Getting started with Azure AI Studio

During the opening keynote of the Microsoft Ignite event, on November 15th, Artificial Intelligence was the undiscussed protagonist. Among the many AI-related announcements, a common flavor has emerged: the idea of making AI and, above all, GenAI more and more consumable and customizable by individuals and organizations.

In this article, I'm sharing some first glimpses of the newly announced Azure AI Studio, now in public preview.

Azure AI Studio can be seen as your enterprise AI hub, an open ecosystem that brings together AI models (including powerful LLMs like the GPT-4 as well as models in other domains such as vision and speech), AI-powered application components (such as prompts, vector indexes, data...), AI evaluation metrics (both built-in and custom) and responsible AI components (content filtering). The idea is that of allowing AI-as-a-service, so that everyone can leverage the best of breed of AI offering in the market within a powerful platform that makes it easier to build AI applications.

Exploring Azure AI Studio

To get started with Azure AI Studio, you can start exploring its model catalog to pick the one you need for your application. You can also produce a quick benchmarking dashboard between two or more models, to better understand their performance in given domains.

You can also explore the different domains of AI (not only generative) where Microsoft developed specialized solutions, such as Speech, Vision, and Language.

You can then leverage the built-in content safety filter and adapt it to your scenario. For example, you can decide to make it more strict toward specific categories of harmful content as I did in the following demo:

As you can see, content that was initially allowed is now filtered out due to my enforcement of stricter triggers.

Finally, you can also navigate a library of pre-built metapromts you can leverage to tailor your LLMs for vertical scenarios. You can also enrich this library so that your organization can benefit from it.

As mentioned above, Azure AI Studio is meant to be your AI Hub and, as such, it needs to be easily integrated with external components and services such as Azure OpenAI Instances, Azure AI Search, and similar. To do so, you can create connections and manage them in the Manage tab, which also allows you to have governance over your AI components (compute instances, permissions, billing etc.).

Now that we have a first overview of the Studio, let's see it in action.

Getting started with a simple application

Let's now see how to build a simple RAG application leveraging Azure AI Studio. I'll start by adding my components to the Studio. I will be using the following components:

• A PDF I want to chat with (I will use the paper Attention is All you Need)

• Connections towards my Azure OpenAI and Azure AI Search Instances

• An Index created over the PDF

Now that we have all our ingredients, let's start experimenting with the well-known playground. Since I've already created my Index, I can connect my GPT-4 model directly to it from the GUI widget:

Below the result:

I can also add multimodality to my chatbot, by enabling the speech2text and text2speech capability, backed by Azure AI speech models:

Now, since Azure AI Studio brings together also the AML capabilities, we can open this first prototype into Prompt Flow seamlessly and make any change we want in each of its components (to do so, you will need a compute instance as runtime):

Last but not least, you can also evaluate your flow with built-in metrics or custom metrics, also providing a test dataset of Q&A pairings:

Once you are happy with your result, the final flow can be deployed to an online endpoint for real-time inference.

This was an easy example of how you can leverage the power of a unified AI ecosystem such as Azure AI Studio, and we've just scraped the surface of its true potential. There are so many components - Responsible AI, metaprompting templates, documents' embedding - that can be made easier and with a greater time to market thanks to this AI hub.

Looking forward to further experiments on Azure AI Studio and discover its capabilities!

References

• Azure AI Studio - Generative AI Development Hub | Microsoft Azure
• Full Keynote: Satya Nadella at Microsoft Ignite 2023 - YouTube
• [1706.03762] Attention Is All You Need (arxiv.org)