or payload details provided return f"Order status for {order_id} retrieved.": This distractor describes the idea that ```python def get_order_status(order_id: str) -> str: # No authentication or payload details provided return f"Order status for {order_id} retrieved.". In "Your company is developing an Azure AI agent to automate customer service inquiries. The agent needs to retrieve order status information from a legacy internal system. This system exposes a REST API that requires specific authentication headers and expects a JSON payload containing the order ID. You are using the Azure OpenAI Python SDK to define a tool for this API call. Which of the following code snippets correctly defines a tool function that includes the necessary authentication and payload details, ensuring the agent can reliably interact with the legacy system?", that misses the required action because the correct response is "```python def get_order_status(order_id: str) -> str: import requests headers = {"Authorization": "Bearer YOUR_API_KEY"} payload = {"orderId": order_id} response = requests.post("https://legacy.example.com/orderstatus", headers=headers, json=payload) return response.text ```". On the job, mixing up that distractor with "```python def get_order_status(order_id: str) -> str: import requests headers = {"Authorization": "Bearer YOUR_API_KEY"} payload = {"orderId": order_id} response = requests.post("https://legacy.example.com/orderstatus", headers=headers, json=payload) return response.text ```" can lead to the wrong implement generative ai and agentic solutions action or troubleshooting path.
• C. ```python def get_order_status(order_id: str) -> str: import requests headers = {"X-Auth-Token": "YOUR_TOKEN"} payload = {"order_id": order_id} response = requests.get("https://legacy.example.com/orderstatus", headers=headers, params=payload) return response.text ```: This distractor describes the idea that ```python def get_order_status(order_id: str) -> str: import requests headers = {"X-Auth-Token": "YOUR_TOKEN"} payload = {"order_id": order_id} response = requests.get("https://legacy.example.com/orderstatus", headers=headers, params=payload) return response.text ```. In "Your company is developing an Azure AI agent to automate customer service inquiries. The agent needs to retrieve order status information from a legacy internal system. This system exposes a REST API that requires specific authentication headers and expects a JSON payload containing the order ID. You are using the Azure OpenAI Python SDK to define a tool for this API call. Which of the following code snippets correctly defines a tool function that includes the necessary authentication and payload details, ensuring the agent can reliably interact with the legacy system?", that misses the required action because the correct response is "```python def get_order_status(order_id: str) -> str: import requests headers = {"Authorization": "Bearer YOUR_API_KEY"} payload = {"orderId": order_id} response = requests.post("https://legacy.example.com/orderstatus", headers=headers, json=payload) return response.text ```". On the job, mixing up that distractor with "```python def get_order_status(order_id: str) -> str: import requests headers = {"Authorization": "Bearer YOUR_API_KEY"} payload = {"orderId": order_id} response = requests.post("https://legacy.example.com/orderstatus", headers=headers, json=payload) return response.text ```" can lead to the wrong implement generative ai and agentic solutions action or troubleshooting path.
• D. ```python def get_order_status(order_id: str) -> str: import requests headers = {"Authentication": "Basic YOUR_CREDENTIALS"} payload = {"orderId": order_id} response = requests.post("https://legacy.example.com/orderstatus", headers=headers, data=payload) return response.text ```: This distractor describes the idea that ```python def get_order_status(order_id: str) -> str: import requests headers = {"Authentication": "Basic YOUR_CREDENTIALS"} payload = {"orderId": order_id} response = requests.post("https://legacy.example.com/orderstatus", headers=headers, data=payload) return response.text ```. In "Your company is developing an Azure AI agent to automate customer service inquiries. The agent needs to retrieve order status information from a legacy internal system. This system exposes a REST API that requires specific authentication headers and expects a JSON payload containing the order ID. You are using the Azure OpenAI Python SDK to define a tool for this API call. Which of the following code snippets correctly defines a tool function that includes the necessary authentication and payload details, ensuring the agent can reliably interact with the legacy system?", that misses the required action because the correct response is "```python def get_order_status(order_id: str) -> str: import requests headers = {"Authorization": "Bearer YOUR_API_KEY"} payload = {"orderId": order_id} response = requests.post("https://legacy.example.com/orderstatus", headers=headers, json=payload) return response.text ```". On the job, mixing up that distractor with "```python def get_order_status(order_id: str) -> str: import requests headers = {"Authorization": "Bearer YOUR_API_KEY"} payload = {"orderId": order_id} response = requests.post("https://legacy.example.com/orderstatus", headers=headers, json=payload) return response.text ```" can lead to the wrong implement generative ai and agentic solutions action or troubleshooting path.

