AI-200 Study Plan: Pass in 30 Days (Day-by-Day Schedule)

Before you start

Two things materially improve the outcome and both take under ten minutes.

• Book the exam now, dated 30 days out. An unbooked exam slips indefinitely. A booked one produces the deadline pressure this plan depends on.
• Create a free Azure account. Every hands-on task in this plan fits inside the free tier or the initial credit.

Time allocation by domain

Study time should mirror exam weight, not personal interest. Domain 2 carries the most marks and contains the material least likely to be covered by prior experience.

Week 1 — Containers (days 1–7)

Days 1–2: Azure Container Registry

• Read the ACR module
• Understand tags versus digests, and why digests are the immutable reference
• ACR Tasks: base image update triggers, scheduled tasks, multi-step YAML
• Authentication: managed identity with the AcrPull role, not the admin user
• Lab: run az acr build to build and push an image without a local Docker daemon

Days 3–4: Azure Container Apps

• Read the Container Apps module
• Revisions, revision modes, and traffic splitting for canary releases
• KEDA scaling rules: HTTP concurrency, CPU, and queue-based scalers
• Scale to zero and the resulting cold start trade-off
• Internal versus external ingress
• Lab: deploy an app, then split traffic 80/20 across two revisions

Days 5–7: Azure Kubernetes Service

• Read the AKS module
• Network plugins: kubenet, Azure CNI, and CNI Overlay, and the IP consumption of each
• Workload Identity and the OIDC federation model
• Node pools: system versus user, GPU pools, and the cluster autoscaler
• Troubleshooting: kubectl logs --previous, kubectl describe pod, CrashLoopBackOff diagnosis
• Lab: deploy a manifest to AKS and deliberately break it to read the events

Week 2 — AI data services (days 8–16)

This is the heaviest domain and the material least likely to be covered by AZ-204 experience. Do not compress it.

Days 8–11: Cosmos DB

• Read the Cosmos DB module
• Partition key selection, and why it is immutable after container creation
• Request Units: what drives cost, and why writes are roughly five times reads
• The five consistency levels and the guarantee each provides
• Vector search: the vector policy, DiskANN, and VectorDistance() in queries
• Change feed: captures inserts and updates, never deletes
• Lab: create a container with a vector policy and run a similarity query

Days 12–14: PostgreSQL and pgvector

• Read the PostgreSQL module
• Flexible Server tiers and high availability modes: same-zone versus zone-redundant
• Enabling the vector extension via the allowlist parameter
• IVFFlat versus HNSW: build cost, memory, recall, and tuning parameters
• Distance operators: <-> for L2, <=> for cosine, <#> for inner product
• Hybrid search combining tsvector full-text with vector similarity
• Lab: create an HNSW index and compare query plans before and after

Days 15–16: Azure Managed Redis

• Read the Redis module
• Tier differences, and that vector search requires Enterprise with RediSearch
• Cache-aside pattern, TTL, and invalidation strategy
• Semantic caching for LLM responses
• Rate limiting with sorted sets in a sliding window

Week 3 — Messaging, security, monitoring (days 17–24)

Days 17–19: Event-driven services

• Read the backend services module
• Service Bus: PeekLock versus ReceiveAndDelete, dead-letter causes, sessions, topics with filters
• Event Grid: retry with exponential backoff, dead-lettering to Storage, handler idempotency
• Event Hubs: partitions, consumer groups, Capture, checkpointing
• Azure Functions: triggers, bindings, hosting plans, and cold start behaviour
• Read the messaging comparison guide

Days 20–21: Secrets and configuration

• Read the secrets and configuration module
• Key Vault: RBAC versus access policies, soft delete, purge protection
• Managed identity: system-assigned versus user-assigned, and when each applies
• App Configuration: labels, feature flags, and Key Vault references
• RBAC: Owner can assign roles, Contributor cannot

Days 22–24: Observability

• Read the observability module
• OpenTelemetry, distributed tracing, and correlated TraceId across services
• KQL: where, summarize, project, extend, bin, top, join
• Application Insights tables: requests, exceptions, dependencies, traces
• Container Insights tables: KubePodInventory, Perf, ContainerLog
• Lab: write five KQL queries by hand against a Log Analytics workspace
• Work through the KQL cheatsheet

Week 4 — Consolidation (days 25–30)

Days 25–26: Full-length mocks

• Take the timed mock exam under exam conditions with no notes
• Review every incorrect answer and write down the specific reason it was wrong
• Re-read the module for any domain scoring below 75%

Days 27–28: Decision matrix and question volume

• Drill the service decision matrix until the mapping is automatic
• Work through the question bank for your two weakest modules
• Retake the mock exam; target above 85%

Day 29: Gotchas and cheatsheets

• Read the common gotchas list in full
• Review the cheatsheets and the memory aids
• No new material from this point

Day 30: Exam day

• Re-read the gotchas list once in the morning; nothing else
• Read every case study scenario fully before looking at its questions
• Flag and move on rather than stalling; 60 questions in 100 minutes allows roughly 100 seconds each
• Eliminate two options first — the remaining choice is usually decided by one constraint in the scenario

If you have less than 30 days

Progress checkpoints

Use these as go/no-go gates rather than as encouragement.

• End of week 1: 80% on Domain 1 questions
• End of week 2: 75% on Domain 2 questions
• End of week 3: 80% on Domains 3 and 4 questions
• Day 26: 75% or better on a full timed mock
• Day 28: 85% or better on a full timed mock

Missing the day 28 gate is a reasonable signal to move the exam back a week rather than sit it and hope.