Blog Post

Azure Architecture Blog
5 MIN READ

Azure OpenAI Landing Zone reference architecture

FreddyAyala's avatar
FreddyAyala
Icon for Microsoft rankMicrosoft
Jul 24, 2023
Azure Landing Zones provide a solid foundation for your cloud environment. When deploying complex AI services such as Azure OpenAI, using a Landing Zone approach helps you manage your resources in a ...
Updated Aug 04, 2023
Version 3.0
artificial intelligence
infrastructure
well architected
FreddyAyala's avatar
FreddyAyala
Icon for Microsoft rankMicrosoft
Sep 19, 2023

Hello wmeints

I understand your concerns and feedback regarding the complexity and organization of the proposed architecture. It's essential to simplify and provide clear guidance for implementing Azure OpenAI effectively. Here's a response to each of your points:

1. Complexity and Multiple Subscriptions:

  • The complexity in the reference architecture, such as splitting components across multiple subscriptions, can indeed increase management overhead. The design can be simplified by consolidating resources into a single subscription, especially when a single team is responsible for the entire ChatGPT-like solution.
  • Multiple subscriptions might be recommended in scenarios where different teams or stakeholders have ownership of specific resources, but it's not necessary for all use cases.

2. Inclusion of Many Azure Services:

  • You are correct that not all Azure services are required for every solution. Simplicity and cost-effectiveness are key considerations. If public networking meets the security and performance needs of your solution, it can be a valid choice.
  • Using Azure Web App behind Azure Front Door integrated with a Virtual Network and a private endpoint for OpenAI can indeed provide a secure and cost-efficient solution.

3. Request for Diagrams and Explanations:

  • Believe it or not it's not my main job to do these architectures 🙂 this is just out of passion, I'm highly encourage by my awesome manager, but doing all of what you request requires some time so you just need to be patient. In the mean time I've upgraded my landing zone reference architecture repository with a simple code example of how to deploy a Private OpenAI Web Application that explains in a more simple a basic scenario, you can check it out: https://github.com/FreddyAyala/AzureAIServicesLandingZone/blob/main/README.md

Your feedback is harsh but valuable and highlights the need for clear, simplified, and practical guidance for implementing Azure OpenAI solutions and that will come very soon. However I suggest you to also get familiar with Cloud Adoption Framework, Landing Zones and also Well Architected Framework so you can grasp more of what has been exposed here.