Azure AI architecture
Four short sessions. One idea per card.
Orchestration calls these layers and receives results ↓
History · retrieved context · generated output · business actions
Original teaching diagram based on Microsoft’s pattern. Logical roles may share a service; this is not a required Azure resource list.
Follow one example request
- An employee asks whether a taxi is reimbursable.
- The API authenticates the request and checks access.
- Orchestration loads relevant history and permitted policy passages.
- The model drafts an answer using that context.
- If a claim must be submitted, a separate tool checks permissions and any required approval.
- The application validates and returns the result, recording appropriate evidence.
This is our illustrative walkthrough, not a claim that every request needs every layer.
TWO KINDS OF AI WORKLOAD
Does a RAG assistant need to train its own model?
Explain it in your own words, then compare.