Build Secure, Observable, Production-ready Agents with a Control Plane [APAC] скачать в хорошем качестве

Build Secure, Observable, Production-ready Agents with a Control Plane [APAC]

Brought to you by Microsoft and NVIDIA Orchestrate, secure and observe agents enabling advanced AI workflows. Reliable agent orchestration requires clear separation between control and execution. In this joint session, Microsoft and NVIDIA present a multi-agent architecture where Foundry Agent Service coordinates GPU-backed agents running on Azure Container Apps. Using a document processing scenario with multiple open-source models including NVIDIA Nemotron models, we show how agents route requests, apply security controls, and monitor execution across GPU workloads, producing consistent and explainable results suitable for production environments. 📌 This episode is a part of a series. Learn more: https://aka.ms/AIAgentsApps/y-MSFT Chapters: 0:06 Welcome & Session Introduction 0:23 Housekeeping & Code of Conduct 0:56 Speaker Introductions 1:57 Session Agenda & Overview 4:03 NVIDIA + Azure Infrastructure Overview 7:17 GPUs, Networking & Confidential Computing 8:01 Microsoft + NVIDIA AI Integrations 9:38 Introduction to NVIDIA Nematron 11:21 Nematron Models for Agentic AI 12:29 Nematron Parse & Model Families 14:42 Better Together: Microsoft + NVIDIA Stack 17:52 Transition to Live Demo 18:19 What Is Agent Orchestration? 20:09 Live Demo: Invoice Processing Agent 22:45 Anomaly Detection Results Explained 25:44 Azure Container Apps & GPU Scaling 27:05 Foundry Agent Orchestrator Architecture 28:36 Microsoft Agent Framework vs Semantic Kernel 30:01 Model Selection in Azure AI Foundry 31:24 Nematron Parse Model Deep Dive 33:59 Why GPT‑4 for Orchestration 36:33 Using NVIDIA Build (No Deployment Required) 39:57 Control Plane, Security & Scalability 42:48 Observability, Tracing & Monitoring 47:33 Policy Enforcement & Auditability 51:49 Design Principles & Best Practices 55:13 Choosing the Right Model for the Job 58:23 Orchestrating Multi‑Model AI Systems 1:01:39 Resources, Code & Next Steps #microsoftreactor #learnconnectbuild [eventID:26683]