From Test Kitchen to Table: A Demo-driven Tour of Foundry Portal for AI Developers | OD836

Nitya Narasimhan and Carlotta Castelluccio walk through an end-to-end, demo-driven workflow in the Foundry portal: starting with rapid experimentation in the playground and ending with production-ready agent deployment. The session maps portal features to familiar developer tasks (prototype, iterate, evaluate, deploy, monitor) and highlights observability and evaluation capabilities that complement a code-first workflow.

Overview

What the session covers

• A live journey from experimentation in the Foundry portal playground to deploying an agent.
• How the portal UI supports:
  • Workflow design
  • Model management
  • Monitoring/observability
• How portal capabilities map to common developer tasks, while still supporting a code-first mindset.

Scenario used in the demo

• Building “Cora”, a retail AI assistant (for “Zava”).
• Explaining the AI app structure and how it transitions toward agent-based microservices.

Key portal capabilities highlighted

• Creating a project and building an initial retail agent prototype.
• Debugging and performance measurement using:
  • Traces
  • Evaluation tools
• Iteration features:
  • Customizing instructions
  • Versioning
  • Web-based agent preview
• Model selection and optimization:
  • Using the Foundry model catalog
  • Comparing models via a leaderboard
• Improving grounding and safety:
  • Knowledge grounding improvements
  • Evaluations
  • Red teaming for safety testing

Resources

• https://aka.ms/build26/OD836
• https://aka.ms/build/foundrydiscord

Session context

• Microsoft Build 2026 on-demand session (OD836)
• Track/topic label: “Agents & apps”

Chapter timestamps (from the video description)

• 0:00 - Introduction and session overview – ‘From Test Kitchen to Table’
• 00:00:34 - AI development scenario: building Cora, a retail AI assistant for Zava
• 00:01:15 - Explaining AI app structure and transition to agent-based microservices
• 00:03:39 - Introducing Microsoft Foundry Portal as the ‘AI Test Kitchen’
• 00:10:00 - Hands-on demo: creating first project and retail agent prototype
• 00:17:00 - Using traces and evaluation tools for debugging and performance measurement
• 00:26:00 - Customizing instructions, versioning, and web-based agent preview
• 00:31:00 - Model comparison and optimization using Foundry model catalog and leaderboard
• 00:50:00 - Improving knowledge grounding, evaluations, and red teaming for safety testing
• 01:15:00 - Final summary: end-to-end agent development, code-first transition, and community resources