Training course

This is a free, self-paced reading course for AI-200: Microsoft Azure AI Cloud Developer Associate, the successor to the retiring AZ-204 Azure Developer track, and a rebuild of the associate developer tier around AI-integrated applications rather than generic compute and storage. The distinction shapes everything about how you should study for it. AI-200 assumes you already deploy containers, wire up messaging, and instrument applications; what it tests is whether you make the right call when a foundation model, a vector store, and an event-driven pipeline all have to work together under a production SLA.

The course is organized as one module per official exam domain, in the order Microsoft publishes them, and each module carries the domain’s published weight. That keeps the time you spend proportional to the time the exam spends: AI data management services is the heaviest domain at 28%, and vector-backed storage gets the most room here as a result. Every lesson ends with its key terms and suggested further reading. On an exam this service-dense (diskANN against HNSW, PeekLock against Event Grid delivery semantics, RBAC against vault access policies) the further reading is where you turn a recognized name into a decision you can defend.

AI-200 Associate level 4 modules Domain-weighted Self-paced Free account

What the course covers

Containerized AI solutions on Azure

Module 1 · 23%

Deploying and operating AI workloads across Azure Container Registry, Azure Container Apps, AKS, and App Service. ACR Tasks with base-image triggers and geo-replication; Container Apps multi-revision canaries and KEDA scale-to-zero rules; Azure Workload Identity federated to Kubernetes service accounts; tag immutability; and managed-identity image pulls instead of stored registry credentials.

AI data management services

Module 2 · 28%

The heaviest domain, and the one that most distinguishes AI-200 from the developer exam it replaces. Cosmos DB for NoSQL, session consistency, tenant partitioning, diskANN vector indexes, and the change feed. Azure Database for PostgreSQL Flexible Server with pgvector: HNSW indexes, cosine operators, Memory-Optimized sizing, and PgBouncer connection pooling. And Azure Managed Redis, cache-aside, FT.CREATE VECTOR indexes, and volatile-lru eviction.

Azure services integration

Module 3 · 23%

Wiring AI workloads into the rest of Azure. Service Bus with PeekLock, sessions, correlation filters, and dead-letter queues; Event Grid with at-least-once delivery, subjectBeginsWith filters, and dead-lettering; and Azure Functions: Cosmos DB triggers on the change feed, the Premium plan where you need VNet integration without cold starts, Durable Functions fan-out and fan-in, and output bindings. The domain consistently favors native bindings over hand-written client code, and the questions are built to reward that instinct.

Security, monitoring, and troubleshooting

Module 4 · 25%

Securing and observing what you have built. Key Vault references from App Configuration, SecretNearExpiry rotation through Event Grid, the DefaultAzureCredential chain, Azure RBAC set against vault access policies, soft delete recovery, and sharing a user-assigned managed identity across services. Then Application Insights through the Azure Monitor OpenTelemetry Distro, adaptive sampling, Live Metrics streams, KQL for error triage, and Standard availability tests.

How to use it

Read a module, then take the matching portion of the AI-200 practice exam rather than waiting until you have read everything. On this exam that sequencing matters more than usual, because a large share of the items turn on picking between two services that both work, and the only way to find out whether you can do that is to be asked. Expect the difficulty to sit in module 2’s vector-data questions and module 4’s security and observability judgment calls; the container-lifecycle material in module 1 is comparatively light. The course modules themselves open once you are signed in to a free Certifym account.

Pair the reading with an Azure subscription if you can. Standing up a Container Apps revision, indexing a few thousand embeddings in pgvector, and watching a Cosmos DB change feed trigger a Function does more for retention than any amount of re-reading, and the case-study format rewards people who have seen these services behave. For exam logistics (format, duration, question count, the 700 out of 1000 scaled pass mark, and the blueprint version this material tracks) see the AI-200 certification guide.

AI-200: Microsoft Azure AI Cloud Developer Associate

One module per AI-200 exam domain, weighted the way Microsoft weights them.

A self-paced reading course for AI-200: Microsoft Azure AI Cloud Developer Associate, the associate-level credential for developers who build AI-integrated applications on Azure. The course is organized as one module per official exam domain, in the order Microsoft publishes them, and each module carries the domain's published weight, so the time you spend maps to the time the exam spends.

You work through containerized AI solutions on Azure Container Registry, Container Apps, AKS, and App Service; the AI data layer across Cosmos DB vector search, PostgreSQL Flexible Server with pgvector, and Azure Managed Redis; integration through Service Bus, Event Grid, and Azure Functions; and finally security, monitoring, and troubleshooting with Key Vault, App Configuration, managed identity, and Application Insights. Every lesson ends with its key terms and further reading, so the course works as a reference after the first pass as well as a syllabus during it.

