AI-200 study plan

Provision an Azure AI Foundry project and deploy both a chat-completion model and an embedding model using the Python SDK

Build a retrieval-augmented generation pipeline that grounds answers in a document set stored in Azure AI Search and returns citations

Compare three prompt variants (zero-shot, few-shot with grounding, and chain-of-thought) and record accuracy on a 20-question evaluation set

Configure and document the effect of temperature, top_p, max_tokens, and content-filter severity levels on model output

Implement a multimodal request that sends an image to a vision-enabled model and forces a structured JSON response validated against a schema

Secure an AI solution using managed identity, Key Vault-backed secrets, and role-based access control with no keys in code or config

Enable diagnostic logging, metrics, and an alert rule for an AI resource and verify telemetry arrives

Estimate monthly cost for two deployment options and justify the selection in writing

Match six scenario descriptions to the correct Azure AI service and deployment option, stating the deciding constraint for each

Author infrastructure-as-code that provisions the AI resources and model deployments reproducibly across two environments

Build a CI/CD pipeline that deploys the application and model configuration together and rolls back on failure

Configure quota, scale limits, and a failover or retry strategy for a model deployment under load

Create an agent with the Azure AI Agent Service and give it persistent instructions, a defined tool set, and file-based knowledge

Wire at least three tools to the agent, including function calling to a custom API, code interpretation, and file search

Implement a connected or multi-agent flow where a routing agent delegates to at least two specialist agents

Add guardrails and tracing, then diagnose one failing run from the trace and correct the agent configuration

Extract printed and handwritten text from images and scanned pages and return the results with bounding-box coordinates

Classify and detect objects by training a custom image model on a labeled dataset and report precision and recall

Generate images from text prompts and apply responsible-AI settings that block disallowed content

Analyze a video or image stream for spatial and temporal insights and summarize the returned metadata

Run sentiment, key-phrase, entity, and PII detection over a text corpus and redact sensitive values in the output

Summarize long documents with both extractive and abstractive approaches and compare fidelity

Transcribe audio to text and synthesize speech with configurable voice and format settings

Build a conversational language understanding model with intents and entities, train it, and evaluate it against test utterances

Extract fields, tables, and key-value pairs from forms with a prebuilt document model and compare the output to a custom-trained model

Assemble an enrichment pipeline that runs OCR, key-phrase extraction, and entity recognition over blob storage and writes results into a search index

Design an index schema, indexer schedule, and query strategy that support filtered and semantic search over enriched content

Expose the enriched index to a generative app and verify grounded answers cite the correct source documents

Complete a full-length timed practice assessment and log every missed question by domain and concept

Re-study the two lowest-scoring domains and re-answer the missed items without notes

Integrate a generative model, an agent, and a vision or extraction service into one capstone solution that satisfies a stated business scenario

Explain aloud, in under two minutes each, the design tradeoffs for six common architecture decisions across the domains