Answer choices

1. A. Azure AI Vision's Computer Vision service, utilizing its image analysis features for object detection and captioning.
2. B. Azure OpenAI Service's Visual Analysis capabilities, leveraging multimodal models to understand image content and generate detailed descriptions with contextual reasoning.
3. C. Azure AI Vision's Custom Vision service, trained on a dataset of plant diseases and nutrient deficiencies.
4. D. Azure AI Vision's Read API, to extract text from signs and labels near the plants to identify potential causes.

Wrong-answer review

• A. Azure AI Vision's Computer Vision service, utilizing its image analysis features for object detection and captioning.: Azure AI Vision's Computer Vision service, utilizing its image analysis features for object detection and captioning. is incorrect because While Azure AI Vision's Computer Vision service is powerful for basic image analysis, it lacks the multimodal reasoning capabilities needed to understand the context and provide detailed, actionable insights. It would struggle to connect visual observations with potential causes and recommendations.
• C. Azure AI Vision's Custom Vision service, trained on a dataset of plant diseases and nutrient deficiencies.: Azure AI Vision's Custom Vision service, trained on a dataset of plant diseases and nutrient deficiencies. is incorrect because Custom Vision is excellent for building custom image classifiers, but it requires a significant investment in training data and ongoing maintenance. It's not the ideal solution when a pre-trained model with multimodal reasoning capabilities is available and better suited for the task.
• D. Azure AI Vision's Read API, to extract text from signs and labels near the plants to identify potential causes.: This distractor describes the idea that Azure AI Vision's Read API, to extract text from signs and labels near the plants to identify potential causes. In "Your company, 'EcoHarvest,' is developing a mobile application to help farmers identify plant diseases and nutrient deficiencies in their crops. The app needs to analyze images captured by farmers' smartphones, providing detailed descriptions of the observed issues, including identifying specific disease types, suggesting potential causes (e.g., fungal infection, lack of nitrogen), and recommending corrective actions. The application also needs to be able to understand the context of the image – for example, distinguishing between a healthy leaf and a diseased leaf on the same plant. Which Azure service is the most appropriate choice for this scenario?", that misses the required action because the correct response is "Azure OpenAI Service's Visual Analysis capabilities, leveraging multimodal models to understand image content and generate detailed descriptions with contextual reasoning.". On the job, mixing up that distractor with "Azure OpenAI Service's Visual Analysis capabilities, leveraging multimodal models to understand image content and generate detailed descriptions with contextual reasoning." can lead to the wrong implement computer vision solutions action or troubleshooting path.

Answer choices

1. A. Send a synchronous HTTP POST request to the endpoint and retrieve the extracted table and text data directly in the response body.
2. B. Call the `begin_analyze_document` method (POST) which starts an asynchronous operation, retrieve the operation ID from the response header, and then poll the `get_analyze_document_result` method (GET) using the operation ID until it succeeds to retrieve the data.
3. C. Call the `analyze_document_sync` method, passing the file stream, and wait for the thread to block until the analysis finishes.
4. D. Use the `AnalyzeDocument` method to download the layout model locally, perform the extraction in the application container, and upload the output JSON.

Wrong-answer review

• A. Send a synchronous HTTP POST request to the endpoint and retrieve the extracted table and text data directly in the response body.: Send a synchronous HTTP POST request to the endpoint and retrieve the extracted table and text data directly in the response body. is incorrect because document analysis can take time and is therefore handled asynchronously by the API; you cannot retrieve results directly in the initial POST response.
• C. Call the `analyze_document_sync` method, passing the file stream, and wait for the thread to block until the analysis finishes.: Call the `analyze_document_sync` method, passing the file stream, and wait for the thread to block until the analysis finishes. is incorrect because the SDK does not provide a synchronous blocking method for document analysis due to execution time limits.
• D. Use the `AnalyzeDocument` method to download the layout model locally, perform the extraction in the application container, and upload the output JSON.: Use the `AnalyzeDocument` method to download the layout model locally, perform the extraction in the application container, and upload the output JSON. is incorrect because Document Intelligence models are hosted as cloud services (or in specific Docker containers) and cannot be downloaded dynamically to a client application for local execution via a standard API call.

