ExamSprintLab

AI-200 study plan: pass in 4 weeks

Updated September 27, 2026

Four weeks is realistic for AI-200 if you already write Python against Azure and study 8–10 hours a week. Spend one week per domain, in an order that builds on itself: containers first so you have somewhere to run code, then the data services it talks to, then the messaging that connects the pieces, then security and monitoring across all of it. Without Azure development experience, double the timeline.

Before you start

  • Read the official skills outline once, end to end. It is short.
  • Create a free or pay-as-you-go Azure subscription and set a budget alert. AKS, PostgreSQL and Managed Redis cost money when left running; delete resource groups at the end of each session.
  • Set up Python with the Azure SDKs and the Azure CLI. Almost every lab below uses DefaultAzureCredential, so sign in with az login once and let it pick up your identity.

Week 1: containerized solutions (20–25%)

Goal: build an image, store it, and run it three ways.

  • Azure Container Registry: push images, tag and version them, understand why a digest is more reliable than a tag.
  • ACR Tasks: build in the cloud with az acr build; set up a task that rebuilds on a source commit or base image update.
  • App Service: deploy a container, supply environment variables through app settings, and pull secrets with Key Vault references.
  • Container Apps: environments, revisions, single versus multiple revision mode, traffic splitting between revisions.
  • KEDA scaling: an HTTP rule, then a Service Bus queue-length rule that scales to zero.
  • AKS: deploy with a Deployment and Service manifest, attach the registry, and debug a pod stuck in ImagePullBackOff or CrashLoopBackOff with kubectl describe and kubectl logs.

Lab: one small Python API image, built by ACR Tasks, running in both Container Apps and AKS.

Week 2: data management services (25–30%)

Goal: store embeddings and run similarity search in all three services. This is the largest domain, so it gets the most hours.

  • Cosmos DB for NoSQL: SDK queries, partition keys, Request Units, indexing policies, the five consistency levels, a container vector policy and vector index, VectorDistance queries, and a change feed processor with a lease container.
  • PostgreSQL: enable pgvector, choose column types, build HNSW and IVFFlat indexes and compare their build time and recall, filter vector queries on metadata columns, and use connection pooling with PgBouncer.
  • Azure Managed Redis: cache-aside reads, TTLs, invalidation on write, and a vector index for similarity search.

Lab: load the same few hundred embedded documents into Cosmos DB and PostgreSQL, run the same query against both, and note where each one is simpler.

Week 3: connecting services (20–25%)

Goal: move work off the request path.

  • Service Bus: queues versus topics and subscriptions, subscription filters, peek-lock, max delivery count, and processing the dead-letter queue.
  • Event Grid: custom topics, event subscriptions, subject and advanced filters, retry policy and dead-lettering to Blob Storage.
  • Azure Functions: HTTP, Service Bus and Event Grid triggers; input and output bindings; the Python v2 programming model; hosting plans; deploying and configuring app settings.

Lab: an HTTP-triggered Function that accepts a document, drops a message on a Service Bus queue, and a second Function that embeds it and writes it to Cosmos DB. Force a failure and find the message in the dead-letter queue.

Week 4: secure, monitor and troubleshoot (20–25%) and review

Goal: remove every secret from your code and see every request end to end.

  • Key Vault: access with a managed identity and RBAC, secret versions, rotation, and reacting to near-expiry events.
  • App Configuration: key-values, labels per environment, Key Vault references, and dynamic refresh.
  • OpenTelemetry: instrument the week 3 app with the Azure Monitor OpenTelemetry distro and follow one request across both Functions.
  • KQL: where, summarize, bin, join on operation ID, and render timechart. Write queries until they stop feeling like SQL.

Review: take the free 20-question practice test mid-week. Spend the remaining days on whichever domain scored lowest, using its guide in this section.

If you held AZ-204

Compress weeks 3 and 4 into one. Functions, Service Bus, Event Grid and Key Vault have barely changed. Spend the time you save on week 2, which has almost nothing in common with AZ-204’s storage domain, and on AKS manifests in week 1.

How to know you are ready

  • You can explain when you would pick Cosmos DB, PostgreSQL or Redis for vector search in one sentence each.
  • You can read a KEDA scale rule, a Kubernetes manifest and a KQL query and say what they do.
  • You score 80% or more on realistic practice questions twice in a row, on different question sets.

Then book it. See the exam format for the mechanics and the exam day guide for the day itself.