The blueprint assumes you already deploy containers, wire up messaging and serverless services, and instrument applications. This material is written for that reader: it concentrates on the judgment calls between services that both work, which is where the exam does most of its separating. Certifym is an independent study publisher and is not affiliated with, endorsed by, or sponsored by Microsoft Corporation.

Module 1: Containerized AI solutions on Azure 23% of exam

Deploying and operating AI workloads across the four Azure container hosts. You start with the choice itself — Container Apps, AKS, App Service, or none of them — then work through Azure Container Registry: tiers, repositories, ACR Tasks with base-image triggers, geo-replication, tag immutability, and managed-identity image pulls instead of stored registry credentials.The second half covers the hosts themselves. Container Apps environments, revisions and traffic-split canaries, KEDA scale rules and scale-to-zero, secrets and Dapr; AKS with Azure Workload Identity federated to Kubernetes service accounts, GPU node pools, and resource governance; and App Service for Linux containers with plans, slots, and sticky settings.

1.1 Choosing a container host for AI workloads on Azure 5 min Free preview
1.2 Azure Container Registry: tiers, repositories, and image lifecycle 5 min 🔒
1.3 ACR Tasks, base-image triggers, and automated rebuilds 5 min 🔒
1.4 Geo-replication, tag immutability, and registry resilience 5 min 🔒
1.5 Managed-identity image pulls and registry access control 5 min 🔒
1.6 Azure Container Apps: environments, ingress, and the app model 5 min 🔒
1.7 Revisions, traffic splitting, and canary releases 5 min 🔒
1.8 KEDA scale rules and scale-to-zero for inference workloads 5 min 🔒
1.9 Secrets, service discovery, and Dapr in Container Apps 6 min 🔒
1.10 AKS and Azure Workload Identity federation 5 min 🔒
1.11 AKS node pools, GPU scheduling, and resource governance 5 min 🔒
1.12 App Service for containers: plans, slots, and configuration 5 min 🔒

Module 2: AI data management services 29% of exam

The heaviest domain, and the one that most distinguishes AI-200 from the developer exam it replaces. It opens with how the three services divide the work in a retrieval-augmented application, then goes deep on Cosmos DB for NoSQL: request units and throughput modes, partition keys and multitenant partitioning, consistency levels and the session token, vector search with diskANN indexes, the change feed as an event source, and indexing policies that keep RU cost under control.It then covers Azure Database for PostgreSQL Flexible Server with pgvector — Memory-Optimized sizing, distance operators, HNSW against IVFFlat tuning, and PgBouncer connection pooling — and Azure Managed Redis for cache-aside, FT.CREATE VECTOR indexes and semantic caching, and eviction policies such as volatile-lru.

2.1 Choosing an AI data store: vectors, documents, and cache 5 min 🔒
2.2 Cosmos DB for NoSQL: resource model, request units, and throughput 5 min 🔒
2.3 Partition keys and multitenant partitioning strategies 5 min 🔒
2.4 Consistency levels and why session is usually the answer 5 min 🔒
2.5 Vector search in Cosmos DB and diskANN indexes 6 min Free preview
2.6 The Cosmos DB change feed as an event source 5 min 🔒
2.7 Indexing policies, queries, and RU cost control 5 min 🔒
2.8 Azure Database for PostgreSQL Flexible Server: tiers, sizing, and HA 5 min 🔒
2.9 pgvector: types, distance operators, and storing embeddings 6 min 🔒
2.10 Tuning vector indexes: HNSW against IVFFlat 5 min 🔒
2.11 Connection pooling with PgBouncer 5 min 🔒
2.12 Azure Managed Redis and the cache-aside pattern 5 min 🔒
2.13 Redis vector indexes and semantic caching 6 min 🔒
2.14 Eviction policies, persistence, and Redis capacity 6 min 🔒

Module 3: Azure services integration 23% of exam

Wiring AI workloads into the rest of Azure, starting with how to tell messages from events and therefore Service Bus from Event Grid. Service Bus covers PeekLock settlement and lock renewal for slow model calls, sessions and ordering, topics with correlation and SQL filters, and dead-letter queues with a replay path.Event Grid covers delivery semantics and retries, subjectBeginsWith filtering, and dead-lettering to blob storage. Azure Functions covers hosting plans and the Premium plan decision, triggers and input bindings including the Cosmos DB change-feed trigger, output bindings, Durable Functions fan-out and fan-in, and the networking and concurrency settings that keep a function app from overwhelming a rate-limited endpoint. The domain consistently favors native bindings over hand-written client code.

