AI-200 · Study cards
Every objective, one card each
27 cards, about 20 minutes of reading. Each card explains one objective from the official outline of April 15, 2026, names the trap, and ends with a question.
1 Develop containerized solutions on Azure
20-25%
▾
Implement container application hosting
- ACR images and tags Use unique image tags for deployments, immutable digests for exact versions, and delete images rather than only untagging them.
- ACR Tasks Build container images in Azure on demand or from commit, base-image, and timer triggers, without a local Docker engine.
- App Service containers Run a custom image in App Service, configure its port and environment, and pull from ACR securely with managed identity.
Implement container-orchestrated solutions
- Container Apps revisions Revision-scope changes create a new revision; application-scope settings do not. Use multiple revisions to split traffic safely.
- KEDA scaling Use KEDA scale rules to turn HTTP, TCP, queue, and other event signals into Container Apps replicas, including scale to zero.
- AKS manifests A Deployment runs pods and a Service exposes them. Their labels and ports must line up for traffic to reach the app.
- Container troubleshooting Troubleshoot from the inside out: platform events, app logs, service endpoints, then ingress, DNS, and network connectivity.
2 Develop AI solutions by using Azure data management services
25-30%
▾
Develop AI solutions by using Azure Cosmos DB for NoSQL
- Cosmos SDK queries Use point reads when id and partition key are known; use parameterized queries when you need filtering or multiple items.
- Cosmos indexing and RUs Tune RU cost with selective indexing, composite indexes, and the right consistency level, then verify with request charge and index metrics.
- Cosmos vector search Store embeddings with items, define a vector policy and index, then rank semantic matches with VectorDistance and TOP N.
- Cosmos change feed Process inserts and updates from the change feed, checkpoint progress in leases, and design handlers for at-least-once delivery.
Develop AI solutions by using Azure Database for PostgreSQL
- PostgreSQL from Python Connect with psycopg over SSL, use a fresh Microsoft Entra token as the password, and pass SQL values as parameters.
- Vector schema design Store chunk text, metadata, and a dimensioned embedding together; choose vector, halfvec, or sparsevec to fit the model and storage needs.
- pgvector index tuning Choose exact search, IVFFlat, or HNSW based on recall and latency, then tune query-time settings and verify the plan.
- Vector workload sizing Size compute, maintenance memory, and storage together, with special attention to large index builds and vector I/O.
- PostgreSQL RAG search Filter metadata, rank chunks by vector distance, apply a relevance threshold, and pass only grounded context into the RAG prompt.
- Connection pooling Reuse PostgreSQL connections through an app pool or PgBouncer, and account for transaction-pooling behavior when session state matters.
Integrate Azure Managed Redis in AI solutions
3 Connect to and consume Azure services
20-25%
▾
Develop event- and message-based AI solutions
- Service Bus messaging Use queues for competing consumers, topics for fan-out through subscriptions, and dead-letter queues for messages that cannot be completed.
- Event Grid routing Publish custom events to a topic, filter them in event subscriptions, and plan retry and dead-letter behavior for failed deliveries.
Develop and implement Azure Functions
- Functions triggers and bindings One trigger starts each Azure Function; bindings bring data in or send it out, while SDK clients handle interactions that need more control.
- Functions hosting and deploy Choose the Functions hosting plan for scaling, networking, and control, then separate app-wide, local, and Azure configuration correctly.