1.2 Application Scenario¶
In this tutorial, we build a retail chat AI (copilot) that uses Retrieval Augmented Generation (RAG) to ground the chat responses in the retailer's own data. Let's briefly review what this means.
1. RAG Chat App (tutorial)¶
This RAG Chat tutorial provides a quickstart for building and evaluating a basic RAG-based copilot using the Azure AI Foundry portal and SDK. The tutorial is grounded in the Contoso Outdoors retailer data and combines both low-code (Portal) and code-first (SDK) steps to teach the latest Azure AI Foundry tools and features. Think of this as a sandbox for open exploration
The figure explains the RAG pattern visually:
- The user question (prompt) is received at the copilot hosted endpoint
- The question is used to retrieve related knowledge from relevant sources.
- The prompt is then augmented with knowledge as context, and sent to the model.
- The model now generates a response that is "grounded" in this knowledge context.
2. Contoso Outdoor (chat UI)¶
Contoso Outdoor is a fictitious enterprise retailer specializing in hiking and camping gear for outdoor enthusiasts. Their website (chat UI) provides customers with a catalog of their products, with product pages offering detailed information for user review. We'll look at the retailer data sources in the next section.
3. Contoso Chat (chat AI)¶
The chat UI shown is not used in THIS workshop - but code is open-source if useful.
Contoso Chat is the open-source reference implementation of a custom RAG-based retail copilot based on the Contoso Outdoor retail scenario. It is implemented as an AI App Template that can be provisioned and deployed to Azure Container Apps to provide a hosted API endpoint. Customer requests on the chat UI (website) are now directed to this chat AI (endpoint) for processing, allowing for the user experience shown below.