3.1 Messaging, events, and serverless: choosing the integration primitive 5 min 🔒
3.2 Service Bus queues and PeekLock settlement 5 min 🔒
3.3 Sessions, ordering, and stateful message processing 5 min 🔒
3.4 Topics, subscriptions, and correlation filters 5 min 🔒
3.5 Dead-letter queues and poison-message handling 5 min 🔒
3.6 Event Grid delivery, retries, and at-least-once semantics 5 min 🔒
3.7 Event filtering and dead-lettering in Event Grid 5 min 🔒
3.8 Azure Functions hosting plans and the Premium plan decision 5 min 🔒
3.9 Triggers and input bindings, including the Cosmos DB change feed 6 min 🔒
3.10 Output bindings: writing results without client code 5 min 🔒
3.11 Durable Functions: fan-out, fan-in, and orchestration patterns 5 min 🔒
3.12 Networking, cold starts, and scale behavior in Functions 5 min 🔒

Module 4: Security, monitoring, and troubleshooting 25% of exam

Securing and observing what you have built. The security half works through the managed identity model and the control-plane against data-plane distinction, the DefaultAzureCredential chain and its failure modes, sharing a user-assigned managed identity across services, Key Vault under Azure RBAC rather than vault access policies, soft delete and purge protection, SecretNearExpiry rotation through Event Grid, and App Configuration with Key Vault references and feature flags.The monitoring half covers Application Insights through the Azure Monitor OpenTelemetry Distro, adaptive sampling and telemetry cost control, Live Metrics streams and Standard availability tests, KQL for error triage, and a systematic method for diagnosing failures that cross services.

4.1 Managed identity and the Azure identity model for AI applications 5 min 🔒
4.2 DefaultAzureCredential and the credential chain 5 min 🔒
4.3 Sharing a user-assigned managed identity across services 5 min 🔒
4.4 Key Vault: Azure RBAC against vault access policies 5 min 🔒
4.5 Soft delete, purge protection, and vault recovery 5 min 🔒
4.6 Secret rotation and SecretNearExpiry through Event Grid 5 min 🔒
4.7 App Configuration, Key Vault references, and feature flags 5 min 🔒
4.8 Application Insights and the Azure Monitor OpenTelemetry Distro 5 min 🔒
4.9 Adaptive sampling and controlling telemetry volume 6 min 🔒
4.10 Live Metrics streams and availability tests 5 min 🔒
4.11 KQL for error triage in Log Analytics 5 min 🔒
4.12 Troubleshooting distributed AI workloads end to end 5 min 🔒

Frequently asked questions about the AI-200 training course

Is the AI-200 training course free?

Yes. The course costs nothing to read and opens once you are signed in to a free Certifym account, no payment and no card.

How is the course structured?

One module per official AI-200 exam domain, in Microsoft’s published order, with each module weighted to the domain’s published percentage. Within each module the material is broken into short lessons, followed by key terms and further reading. Because AI data management services carries 28%, it is the longest module.

Does this replace Microsoft’s official training?

No. Microsoft Learn is the authoritative source for the AI-200 study guide and its learning paths. This course is an independent study companion, written to be read quickly and to slot alongside practice questions, not a substitute for the official material.

Do I need development experience before starting?

Yes, more than most associate courses assume. The AI-200 blueprint takes for granted that you already deploy containers, wire up messaging and serverless services, and instrument applications, rather than learning those things here. If you are newer to the platform, the AZ-900 and AI-900 courses are the sensible way in.

What should I do after finishing the course?

Move to the AI-200 practice exam and work until module 2 and module 4 are as solid as module 1, since those are where the exam does most of its separating. Practice the case-study format specifically, reading a scenario once and answering several questions from it is a different skill from answering standalone items.

Is the course current with the latest AI-200 outline?

The course is built against the four-domain blueprint the certification guide tracks, version v1-2026-05, with the weightings Microsoft publishes for it. AI-200 is a new exam and Microsoft can revise a new blueprint faster than a settled one; check the current study guide on Microsoft Learn before you book.

Trademark notice & independence. Certifym.net is operated by Certifym Exam Services, LLC and is not affiliated with, endorsed by, or sponsored by Microsoft Corporation. Microsoft®, Azure®, and all related marks are trademarks or registered trademarks of Microsoft Corporation, used here only for factual reference to the exam. The AI-200 exam objectives and skill outline are the property of Microsoft Corporation; candidates should review the official, current study guide directly on Microsoft Learn.

All course content, questions, answers, and explanations on Certifym are original content created for study purposes. They are not actual Microsoft training materials or examination questions and are not represented as such. Studying with these materials does not guarantee a passing result on any live certification exam. Exam requirements, format, domain weights, and pricing are set by Microsoft and may change; always verify current details on Microsoft Learn before scheduling your exam.