Build Generative AI Applications End-to-End, code-first, on Azure AI Studio

Building Generative AI applications can seem daunting to traditional app developers. What is the end-to-end application development cycle? Which model should I use and where can I find it? What tools should you use to build, test, and deploy AI applications? This blog post gives you a sneak peek into a week-long series of posts that were just published, giving you a first-hand experience of the process. Let’s find out more!

This week we published a five-part blog about our Building Intelligent Apps initiative. #30DaysOfIA series. Our focus was application developers looking to build custom co-pilots. code first Azure AI gives you greater control over the decisions made in the end-to-end workflow for your generative AI applications. we did this By looking at two key samples: (contoso chat and Contoso Creative Writer) From prompt to production. Along the way, we shared insights into key tasks and developer tools to simplify them.

This blog post will briefly introduce the two applications, provide an overview of what is covered in the series, and provide links to each post to dive deeper. Ready? Let’s go!

Our first application is contoso chatA customer service chatbot that answers user questions about the retailer’s products. Create search augmentation Use patterns (RAGs) to elicit responses from both your product catalog and customer purchase history.

Our second application is Contoso Creative WriterContent Publishing Assistant using multi-agent conversation It is a pattern that autonomously coordinates and executes multiple tasks on behalf of the user.

The figure below shows the AI ​​application architecture. contoso chat Retail co-pilot. User requests are received through endpoints hosted in Azure Container Apps, and then RAG processes user requests using Azure AI Search (product index) and Azure Cosmos DB (customer database) along with Azure OpenAI Services (model deployment). Processed using a workflow-based workflow. Returns the response back to the UI.

The following figure shows the AI ​​application architecture. Contoso Creative Writer A multi-agent co-pilot that follows a similar user interaction flow – Except deal with it now. Coordination between multiple agent AI tasks is required. before the final output is generated.

We’re glad you asked! If you’ve explored generative AI application development before, you’ll probably be familiar with this. GenAIOps The application life cycle breaks down developer workflow into three stages: concept (Prototype construction and verification) increase (Iterate and evaluate with larger input dataset) operationalization (deployed into production).

In this blog series, we’ll map this lifecycle into a very clear developer workflow, shown below, to give you an intuitive sense of what to do at each stage and what tools to use to do it.

Start reading the posts in the following order:

1. Azure AI Week begins – Learn about app scenarios, architecture, and life cycle.
2. Provisioning with AZD – Provision Azure infrastructure and set up your development environment.
3. Quick idea creation – Build app prototypes using Prompty assets and tools.
4. Evaluate with AI – Build custom evaluators and use AI-assisted evaluation flows.
5. Deploy using ACA – Create a FastAPI server and deploy using Azure Container Apps.

Below is a visual summary of what you learned:

If you find this series valuable, please star the repository so others can find it!

Want to create your own co-pilot? Take these steps today!

1. Register for the Microsoft AI Tour – Participate in workshop sessions led by instructors.
2. Register for Microsoft Ignite – Find related labs and breakout sessions on Azure AI.
3. Browse our collection of AI templates – Take a look at our samples for more scenarios.

Have a scenario where you’d like to create a custom co-pilot? Have questions about Prompty, Azure AI Studio, or GenAI Ops workflows? Want to provide feedback on a sample? Let us know by leaving a comment here! Happy learning!