Domain Weighting & Distribution
This exam set mirrors the live Microsoft AI-200 blueprint: Domain 1 (Containers: ACR, ACA, AKS — 25%), Domain 2 (AI Data Services: Cosmos DB vector indexing, PostgreSQL pgvector, Managed Redis — 25%), Domain 3 (Event-Driven & Messaging: Service Bus, Event Grid, Event Hubs, Functions — 27%), and Domain 4 (Security & Observability: Key Vault, Managed Identity, OpenTelemetry, KQL — 23%).
Domain 1: 12 Qs Domain 2: 12 Qs Domain 3: 13 Qs Domain 4: 11 Qs
Architectural Trade-offs Tested in Set 1
- Container Apps Revision Modes (Single vs. Multiple): Selecting single-revision mode guarantees that 100% of HTTP ingress routes to the latest deployment, whereas multiple-revision mode enables canary splits, blue-green cutovers, and zero-downtime A/B testing across active revisions.
- Cosmos DB Vector Indexing: DiskANN vs. Flat Indexing: DiskANN enables scalable approximate nearest neighbor (ANN) vector search over millions of high-dimensional embeddings with minimal RU overhead and low memory footprints, whereas flat vector indexing executes brute-force cosine distance scans suitable only for small collections under 1,000 vectors.
- Key Vault Access Control: Azure RBAC vs. Vault Access Policies: Microsoft recommends Azure RBAC (Key Vault Secrets User / Crypto Officer) for modern cloud-native architectures to achieve granular, subscription-wide governance without hitting the 1,024 policy-per-vault limitation.
Exam Traps & Common Distractor Pitfalls
- Common Trap: Using ACR Admin credentials to authenticate Azure Kubernetes Service (AKS) or Container Apps.
Engineering Reality: Admin credentials fail enterprise zero-trust audits; AKS and Container Apps must always pull images using Microsoft Entra Managed Identities assigned the AcrPull role. - Common Trap: Selecting cosine distance for normalized embeddings in PostgreSQL pgvector when inner product is available.
Engineering Reality: When vector embeddings are already normalized (length 1.0), the inner product operator (<#>) executes significantly faster than cosine distance (<=>) while producing identical ranking results.
★ Priority Questions (Verified Real Exam Questions):
Questions marked with Priority · Real Exam are verified questions that have previously appeared on live Microsoft AI-200 certification exams. If your preparation time is tight, use the Priority · Real Exam filter below to practice these first.
Study Strategy: Target 2 minutes per question. Treat priority questions as full architectural reviews: verify why the distractor options violate high availability, security standards, or cost efficiency before inspecting the correct shortcut.