Answer choices

1. A. Disable the default content filters and write custom regular expression post-processing scripts in the application code to filter out sensitive terms.
2. B. Create a custom content filtering policy in Azure AI Foundry, configure custom blocklists within Azure AI Content Safety to match the prohibited terms, and apply the policy to the model deployment.
3. C. Call the OpenAI Moderation API via a custom application middleware component before sending prompts to the Azure OpenAI Service endpoint.
4. D. Deploy a separate, fine-tuned GPT-4 model specifically trained to moderate financial queries, and route all prompts through it prior to the main model.

Wrong-answer review

• A. Disable the default content filters and write custom regular expression post-processing scripts in the application code to filter out sensitive terms.: Disable the default content filters and write custom regular expression post-processing scripts in the application code to filter out sensitive terms. is incorrect because disabling default filters increases risk, and application-side regex scripts are difficult to maintain, prone to bypasses, and add unnecessary developer overhead compared to native blocklists.
• C. Call the OpenAI Moderation API via a custom application middleware component before sending prompts to the Azure OpenAI Service endpoint.: Call the OpenAI Moderation API via a custom application middleware component before sending prompts to the Azure OpenAI Service endpoint. is incorrect because the OpenAI Moderation API is a third-party service hosted by OpenAI (not Azure), which violates data residency rules on Azure and increases latency and complexity compared to using native Azure AI Content Safety.
• D. Deploy a separate, fine-tuned GPT-4 model specifically trained to moderate financial queries, and route all prompts through it prior to the main model.: Deploy a separate, fine-tuned GPT-4 model specifically trained to moderate financial queries, and route all prompts through it prior to the main model. is incorrect because deploying an entire GPT-4 model for content moderation is extremely expensive, slow, and complex, especially when Azure provides native, low-latency blocklist features for this purpose.

Answer choices

1. A. Modify the user prompt for each interaction to explicitly state the desired persona and tone.
2. B. Create a custom content filter to block responses that deviate from the defined persona.
3. C. Define a detailed system message within the Azure OpenAI deployment, outlining the persona, tone, and expected behavior of the chatbot.
4. D. Implement a post-processing script that analyzes and rewrites the chatbot's responses to align with the desired persona.

Wrong-answer review

• A. Modify the user prompt for each interaction to explicitly state the desired persona and tone.: Modify the user prompt for each interaction to explicitly state the desired persona and tone. is incorrect because While modifying user prompts can influence the response, it's not a reliable method for consistently enforcing a persona. Users may omit the persona instructions, or the model may prioritize other aspects of the prompt. This approach is prone to inconsistency.
• B. Create a custom content filter to block responses that deviate from the defined persona.: Create a custom content filter to block responses that deviate from the defined persona. is incorrect because Content filters are designed to block inappropriate or harmful content, not to shape the persona of a chatbot. They don't provide the level of control needed to define and maintain a specific conversational style.
• D. Implement a post-processing script that analyzes and rewrites the chatbot's responses to align with the desired persona.: Implement a post-processing script that analyzes and rewrites the chatbot's responses to align with the desired persona. is incorrect because Post-processing scripts add complexity and latency to the chatbot's responses. They are also less reliable than defining the persona directly within the system message, as they rely on analyzing and modifying the model's output, which can be unpredictable.

Answer choices

1. A. Azure AI Vision with custom vision models trained on disease images.
2. B. Azure OpenAI Service with Visual Analysis, leveraging its multimodal capabilities to understand image content and generate descriptive text.
3. C. Azure AI Vision with OCR to extract text from labels and then use a custom logic app to determine the disease.
4. D. Azure AI Vision with the Describe Images feature to generate captions and then manually review the results.

