Orchestrate a RAG system

Retrieval-Augmented Generation (RAG) systems combine the power of large language models with efficient retrieval mechanisms to enhance the accuracy and relevance of generated responses. By leveraging LangChain for orchestration and Azure AI Foundry for AI capabilities, we can create a robust pipeline that retrieves relevant information from a dataset and generates coherent responses. In this exercise, you will go through the steps of setting up your environment, preprocessing data, creating embeddings, and building a index, ultimately enabling you to implement a RAG system effectively.

Scenario

Imagine you want to build an app that gives recommendations about hotels. In the app, you want an agent that can not only recommend hotels but answer questions that the users might have about them.

You’ve selected a GPT-4 model to provide generative answers. You now want to put together a RAG system that will provide grounding data to the model based on other users reviews, guiding the chat’s behavior into giving personalized recommendations.

Let’s start by deploying the necessary resources to build this application.

Create an Azure AI hub and project

You can create an Azure AI hub and project manually through the Azure AI Foundry portal, as well as deploy the models used in the exercise. However, you can also automate this process through the use of a template application with Azure Developer CLI (azd).

  1. In a web browser, open Azure portal at https://portal.azure.com and sign in using your Azure credentials.

  2. Use the [>_] button to the right of the search bar at the top of the page to create a new Cloud Shell in the Azure portal, selecting a PowerShell environment. The cloud shell provides a command line interface in a pane at the bottom of the Azure portal. For more information about using the Azure Cloud Shell, see the Azure Cloud Shell documentation.

    Note: If you have previously created a cloud shell that uses a Bash environment, switch it to PowerShell.

  3. In the PowerShell pane, enter the following commands to clone this exercise’s repo:

     rm -r mslearn-genaiops -f
 git clone https://github.com/MicrosoftLearning/mslearn-genaiops

  4. After the repo has been cloned, enter the following commands to initialize the Starter template.

     cd ./mslearn-genaiops/Starter
 azd init

  5. Once prompted, give the new environment a name as it will be used as basis for giving unique names to all the provisioned resources.

  6. Next, enter the following command to run the Starter template. It will provision an AI Hub with dependent resources, AI project, AI Services and an online endpoint. It will also deploy the models GPT-4 Turbo, GPT-4o, and GPT-4o mini.

     azd up
  7. When prompted, choose which subscription you want to use and then choose one of the following locations for resource provision:
    • East US
    • East US 2
    • North Central US
    • South Central US
    • Sweden Central
    • West US
    • West US 3

  8. Wait for the script to complete - this typically takes around 10 minutes, but in some cases may take longer.

    Note: Azure OpenAI resources are constrained at the tenant level by regional quotas. The listed regions above include default quota for the model type(s) used in this exercise. Randomly choosing a region reduces the risk of a single region reaching its quota limit. In the event of a quota limit being reached, there’s a possibility you may need to create another resource group in a different region. Learn more about model availability per region

    Troubleshooting tip: No quota available in a given region

    If you receive a deployment error for any of the models due to no quota available in the region you chose, try running the following commands:

      azd env set AZURE_ENV_NAME new_env_name
azd env set AZURE_RESOURCE_GROUP new_rg_name
azd env set AZURE_LOCATION new_location
azd up
      Replacing new_env_name, new_rg_name, and new_location with new values. The new location must be one of the regions listed at the beginning of the exercise, e.g eastus2, northcentralus, etc.

  9. Once all resources have been provisioned, use the following commands to fetch the endpoint and access key to your AI Services resource. Note that you must replace <rg-env_name> and <aoai-xxxxxxxxxx> with the names of your resource group and AI Services resource. Both are printed in the deployment’s output.

     Get-AzCognitiveServicesAccount -ResourceGroupName <rg-env_name> -Name <aoai-xxxxxxxxxx> | Select-Object -Property endpoint
 Get-AzCognitiveServicesAccountKey -ResourceGroupName <rg-env_name> -Name <aoai-xxxxxxxxxx> | Select-Object -Property Key1
  10. Copy these values as they will be used later on.

Set up your local development environment

To quickly experiment and iterate, you’ll use a notebook with Python code in Visual Studio (VS) Code. Let’s get VS Code ready to use for local ideation.

  1. Open VS Code and Clone the following Git repo: https://github.com/MicrosoftLearning/mslearn-genaiops.git
  2. Store the clone on a local drive, and open the folder after cloning.
  3. In the VS Code Explorer (left pane), open the notebook 04-RAG.ipynb in the Files/04 folder.
  4. Run all cells in the notebook.

Clean up

If you’ve finished exploring Azure AI Services, you should delete the resources you have created in this exercise to avoid incurring unnecessary Azure costs.

  1. Return to the browser tab containing the Azure portal (or re-open the Azure portal in a new browser tab) and view the contents of the resource group where you deployed the resources used in this exercise.
  2. On the toolbar, select Delete resource group.
  3. Enter the resource group name and confirm that you want to delete it.