Six jobs. Six specialized models.

Frequently asked questions

Why does semantic search find a policy that shares no words with the question?

Embedding models like Ada and Cohere convert both the question and the documents into vectors that represent meaning, not exact text. "Can I return shoes that got wet in the rain?" and "items damaged by weather are refunded in full" share zero keywords but sit near each other in vector space, so the match still surfaces — a keyword search would return nothing.

Can one Azure AI Foundry project really cover search, images, video, claims, and support calls?

Yes. Embeddings, GPT-image-1, Sora 2, GPT-4.1, and the GPT-4o transcribe/tts pair are all separate task-built models sitting in the same catalog under the same project, so one team's Foundry setup can back retail search, marketing creative, claims triage, and voice support without standing up separate infrastructure for each.

Read deeper on Azure

Twenty seconds gets the shape of an idea across. These go into how it behaves in production.

• Azure AI Foundry Model Catalog: How to Choose the Right Model

  A decision framework for the Azure AI Foundry model catalog: LLMs vs SLMs, chat vs reasoning models, and specialized models, with real business use cases.

• How Azure Protects a Mobile App: The Full Request Flow, Layer by Layer

  Front Door, your API, Microsoft Entra ID, authorization and a private database — the five layers that stand between a mobile app and its data on Azure.

• Your Mobile App Is Leaking Its API Key — and Obfuscation Will Not Save It

  An APK or IPA is a zip file. Anything hardcoded inside it, including your API key, is already public — and here is how to get it off the phone for good.