Wrong-answer review

• A. Azure AI Vision with custom vision models trained on disease images.: This distractor describes the idea that Azure AI Vision with custom vision models trained on disease images. In "Your company, 'EcoHarvest,' is developing a mobile application to help farmers identify plant diseases from images taken with their smartphones. The app needs to not only identify the disease but also provide contextual information, such as potential causes (e.g., fungal infection due to humidity) and recommended treatments (e.g., specific fungicide application). The images often include multiple plants and varying lighting conditions. Which Azure AI service is the most appropriate choice for this scenario?", that misses the required action because the correct response is "Azure OpenAI Service with Visual Analysis, leveraging its multimodal capabilities to understand image content and generate descriptive text.". On the job, mixing up that distractor with "Azure OpenAI Service with Visual Analysis, leveraging its multimodal capabilities to understand image content and generate descriptive text." can lead to the wrong implement computer vision solutions action or troubleshooting path.
• C. Azure AI Vision with OCR to extract text from labels and then use a custom logic app to determine the disease.: Azure AI Vision with OCR to extract text from labels and then use a custom logic app to determine the disease. is incorrect because This approach is overly complex and inefficient. OCR would only extract text from labels, not the image itself. Relying on a logic app to determine the disease would be cumbersome and lack the nuanced understanding of a visual analysis model.
• D. Azure AI Vision with the Describe Images feature to generate captions and then manually review the results.: This distractor describes the idea that Azure AI Vision with the Describe Images feature to generate captions and then manually review the results. In "Your company, 'EcoHarvest,' is developing a mobile application to help farmers identify plant diseases from images taken with their smartphones. The app needs to not only identify the disease but also provide contextual information, such as potential causes (e.g., fungal infection due to humidity) and recommended treatments (e.g., specific fungicide application). The images often include multiple plants and varying lighting conditions. Which Azure AI service is the most appropriate choice for this scenario?", that misses the required action because the correct response is "Azure OpenAI Service with Visual Analysis, leveraging its multimodal capabilities to understand image content and generate descriptive text.". On the job, mixing up that distractor with "Azure OpenAI Service with Visual Analysis, leveraging its multimodal capabilities to understand image content and generate descriptive text." can lead to the wrong implement computer vision solutions action or troubleshooting path.

Answer choices

1. A. Custom Neural model (unstructured), because it requires fewer training documents and handles structured data more accurately.
2. B. Custom Template model (structured), because it is optimized for structured documents with highly consistent layouts and can be trained with as few as 5 documents.
3. C. Prebuilt Layout model, configured with custom geometric bounding boxes in the API request payload.
4. D. Prebuilt Invoice model, since it supports training custom text fields.

Wrong-answer review

• A. Custom Neural model (unstructured), because it requires fewer training documents and handles structured data more accurately.: This distractor describes the idea that Custom Neural model (unstructured), because it requires fewer training documents and handles structured data more accurately. In "Your company needs to extract unique customer IDs, service codes, and total fees from a standard company-internal form. The forms are always generated from the same PDF generator, and the layout is 100% consistent across all documents (every field is always in the exact same geometric coordinates on the page). You have a set of 5 sample forms. Which Document Intelligence custom model type is the most appropriate and cost-effective choice?", that misses the required action because the correct response is "Custom Template model (structured), because it is optimized for structured documents with highly consistent layouts and can be trained with as few as 5 documents.". On the job, mixing up that distractor with "Custom Template model (structured), because it is optimized for structured documents with highly consistent layouts and can be trained with as few as 5 documents." can lead to the wrong implement information extraction solutions action or troubleshooting path.
• C. Prebuilt Layout model, configured with custom geometric bounding boxes in the API request payload.: Prebuilt Layout model, configured with custom geometric bounding boxes in the API request payload. is incorrect because the prebuilt Layout model only extracts raw tables and text; it does not accept custom semantic fields or bounding box definitions in the query payload.
• D. Prebuilt Invoice model, since it supports training custom text fields.: Prebuilt Invoice model, since it supports training custom text fields. is incorrect because prebuilt models have fixed schemas and do not support custom training or adding user-defined fields.