I started by creating a new resource group in the Azure portal to contain all the resources for this project.

I searched the Azure Marketplace for "foundry" and found the Microsoft Foundry service.

I filled in the Foundry resource details, naming the resource Haddley-Foundry with a default project of proj-haddley-foundry.

I reviewed the configuration and clicked Create.

The Foundry deployment completed successfully.

I navigated to the Foundry resource overview in the Azure portal.

I opened the Microsoft Foundry portal and saw the project overview with the API key and endpoint details.

A dialog prompted me to select the project to continue with.

The Microsoft Foundry home page loaded, showing the latest model arrivals.

Setting up Storage

I returned to the Azure Marketplace and searched for a storage account to hold the documents I wanted to index.

I configured a new storage account named haddleystorageaccount for machine learning workloads.

I reviewed the storage account settings and clicked Create.

The storage account deployed successfully.

I created a new blob container named haddleystoragecontainer inside the storage account.

The container appeared in the list alongside the system $logs container.

I opened the container and prepared to upload files.

I selected the health-plan PDF documents from my local machine to upload.

All six PDFs uploaded successfully to the container.

Deploying an Embedding Model

I searched the Microsoft Foundry model catalog for the text-embedding-ada-002 model.

I reviewed the text-embedding-ada-002 model details page.

The model deployed successfully with a 500,000 tokens-per-minute rate limit.

Creating an Azure AI Search Service

I searched the Marketplace for Azure AI Search to create a search index.

I configured a search service named haddleyaisearch on the Free pricing tier.

The search service deployed successfully.

I viewed the haddleyaisearch overview showing the service is Running.

Indexing with RAG

I clicked Import data (new) and selected Azure Blob Storage as the data source.

I selected RAG as the scenario to enable AI-powered answers.

I connected the wizard to my haddleystoragecontainer.

The vectorization step initially showed no Azure OpenAI service available, so I needed to create one.

I filled in the Create Azure OpenAI form with the name haddley-azure-openai on Standard S0.

I reviewed and submitted the Azure OpenAI deployment.

The deployment started and the resource was being created.

Back in the RAG wizard, the haddley-azure-openai service appeared but had no deployments yet.

I navigated to the Azure OpenAI model catalog in Foundry and found text-embedding-ada-002.

I deployed the model with a Standard GlobalStandard deployment type.

Returning to the RAG wizard, text-embedding-ada-002 was now available to select.

I skipped image vectorization and moved to the advanced settings.

I enabled the semantic ranker and kept the indexing schedule set to Once.

I reviewed the RAG configuration and clicked Create.

The index was created successfully and indexing began.

I opened the Search explorer and tested the index with a sample question about health insurance costs.

Creating the Agent

Back in the Foundry home, I clicked Start building and selected Create agent.

I named the new agent haddley-health-plan-agent.

The agent playground opened and I clicked to add a tool.

I selected Azure AI Search from the tool catalog.

I confirmed the Azure AI Search tool selection.

No existing connections were available so I created a new one.

I connected to the haddleyaisearch service using an API key.

The rag-1773125823872 index appeared and I selected it.

With the Azure AI Search tool connected and pointing at the RAG index, I typed a question in the chat.

The agent responded with a clear answer citing the Benefit_Options.pdf document.