Link to the previous post

Flight school students are often overloaded with theory material. Instructors answer the same questions again and again. Approved manuals, notes, and procedures contain the right information, but finding the right section at the right time is not always easy.

That is where AI becomes useful.

Not as a replacement for instructors.

Not as an uncontrolled chatbot.

But as a guided learning assistant that can help students ask questions, retrieve approved content, and receive a clear explanation based on trusted material.

The goal is simple:

Build a controlled AI assistant using Azure AI Foundry that can answer student questions based on approved aviation training material.

The Target Architecture

Before touching the configuration, I always like to define the architecture in plain English. The solution has a few simple parts:

• Approved training documents are uploaded into the environment
• Those documents are processed and made searchable
• A model is deployed inside Azure AI Foundry
• A project is created to manage the AI workload
• The assistant is configured with clear instructions
• The assistant retrieves relevant content before answering
• Information retrived only from approved materials
• The output is tested, reviewed, and improved before being used by students

In simple terms/flow:

This is the important difference between a normal chatbot and a governed AI learning assistant, a normal chatbot answers from its general model knowledge, a grounded assistant answers using the material you provide.

So lets get our hands dirty

Step 1: Create the Azure Resource Group

The first step is to create a dedicated Azure Resource Group to host all components for this AI solution.

For this demo, I will use: <rg-flight-school-ai-foundry-dev>

This Resource Group will keep the environment clean and easier to manage. It will also help with access control, cost tracking, and future cleanup.

All related resources, such as Azure AI Foundry, model deployment, search, storage, and monitoring, can be grouped under the same place.

At this stage, I am only preparing the foundation before building the AI assistant.

Step 2: Create the Storage Account

The second step is to create an Azure Storage Account to store the approved training files.

For this demo, I will use: <stflightschoolaifoundry>

This storage account will hold the documents that will later be used by the AI assistant, such as manuals, notes, procedures, and approved learning material.

Keeping the files in a dedicated storage account makes the solution easier to manage, secure, and update as the knowledge base grows.

At this stage, I am only preparing the document location before connecting it to Azure AI Foundry.

Once the storage account is created, I will create a container and upload the required documents.

For this example, I uploaded several relevant documents. This material covers many of the theory areas needed for the examples in this blog.

If you are following along as a reader, make sure you upload your own approved materials that match your own use case.

Step 3: Create the Azure AI Foundry Project

The first step is to create a project in Azure AI Foundry by accessing the following url: https://ai.azure.com

A project gives you a workspace where you can manage the AI application, model deployments, tools, connections, testing, and monitoring.

For this type of scenario, I would create a dedicated project such as: <FlightSchool-AI-Assistan>

This project is designed to support students with aviation theory questions using approved school material only.

That clarity matters because AI projects can quickly become messy if the project

Step 4: Create and Deploy the Agent

After creating the Azure AI Foundry project, the next step is to create the agent.

The agent is the AI assistant that students will interact with. It will receive the question, search the approved material, and generate a grounded answer.

In the Azure AI Foundry project, I open the Agents section and create a new agent called: <agent-flight-school-theory-assistant>

Then I select the model deployment and add clear instructions:

“You are a flight school theory assistant. Answer using the approved training material only. Do not replace an instructor. Do not invent procedures. If the answer is not found, say so and recommend checking with an instructor.”

Next, I connect the agent to the uploaded knowledge source. For this demo, the source contains several documents which i uploaded at step 2.

Once saved, the agent is ready for initial testing. However, this is only the first configuration stage.

We will return to this agent configuration window later after completing a few required steps, so we can align the agent with the final architecture and adjust the settings properly.

Step 5: Deploy and Configure Azure AI Search

The next step is to deploy Azure AI Search.

Azure AI Search will be used to index the uploaded training material and make it searchable by the agent. This is an important part of the architecture because the agent needs a reliable way to retrieve relevant content before generating an answer.

For this demo, I will create a search service in the same Resource Group: <srch-flight-school-ai-foundry>

After the service is created, I will connect it to the storage account that contains the uploaded documents.

The goal is to allow the document content to be indexed, searched, and later used by the agent as a grounded knowledge source.

At this stage, we are not only deploying a search service. We are preparing the retrieval layer that will help the AI assistant answer from approved material instead of relying only on the model’s general knowledge.

Step 6: Import and Vectorize the Training Data

After the Azure AI Search deployment is completed, I open the search resource and select Import data from the menu.

For the data source, I choose Azure Blob Storage, because this is where I uploaded the training material earlier.

Next, I select the storage account and container created in the previous steps.

For this configuration, I select the RAG option. This prepares the content to be used in a retrieval-augmented generation flow, where the agent can search the material before generating an answer.

I then continue to the Vectorize your text step and complete the configuration:

This step is important because it converts the uploaded document into searchable content that the agent can retrieve and use when answering student questions.

After completing the configuration, you need to wait for the indexer to finish before testing or integrating it with the agent.

Depending on the number and size of your source documents, this process may take some time.

In the example below, I asked the assistant:

How do I teach hover?

The response came back from the uploaded documentations, which confirms that Azure AI Search indexed the document and the agent was able to retrieve relevant content from the approved material.

Step 7: Connect Azure AI Search to the Agent

After the Azure AI Search index is completed, I return to the agent configuration window.

Under the Knowledge section, I select Add, then choose Connect to Foundry IQ.

From there, I select the Azure AI Search service that was created and configured in the previous steps.

This connects the agent to the indexed training material, allowing it to retrieve relevant content from the uploaded documents before generating an answer.

I also make sure to go to the Tools option and disable Web Search.

This is important because I want the agent to retrieve information only from the approved documents, not from the public internet.

At this stage, the agent is no longer working as a general chatbot. It is now connected to the approved knowledge source and ready for grounded testing.

Step 8: Publish and Preview the Web App

After connecting the agent to Azure AI Search and disabling Web Search, the next step is to test the experience.

From the agent configuration window, I select Publish, then choose Preview web app.

This opens a simple web interface where I can test the agent as a student would use it.

For this example, I ask:

How do I teach hover?

The purpose of this test is to confirm that the agent retrieves the answer from the uploaded documents and not from the public internet.

If the answer is clear, relevant, and grounded in the approved material, then the initial configuration is working as expected.

Lets make sure these info comes from the books by asking “Where these info come from” – as you can see the source come from our RAG

Summary

In this blog, I walked through an initial hands-on configuration using Azure AI Foundry.

The goal was not to build a full production solution in one article, but to demonstrate the capability of Azure AI Foundry and how quickly an approved document can become part of a grounded AI assistant.

We created the foundation, uploaded training material, configured Azure AI Search, connected the knowledge source to the agent, disabled Web Search, and tested the experience using the preview web app.

The important point is that the assistant was not answering as a general chatbot. It was retrieving information from the approved document provided in the example.

Where From Here?

As you noticed, this blog post was designed to show the capability of Azure AI Foundry at a high level.

I did not go through every single configuration option in detail, because that was not the main purpose of this post. The purpose was to show how Azure AI Foundry can be used to move from an idea into a working AI assistant that uses approved business content.

From here, the next steps would include improving the instructions, reviewing security, validating the responses, adding more approved documents, monitoring usage, and preparing the solution for a more production-ready deployment.

If you are a business and would like to learn more about Azure AI Foundry, or you are exploring how AI can be safely deployed in your organisation, I would be more than happy to hear from you.