Collaboratively build AI apps and share resources with hubs and projects

[Guest deeikele]

At Microsoft Build '24 last week, we announced the general availability of Azure AI Studio. With this release, we are also introducing hubs, which enable developers to self-service create project workspaces and access shared company resources without needing an IT administrator's repeated help.

 

 

 

Hubs provide a central way for a team to govern security, connectivity, and computing resources across playgrounds and projects. Project workspaces that are created using a hub inherit the same security settings and shared resource access. Teams can create as many project workspaces as needed to organize their work, isolate data, and/or restrict access.

 

 

 

By balancing easy, distributed project creation and resource management with centralized controls for security, compute governance, and compliance, Azure AI empowers developer agility and enterprise IT governance at scale.

 

 

 

[HEADING=1]Rapid AI use case exploration without IT bottlenecks[/HEADING]

 

Successful AI applications and models typically start as prototypes, with developers testing the feasibility of an idea or assessing the quality of data or a model for a particular task. This is a steppingstone towards project funding or a full-scale implementation.

 

 

 

The transition from proving the feasibility of an idea to a funded project is where many organizations encounter a bottleneck in productivity, because a single platform team is responsible for the setup of cloud resources. Such a team may be the only one authorized to configure security, connectivity or other resources that may incur costs. This can cause a huge backlog, resulting in development teams getting blocked on innovating with a new idea.

 

 

 

Azure AI Studio hubs help mitigate this bottleneck. IT can set up a pre-configured, reusable environment, or hub, for a team one time, and a team can use that hub to create their own project workspaces for prototyping, building, and operating AI applications.

 

 

 

[HEADING=1]Set up and secure a hub for your team[/HEADING]

 

Get started by creating a hub in Azure AI Studio, or use Azure Portal for advanced configuration options. You can customize networking, identity, encryption, monitoring or tags, to meet compliance with your organization’s requirements. Step by step guidance on how to create a hub with customized security settings can be found in our documentation.

 

 

 

Often, projects in a business domain require access to the same company resources such as vector indices, model endpoints or repos. As a team lead, you can pre-configure connectivity with these resources within a hub, so developers can access them from any new project workspace without delay on IT.

 

 

 

Connections let you access objects in AI Studio that are managed outside of your hub. For example, uploaded data on an Azure storage account, or model deployments on an existing Azure OpenAI resource. A connection can be shared with every project or made accessible to one specific project, with the option to configure key-based access or EntraID-passthrough to authorize access to users on the connected resource. Plus, as an administrator, you can easily track, audit, and manage connections across the organization from a single view in AI Studio.

 

 

 

 

[HEADING=1] [/HEADING]

[HEADING=1]Prototype on your idea in the playground[/HEADING]

 

Once your hub is set up, you can immediately discover and deploy models from the Azure AI model catalog, to start experimenting in the playground without the need for a defined project. The playground in AI Studio helps developers understand how a model behaves in a controlled environment. For example, you can experiment with different system message prompts to assess the feasibility of a new idea.

 

 

 

Once you’re ready to integrate a model into code, you can transition to working in a project. You can get the API key from the deployment page, and select AI services, such as Azure AI Speech and Azure AI Language, come with additional REST APIs and SDKs to ease integration.

 

 

[HEADING=1] [/HEADING]

[HEADING=1]Build customized AI applications using a project[/HEADING]

 

Projects are containers to help organize work and collaborate on a single use case. They help organize components, let you upload or connect with data in isolation, and restrict access to you and other project members. Projects inherit the hub’s security settings and shared resources. This includes virtual network, encryption settings, computing and storage.

 

 

 

In AI Studio, a project grants access to customization tools including prompt flow, content safety, fine-tuning, and assistants. With these tools, you can connect or upload your company data, add content filters for problematic inputs or outputs, evaluate your entire application for quality and safety metrics, then deploy as a webapp and monitor it in production.

 

 

 

 

If you are working in Azure Machine Learning, you can now create a project workspace from within the studio experience, too. This feature is now available in preview. By selecting a hub, you get access to shared company resources made available by your administrator including compute, connections and network connectivity.

 

 

The same project workspace can be accessed from both AI Studio and Azure Machine Learning. For teams with a mix of developers and data scientists, this means developers can use AI Studio to build and operate GenAI applications responsibly, and data scientists can user Azure Machine Learning to build and operate custom machine learning models, all while sharing components, including datasets and flows.

 

[HEADING=1] [/HEADING]

[HEADING=1]Organize for your team’s development needs[/HEADING]

 

The number of hubs and projects you need depends on your way of working. Typically, customers aim to create a hub for a large team with similar data access needs to maximize cost efficiency, resource sharing, and minimize setup overhead. For example, you might create a hub for all projects related to customer support.

 

 

 

If your organization requires segregation between dev, test and production environments as part your LLMOps or MLOps strategy, consider creating a hub for each environment. Dependent on the readiness of your solution for production, you may decide to replicate your project workspace in every environment or keep it just in one.

 

 

 

Use Azure Role-Based Access Control (RBAC) to configure with granularity who can create hubs, projects and connections in your organization. The built-in Azure AI developer role grants permissions for common development tasks in AI Studio, so is ideal for a contributing user on a project workspace. Any user with an AI developer role assigned on a hub workspace can also create new project workspaces.

 

[HEADING=1] [/HEADING]

[HEADING=1]Get started in Azure[/HEADING]

 

To sum up, hubs provide a shared Azure environment for a team with shared connectivity, compute and governance, yielding enhanced self-serve capabilities and business agility for development teams.

 

 

 

The following links provide further information on concepts and how to get started in AI Studio or Azure Machine Learning, including infrastructure template examples.

 

• Develop and deploy generative AI apps responsibly with Azure AI Studio
  
• Azure AI hub resource concepts - Azure AI Studio
  
• How to create and manage an Azure AI hub resource - Azure AI Studio
  
• Create an AI hub using a Bicep template - Azure AI Studio
  
• What are hub workspaces? - Azure Machine Learning
  
• Manage workspaces in portal or Python SDK (v2) - Azure Machine Learning
  

 

Continue reading...