Putting Microsoft AI into production. Twenty courses covering architecture, Azure AI, Copilot Studio, security and compliance, integration, data, testing, monitoring and cost.
The Deploy pillar is the build half of the practice. It assumes the case is made and the pilot is over — what follows is architecture, security, integration, data, testing, monitoring and the cost of running it. Take them in order or jump to the one in front of you.
Why architecture comes first · The three core patterns · Azure AI Foundry hub and project model
From pilot to production · Reliability and resilience · Capacity and quota planning
The service in context · Model catalogue and deployments · Endpoints, tokens, and rate limits
The Azure AI service family · Language, vision, speech, documents · RAG with Azure AI Search
What production hardening means · Monitoring and evaluation · Resilience and safe deployment
What Copilot Studio is for · Topics, triggers, and entities · Knowledge sources and actions
Beyond the first agent · Generative answers and knowledge · Variables, conditions, and Power Fx
Why extend Microsoft 365 Copilot · Declarative vs custom agents · Plugins and Graph connectors
The AI threat surface · Prompt injection and jailbreaks · Exfiltration and oversharing
Compliance for AI at work · Data residency and the EU Data Boundary · Labels, retention, and DLP
The integration problem · Connectors and Graph connectors · APIs and custom connectors
The low-code AI stack · AI Builder models · Document processing patterns
Why pipelines decide AI quality · Ingestion and the medallion pattern · Transformation and quality gates
Fabric as the data foundation · OneLake, lakehouses, and Delta · Notebooks and real-time intelligence
Why AI testing is different · Evaluation sets and metrics · Red-teaming and quality gates
Why AI needs its own quality discipline · Standards: ISO/IEC 42001 and the RAI principles · Internal audit and evidence
What observability means for AI · The four signal layers · Drift and groundedness monitoring
Why AI incidents are different · Severity classification · Response playbooks
The pilot-to-production gap · What breaks at scale · Governance and ALM
Token economics · Prompt and context efficiency · Caching, batching, and model selection