Stand up the Foundry environment and get a working chat model before building anything on top of it. You'll create the Foundry resource and project, deploy the contoso-chat model, shape its behavior with help desk instructions in the chat playground, and put guardrails and a custom blocklist in place to keep it on task.
Turn a general-purpose chat model into a Contoso help desk agent that answers from real policy documents. You'll upload IT policy docs to a vector index, save the grounded model as a reusable agent, and connect a Foundry IQ knowledge base backed by Azure AI Search.
Give the agent reach beyond Contoso's own documents, then measure how well it performs. You'll add web search with citations for questions the knowledge base can't answer, then run an evaluation against a test dataset using contoso-chat as the judge.
Get the agent ready for real users. You'll assign Foundry roles to Entra ID security groups and attach the guardrail, wire up Application Insights tracing and a monthly budget, then publish the agent to Microsoft Teams and Microsoft 365 Copilot.
