Practice Free AI-200 Exam Online Questions
An Azure AI solution must detect the language of text before routing to language-specific models. The text can be very short (under 10 words).
Which Azure AI Language feature is most reliable for short-text language detection?
- A . Azure AI Language language detection API
- B . Use Azure OpenAI with a language detection system prompt
- C . Azure AI Language with confidenceScores threshold filtering
- D . Azure AI Translator detect API
- E . Hardcode language detection rules for known short phrases
An AI workload on AKS needs to access a private Azure Container Registry (ACR). You want to avoid using a fixed username and password for authentication.
What is the most secure method to grant AKS access to ACR?
- A . Use a Shared Access Signature (SAS)
- B . Enable the ‘AcrPull’ role for the AKS cluster’s Managed Identity
- C . Make the ACR public
- D . Store the ACR password in an environment variable
You are deploying a global AI application using Azure Front Door. You want to ensure that users from Europe are routed to a different backend than users from North America.
Which routing method should you use?
- A . Priority-based
- B . Weighted-based
- C . Latency-based
- D . Geographic-based
You are building an AI agent that uses a ‘State Store’. You need to ensure that if two instances of the agent try to update the same user’s state at the exact same time, one is prevented from overwriting the other’s changes.
Which concept handles this?
- A . Atomic Increments
- B . Global Replication
- C . Read-only access
- D . Optimistic Concurrency (ETags)
An enterprise requires that all Azure OpenAI API traffic routes through their private network without traversing the public internet.
Which configuration achieves this?
- A . Use Azure Front Door with WAF policy
- B . Configure an API Management gateway in front of Azure OpenAI
- C . Enable VNet integration on the Azure OpenAI resource
- D . Configure a Private Endpoint for the Azure OpenAI resource
- E . Disable public network access on the Azure OpenAI resource
You are building a pipeline that uses Azure AI Document Intelligence to process receipts and extract totals. The extracted amounts must be validated (total = sum of line items).
Where should validation logic be placed in the architecture?
- A . In the Azure Data Factory transformation activity
- B . In the Azure AI Document Intelligence skill configuration
- C . In the Azure AI Search indexer mapping
- D . In the Azure Logic Apps workflow after the Document Intelligence step
- E . In a post-processing Azure Function after extraction
An AI solution on Azure Container Apps needs to share a set of configuration files that are updated frequently. You want to avoid rebuilding the container image for every config change.
How should you provide these files?
- A . Mount an Azure Files share to the Container App
- B . Include them in the Docker image
- C . Use a Git-based CI/CD pipeline
- D . Paste them into the ‘Startup’ command
Your company has deployed an interactive voice response (IVR) chatbot using Azure AI Speech. Users complain that when they speak alphanumeric account numbers (e.g., ‘A73-9X’), the system often misinterprets individual letters as full words (e.g., transcribing ‘A’ as ‘Eh’ or ‘Hey’). You need to drastically improve transcription accuracy for these specific account number strings without rewriting the core application.
Which two actions should you take? (Select TWO)
- A . Deploy an Azure AI Document Intelligence Layout model to analyze the telephone audio waveforms.
- B . Configure an Azure AI Translator glossary to map incorrect phonetic text strings to the correct characters.
- C . Develop a Custom Speech model trained on audio recordings and matching text transcripts that reflect the account number formatting.
- D . Create a Phrase List containing common variations and examples of your account number formats and associate it with the SpeechRecognizer.
- E . Increase the max_tokens parameter within the Azure OpenAI endpoint query configuration.
A company is implementing a RAG solution and notices that retrieved chunks are semantically relevant but the final generated answer contains information not present in any retrieved chunk (hallucination).
Which two mitigations are most effective?
- A . Improve the system prompt to instruct the model to answer only from provided context
- B . Add a groundedness evaluation step using Azure AI Studio evaluators
- C . Switch from GPT-4 to a fine-tuned model
- D . Increase the number of retrieved chunks (top-k)
- E . Use a lower temperature setting
You are deploying a global AI application using Azure Front Door. You want to protect the application from ‘Prompt Injection’ attacks and common web vulnerabilities at the edge.
Which feature should you enable?
- A . Azure Bastion
- B . Network Security Group (NSG)
- C . Web Application Firewall (WAF)
- D . Azure Firewall