AI-200 Sample Questions

AI-200 Sample Questions & Answers

Building AI solutions on Azure data services such as Managed Redis, PostgreSQL and Cosmos DB takes the biggest share, alongside hosting and orchestrating containers, connecting to event- or message-based services and Functions, and securing and troubleshooting it.

Launch the full AI-200 simulator →

Showing 6 of 12 free samples.

  1. Question 1Advanced

    Develop containerized solutions on Azure · Deploy applications to Azure Container Apps, including environment configuration and revision management

    Contoso Financial Services runs an intelligent document processing system deployed on Azure Container Apps. The system utilizes multiple microservices, with the core extraction engine running as a containerized Python workload.

    The extraction engine is currently deployed in single revision mode. During peak business hours, the container app maintains 12 active replicas to process incoming loan underwriting documents. The development team has committed an updated revision containing performance optimizations and an updated vector retrieval library.

    The deployment of the new revision must achieve zero downtime. Active incoming customer traffic must not experience dropped connections or HTTP 503 Service Unavailable errors. Traffic must transition to the new revision only when the new revision is fully capable of handling production load.

    Under Azure Container Apps single revision mode, what specific criteria must be met before Azure Container Apps deprovisions the previous revision and routes 100% of traffic to the newly deployed revision?

    Show answer & explanation

    Correct answer: A

    In Azure Container Apps single revision mode, zero-downtime deployment is achieved because traffic automatically switches to the new revision only after three conditions are satisfied: 1) provisioning succeeds, 2) the replica count of the new revision scales to match the replica count of the previously active revision (subject to minReplicas and maxReplicas constraints), and 3) startup and readiness probes pass successfully. Once ready, traffic shifts 100% to the new revision and the old revision is deactivated and deprovisioned.

  2. Question 2Beginner

    Develop containerized solutions on Azure · Deploy applications to Azure Container Apps, including environment configuration and revision management

    A systems administrator needs to configure revision labels for an Azure Container App in multiple revision mode. According to Azure Container Apps naming rules, which of the following is a valid revision label?

    Show answer & explanation

    Correct answer: D

    In Azure Container Apps, revision labels must consist of lowercase alphanumeric characters or hyphens (-), must start and end with an alphanumeric character, cannot exceed 64 characters, and cannot include consecutive hyphens (--). Therefore, qa-test-v2 is valid. -staging-release is invalid because it begins with a hyphen, prod--preview is invalid because of consecutive hyphens, and Feature_Experiment_1 is invalid because it contains uppercase letters and underscores.

  3. Question 3Intermediate

    Develop containerized solutions on Azure · Implement event-driven scaling by using Kubernetes Event‑driven Autoscaling (KEDA) in Container Apps

    An engineer is configuring an HTTP scale rule for a conversational AI backend hosted on Azure Container Apps. The container app currently has minReplicas set to 0, maxReplicas set to 15, and --scale-rule-http-concurrency configured to 25. During an idle period after high traffic, incoming requests cease entirely. The team notices that the app takes exactly 300 seconds to scale from 1 replica down to 0 replicas, whereas earlier reductions between 10 replicas and 4 replicas occurred much faster. Why does this behavior occur?

    Show answer & explanation

    Correct answer: D

    In Azure Container Apps KEDA-based scaling, the cooldownPeriod (which defaults to 300 seconds) applies strictly when scaling from the last active replica down to 0 replicas. Scale-down operations between non-zero replica counts (such as from 10 down to 4) do not wait for the cool down period; they scale down based on the scale-down stabilization window and current metric evaluations.

  4. Question 4Advanced

    Develop containerized solutions on Azure · Implement event-driven scaling by using Kubernetes Event‑driven Autoscaling (KEDA) in Container Apps

    A batch AI image-embedding service runs on Azure Container Apps and scales using a KEDA custom scale rule connected to an Azure Storage queue. The scale rule is configured with queueLength=10, min-replicas=0, and max-replicas=40. Currently, the container app has 0 running replicas. Suddenly, an upstream producer publishes 85 messages into the queue. How many replicas will Container Apps scale up to during its initial scale step, and what is the final calculated target replica count?

    Show answer & explanation

    Correct answer: C

    The target replica count formula in Azure Container Apps is desiredReplicas = ceil(currentMetricValue / targetMetricValue) = ceil(85 / 10) = 9 replicas. When scaling up from 0 replicas, Azure Container Apps scales up in geometric steps: 1, 4, 8, 16, 32, ... up to the max replica count. Therefore, the very first step from 0 brings the container app to 1 replica before progressing toward the target of 9 replicas.

  5. Question 5Intermediate

    Develop AI solutions by using Azure data management services · Store and retrieve embeddings and execute vector similarity search for semantic retrieval

    A developer is designing a semantic document retrieval container in Azure Cosmos DB for NoSQL. The documents contain 1,536-dimensional embeddings generated by the text-embedding-3-small model. When defining the vector index policy, the developer attempts to configure an index of type flat. What issue will occur, and what is the appropriate resolution?

    Show answer & explanation

    Correct answer: A

    In Azure Cosmos DB for NoSQL vector search, the flat index type performs an exact brute-force scan and is limited to a maximum of 505 dimensions. For high-dimensional embeddings (such as 1,536-dimensional vectors from OpenAI models), developers must use either quantizedFlat or diskANN, both of which support up to 4,096 dimensions.

  6. Question 6Advanced

    Develop AI solutions by using Azure data management services · Store and retrieve embeddings and execute vector similarity search for semantic retrieval

    A lead AI engineer creates a new container in Azure Cosmos DB for NoSQL with a vector index policy using quantizedFlat to store 1,536-dimensional vectors. During early development, the developer seeds 450 document vectors into the container and runs semantic similarity queries using VectorDistance(c.embedding, @queryVector). How does Cosmos DB execute this query given the current data volume?

    Show answer & explanation

    Correct answer: D

    In Azure Cosmos DB for NoSQL, the quantizedFlat and diskANN vector index types require a minimum of 1,000 vectors to train the vector quantization dictionary. If the container contains fewer than 1,000 vectors, the indexing engine automatically falls back to full brute-force scans (acting like a flat index) without failing the queries.

Ready for the real thing?

The full AI-200 simulator has every exam-style question, timed mode, and instant scoring.

Go to the AI-200 simulator →