Agentic AI - From Theory to Practice: Developing Multi-Agent AI Systems on Azure

"I have a better feeling about that one." If this is your AI evaluation strategy, you need a better way. Learn to measure agent performance with concrete KPIs.

#1about 3 minutes

Understanding the core components of an AI agent

AI agents evolve beyond simple chatbots by using large language models, instructions, and tools to integrate directly into business processes.

#2about 2 minutes

Solving developer challenges with Azure AI Foundry

Azure AI Foundry provides a comprehensive platform to address common developer challenges like model selection, security, and observability when building AI agents.

#3about 1 minute

Exploring the Azure AI Foundry Agent Service

The Agent Service offers enterprise-grade features including orchestration, SDK integrations, knowledge tools, and built-in content safety for robust agent development.

#4about 2 minutes

The development lifecycle from ideation to production

The AI application lifecycle consists of ideation, implementation, and operations, starting with low-cost experimentation in GitHub Models before moving to Azure Foundry.

#5about 4 minutes

Experimenting with prompts and models in GitHub

Use GitHub Models to rapidly prototype by comparing different prompts and models against test datasets and using evaluators to generate performance KPIs.

#6about 3 minutes

Integrating evaluation and monitoring into your workflow

Implement a robust evaluation strategy by incorporating KPI checks into CI/CD pipelines and using continuous monitoring with end-to-end tracing in production.

#7about 7 minutes

Building a multi-agent contract analysis application

A practical example demonstrates a multi-agent system that analyzes contracts, checks compliance, and uses automated evaluations for continuous quality assurance.

#8about 2 minutes

Choosing the right multi-agent interaction pattern

Design effective multi-agent systems by selecting the appropriate interaction pattern, such as sequential, concurrent, or group chat, based on your process and outcome goals.

#9about 2 minutes