Introducing Azure AI Foundry - Everything you need for AI development

Azure AI Foundry: A Comprehensive Overview

Introduction to Azure AI Foundry

• Azure AI Foundry is a unified platform designed for efficient AI development, providing essential building blocks for creating agentic solutions.

• The platform supports the entire AI development lifecycle, from concept and experimentation to production management.

Key Features of Azure AI Foundry

• Users can access a model catalog with thousands of models, including premium large language models from various providers like OpenAI and Meta.

• Models are categorized by specialization, supporting different languages and industries, hosted on Microsoft's supercomputer infrastructure for optimized performance.

Model Deployment and Customization

• Users can deploy models on managed hardware or bring their own models to run on Azure infrastructure. Experimentation is facilitated through an interactive playground.

• Knowledge integration options include uploading files, using search indexes, or adding web knowledge via Microsoft Bing and Microsoft 365 data sources. Actions can be defined for agents to perform tasks like API calls or running Python code.

Enhancing Agent Capabilities

• The new Azure AI Agents service allows users to orchestrate agents without managing underlying resources while integrating seamlessly with coding workspaces such as GitHub and Visual Studio.

• A single API enables easy connection to various models through the Azure AI model inference endpoint, allowing comparison without altering codebase. Continuous assessment tools help improve user experience based on feedback metrics.

Monitoring and Safety Features

• Centralized observability features include application tracing for debugging and automated evaluations based on key metrics like relevance and fluency of outputs. Reporting tools facilitate stakeholder communication through dashboards and alerts.

• Built-in safety controls automatically detect unwanted content across text, image, and multimodal inputs while offering advanced techniques like model fine-tuning for improved accuracy in real-world applications. Integrations with Semantic Kernel enhance multi-agent process orchestration as applications transition into production environments.

Creating Multi-Agent Applications with Azure AI Foundry

Overview of Microsoft's Security and Governance Stack

• Microsoft’s security and governance stack allows enforcement of organizational standards through Azure policy, identity management via Microsoft Entra, data security with Microsoft Purview, and threat detection for AI applications using Microsoft Defender.

Introduction to Multi-Agent Application Development

• The session introduces the creation of a multi-agent application utilizing Azure AI Foundry and Semantic Kernel for orchestration.

• A four-agent solution is described: a researcher agent gathers information, a writer agent composes the report, an editor agent reviews it, and a sender agent emails the final output.

Agent Functionality Breakdown

• Each agent operates in a modular fashion akin to microservices; this structure enhances process efficiency by breaking down monolithic tasks into manageable components.

Building Agents in Azure AI Foundry

• The speaker begins building agents within Azure AI Foundry, starting with the researcher agent while noting that other agents will be created subsequently.

• Configuration options are provided for each agent; for instance, the researcher agent is set up to use Bing search as its knowledge source without attempting to write reports.

Testing the Researcher Agent

• The research agent is instructed to gather concise information from Bing. It generates summarized results based on queries like "What is dot net," ensuring outputs are dense with relevant knowledge.

Creating the Sender Agent

• Transitioning to VS Code, the speaker demonstrates how to create an email sender agent using Python SDK. This includes defining its name and tools (Outlook).

Wiring Up Agents Using Semantic Kernel

• After creating all four agents, they are connected through Semantic Kernel. Each configuration file links back to their respective IDs in Azure AI Foundry.

Demonstrating Agent Interaction

• The program runs an example where users can request reports. The interaction between agents is observed live as they collaborate on generating content based on user prompts.

Example Scenario: Report Generation Process

• In a demonstration scenario about Python snakes (the animal), each agent's role is highlighted: gathering information by the researcher followed by writing and editing processes until final approval triggers sending via email.

Conclusion: Benefits of Using Azure AI Foundry

• The session concludes by emphasizing how Azure AI Foundry streamlines creating powerful agents efficiently across various stages of AI development. Viewers are encouraged to explore further at ai.azure.com.