Chapter Eight

AI & Machine Learning

Six services spanning the AI surface of Azure — a platform for building models, the OpenAI models behind generative apps, and pre-trained APIs for vision, language, speech, and documents.

6 services

Core Terminology

Azure's AI services range from raw model training to ready-made APIs. These terms recur across the chapter.

Azure AI Foundry
The umbrella studio for building generative-AI applications — model catalog, deployments, evaluation, and prompt management.
Pre-trained API
A managed model you call over HTTP without training anything — vision, language, speech, document — billed per transaction.
Deployment
A provisioned instance of a model (such as GPT-4o) in Azure OpenAI with its own throughput quota, measured in tokens per minute.
Token
The unit generative models read and write text in, and the unit Azure OpenAI bills. Prompt plus completion tokens determine cost and latency.
Endpoint
The deployed, network-addressable target that serves a model for real-time or batch inference in Azure Machine Learning.
Grounding
Supplying a model with retrieved, authoritative context (often via Azure AI Search) so its answers reflect your data, not just its training.

Services in This Chapter