The Future of AI: Reduce AI Provisioning Effort - Jumpstart your solutions with AI App Templates

The Future of AI blog series is an evolving collection of posts from the AI Futures team in collaboration with subject matter experts across Microsoft. In this series, we explore tools and technologies that will drive the next generation of AI. Explore more at: https://aka.ms/the-future-of-ai 

In the previous post, we introduced Contoso Chat – an open-source RAG-based retail chat sample for Azure AI Foundry, that serves as both an AI app template (for builders) and the basis for a hands-on workshop (for learners). And we briefly talked about five stages in the developer workflow (provision, setup, ideate, evaluate, deploy) that take them from the initial prompt to a deployed product. But how can that sample help you build your app? 

The answer lies in developer tools and AI App templates that jumpstart productivity by giving you a fast start and a solid foundation to build on.

Imagine this familiar scenario.  

You are a traditional application developer in an enterprise and have been asked to build an AI-powered chat application that answers questions about your products. Where do you even start?  

1. Do you know what architecture to use? (Think Retrieval Augmented Generation) 
2. Do you know what “models” to use? (Chat model, Embeddings model) 
3. Do you know what “services” you might need? (Safety, Search, Model Hosting) 
4. Do you know how to “build” the application around them? (ideate-evaluate-deploy) 
5. Can you make the development workflow repeatable across teams? (collaborative) 

If this is not complex enough, consider the fast-growing ecosystem of models, frameworks and tools that are coming up around AI. How can you flatten your learning curve? 

Azure AI App Templates can help in three ways: 

1. They implement infrastructure as code – with template files that can be version controlled and activated consistently across teams, with the Azure Developer CLI. 
2. They use configuration as code – with dev container files for a Docker container with all dependencies pre-installed, that can be activated consistently across teams, in the cloud (with GitHub Codespaces) or locally (with Docker Desktop). 
3. They provide a working app foundation with a defined application architecture. Now, instead of having to figure out your design from scratch, you can start with a template that has the key requirements for your scenario – and customize it for your needs (with updated models, data, app source and evaluation metrics). 

Let’s revisit that scenario now. Want to build a custom retail chatbot grounded in your own data? Here’s how you can make that happen with Contoso Chat. 

Discover it - with AI App Template Gallery 

Let’s start with the discovery process. How would you have found the right template for your needs if I hadn’t told you about it? You’d start with the AI app templates gallery as shown below.

Simply use the filters in the gallery to find the AI template that supports your use case. Let’s say you came in with the following criteria: 

• You have your customer data in an Azure CosmosDB 
• You have product indexes built with Azure AI Search 
• You want to use Azure OpenAI Service models for chat and embeddings 
• You want to build your application using the Azure AI Foundry SDK in Python 

Fill those requirements in – and you will see the recommended template is for the Contoso Chat sample. Click on the tile to get more details like the resources used, as shown below. 

Develop with GitHub Codespaces 

You have a template – what do you do now? The first thing you want to do is to take the template for a spin and see if the features and experience match your needs.  

Activating that template requires you to use the Azure Developer CLI tool (more on that in a minute) – and install additional dependencies (for example: Azure AI Foundry SDK and individual Python SDKs for services used, and VS Code extensions to boost productivity). Built-in devcontainer support in template repos makes this a 1-click experience, as we’ll see in a minute. 

But you also have a choice – you can fork the existing sample to get a sandbox copy that you can periodically sync with the original for updates. Or you can use “azd init” to create an instance of that template (at the current time) and use that as the basis for a new repo. We recommend the first approach for learners, and the second for builders. The first approach allows you to track updates to the sample and learn about new features or tools.

Contoso Chat has a prebuild-ready branch used with this workshop, as shown in the figure on the right. Want to jumpstart your learning journey? Use this prebuild link to launch the wizard below - and setup your GitHub Codespaces environment in minutes, with 1 click. 

Provision with Azure Developer CLI 

Okay, so you found the right template for your needs. And you have your development environment running in GitHub Codespaces to start building. And all this took minutes. So, what do you need to do to provision, deploy, and explore, the sample app? You need just one tool (azd) – and it’s already pre-installed in your GitHub Codespaces by default! 

The comic below gives you a visual guide to the Azure Developer CLI documentation explaining what it does, how it works, why it matters, and how to use it with templates like Contoso Chat. 

Want to get a more structured understanding of the Azure Developer CLI workflow? Check out this free learning path that covers the same information with hands-on labs.  

For now, we just want to deploy the template and explore the application. To do that, launch the GitHub Codespaces session as explained earlier, then wait till you see the Visual Studio Code environment become active in that browser tab. Going from template to deployment is just two steps away: 

1. Authenticate with Azure (using “azd auth –use-device-code”) to connect the development environment with an active Azure subscription. 
2. Deploy the application with one command (“azd up”) – which provisions the required resources, populates required data, and deploys the application. 

You will now have a RAG-based retail chat AI deployed to an Azure Container Apps hosted endpoint that you can test using the built-in Swagger (“/docs”) endpoint – or integrate with your external applications or clients for driving a better user experience.

The deployment process will take a few minutes to complete with minimal involvement needed from you or your IT admins at this stage! You can now visit the following “portals” to explore the deployment in more detail: 

• Visit Azure Portal to understand the resource deployments associated with this architecture - specifically the Azure AI hub, project, and services resources that are typical for an Azure AI Foundry project. You can also explore the data samples used (product index in Azure AI Search, customer database in Azure CosmosDB) to get a sense for the schema and usage (e.g., vector search with semantic ranking). 

• Visit Azure AI Foundry portal to manage your generative AI application needs in one place – from discovering and deploying new models, to activating content filters for safety, to viewing application traces or evaluation results when enabled. 

The Azure AI Foundry portal helps you monitor your application with enterprise-grade management features.

Recap and Next Steps 

We started off this post by asking “how can an AI template help you build your app?” with specific focus on improving developer productivity for jumpstarting new projects. And we saw how AI app templates solved three challenges for us: 

• Reuse vs. Build from scratch – knowing the right AI architecture and components to use can be complicated. Start with a foundation template and customize instead. 

• Configuration as code – get a consistent, reproducible development environment with a prebuilt dev container that can be activated in the cloud, or on local device. 

• Infrastructure as code – use AI app templates with the Azure Developer CLI, to ensure a consistent and reproducible provisioning experience, with minimal developer effort. 

Now, you have a working app and development environment. Next, it’s time to customize it to your needs. And that means understanding how that application was designed and evolved from prompt to prototype. Join me next time to look at how we can ideate with Prompty! 

Are you ready to start developing? Here are some resources that can help! 

1. AI app templates gallery - Discover other AI solution templates to deconstruct.  
2. Contoso Chat repository - Browse the README for a self-guided quickstart. 
3. Azure AI Foundry - Discover AI models and services tailored to your use case. Explore the management center to manage resources, quotas and more throughout the dev lifecycle.