Building a Grounded AI Concierge with GitHub Copilot CLI and Microsoft Foundry

The video demonstrates building a conversational AI agent for Tailspin Toys, a crowdfunding site for developer-themed board games, that answers questions grounded in a catalog of 21 games. The agent uses GitHub Copilot CLI, Microsoft Foundry, and Azure skills to plan, build, test, and deploy, then connects to a static website via a secure proxy. The agent only answers from the catalog and never fabricates missing information.

■ Planning and Setup
- Using GitHub Copilot CLI in plan mode with Microsoft Foundry skill
- Creating Azure resource group and Foundry project
- Selecting an Azure region and handling resource provisioning issues

■ Model Selection and Testing
- Recommending and deploying a suitable model (GPT-5.4 mini) based on criteria
- Testing the model for grounding and token usage

■ Agent Development and Deployment
- Scaffolding the agent code and local testing with an evaluation set
- Improving the agent prompt and rate limit, then deploying to Microsoft Foundry

■ Integration with Website
- Building an Azure Functions proxy for secure client-server communication
- Adding a check widget to the website for user queries

This is for developers interested in building grounded AI agents using Azure and GitHub Copilot. Viewers will learn a practical workflow from planning to deployment, and how to ensure answers are based solely on provided data. The full workshop is available for hands-on practice.