What You Will Build
Most RAG tutorials skip the part that actually trips people up: the cloud setup. This guide walks through every screen in the Azure portal so you can stand up the retrieval layer for an agentic project without guessing.
Developers and students building their first retrieval-augmented or agentic application on Azure, especially those working in OutSystems and ODC.
A free Azure account, a resource group, an AI Search service with a vector index, and a deployed text-embedding-3-small model.
Six credentials: two endpoints, two keys, an index name, and a deployment name. Drop them into your application and your pipeline can retrieve.
Sign Up for a Free Azure Account
Before you can build anything, you need a verified Azure subscription. This part gets you from zero to a working portal with $200 in free credits.
Sign Up or Log In at portal.azure.com
Open portal.azure.com in your browser. If you already have a Microsoft account, log in. If not, start the sign-up flow from the welcome screen.
Click "Start with an Azure Free Trial"
On the landing screen, select the first card — Start with an Azure free trial — to begin the free trial path.
Create Your Azure Free Account
Click the second card — Azure free account — to create a brand-new free account with $200 credit valid for 30 days.
Enter Your Details
Provide your name, email address, phone number, and address, then click Next through each step.
Add a Payment Method and Sign Up
Enter your credit card information, click Next, and complete sign-up.
Enable Multi-Factor Authentication
If prompted to enable MFA, go ahead and set it up. It keeps the account secure and is increasingly required by Azure.
Land on the Azure Home Page
You should be redirected to portal.azure.com/home and see the welcome message. Your subscription is now live.
Create a New Resource Group
A resource group is a container that holds everything for this project, so billing, access control, and clean-up stay in one place. Always group related Azure resources together.
Open Resource Groups
From the left-side panel of the Azure portal home page, click Resource groups.
Create a New Resource Group
Click Create (or the + Create button at the top) to start a fresh resource group.
Name It and Create
Fill out the form with the following settings, then click Review + Create:
- Subscription — Azure subscription 1
- Resource group name — e.g.
azure-ai-resource-group - Region — (US) East US
Create the Azure AI Search Service and Index
This is the retrieval backbone of your RAG pipeline. You will provision the search service, then define an index that stores both text and 1536-dimension vector embeddings.
Create a Resource
From the home page, click Create a resource in the left-side panel.
Find Azure AI Search
In the search bar, type AI Search and select Azure AI Search from the Marketplace results.
Fill Out the Service Form
Configure the new AI Search service with these settings:
- Select your Subscription
- Select the resource group you just created
- Give a unique service name (e.g.
lowcademy-search-service) - Set Location to (US) East US
- Set Pricing tier to FREE
- Click Review + Create
- Wait 2–3 minutes for deployment to complete
Open the New Search Service
Go back to the home page, search for the search service you just created, and open it. You should see the service Overview with the URL shown in Essentials.
Go to Indexes
In the left navigation of your search service, under Search management, click Indexes to create a new index.
Add an Index via JSON
Click + Add index → Add index (JSON) to open the JSON editor.
Paste the Index Definition and Save
Paste the JSON below into the JSON box and click Save. This defines your fields, the 1536-dimension vector field for embeddings, the semantic configuration, and the HNSW vector search profile.
{
"name": "ai-search-index",
"fields": [
{
"name": "id",
"type": "Edm.String",
"key": true,
"searchable": false,
"filterable": false,
"retrievable": true,
"stored": true,
"sortable": false,
"facetable": false
},
{
"name": "tenant_id",
"type": "Edm.Int64",
"searchable": false,
"filterable": true,
"retrievable": true,
"stored": true,
"sortable": true,
"facetable": false
},
{
"name": "document_id",
"type": "Edm.Int64",
"searchable": false,
"filterable": true,
"retrievable": true,
"stored": true,
"sortable": true,
"facetable": false
},
{
"name": "sourceName",
"type": "Edm.String",
"searchable": false,
"filterable": true,
"retrievable": true,
"stored": true,
"sortable": true,
"facetable": false
},
{
"name": "section",
"type": "Edm.String",
"searchable": false,
"filterable": true,
"retrievable": true,
"stored": true,
"sortable": true,
"facetable": false
},
{
"name": "content",
"type": "Edm.String",
"searchable": true,
"filterable": false,
"retrievable": true,
"stored": true,
"sortable": false,
"facetable": false
},
{
"name": "embedding",
"type": "Collection(Edm.Single)",
"searchable": true,
"retrievable": true,
"stored": true,
"dimensions": 1536,
"vectorSearchProfile": "default-vector-profile"
}
],
"scoringProfiles": [],
"suggesters": [],
"analyzers": [],
"normalizers": [],
"tokenizers": [],
"tokenFilters": [],
"charFilters": [],
"similarity": {
"@odata.type": "#Microsoft.Azure.Search.BM25Similarity"
},
"semantic": {
"configurations": [
{
"name": "default",
"rankingOrder": "BoostedRerankerScore",
"prioritizedFields": {
"titleField": { "fieldName": "section" },
"prioritizedContentFields": [ { "fieldName": "content" } ],
"prioritizedKeywordsFields": [ { "fieldName": "sourceName" } ]
}
}
]
},
"vectorSearch": {
"profiles": [
{ "name": "default-vector-profile", "algorithm": "default-hnsw" }
],
"algorithms": [
{ "name": "default-hnsw", "kind": "hnsw" }
]
}
}
text-embedding-3-small model outputs 1536-dimension vectors. The index field must match this dimension count exactly for vector search to work.
Verify the Index Fields
After saving, open the newly created index, go to the Fields tab, and confirm the fields match the JSON you pasted — including the embedding field with dimension 1536.
Grab the URL and Admin Key
You now need two values from this service:
- URL — visible on the search service Overview page (format:
https://<name>.search.windows.net). Copy and save it. - Admin key — go to Security + Networking › Keys and copy the Primary admin key.
Create the Azure OpenAI Resource for Embeddings
Retrieval needs vectors. Here you provision Azure OpenAI and deploy the embedding model that turns your text into the 1536-dimension vectors your index expects.
Find Azure OpenAI
- From the portal search bar, type OpenAI
- Click Azure OpenAI from the results
Create the Resource
- Click Create → choose Azure OpenAI
- Fill the form: select your subscription, resource group, set a unique name, choose region (US) Central US, pricing tier Standard S0
- Click Next through Network and Tags, then click Create
- Wait 3–4 minutes for the deployment to complete
Open the Foundry Portal
- Open the newly created Azure OpenAI resource
- Click Go to Foundry portal
Open the Model Catalog
In the Foundry portal, under the new resource, open the Model catalog.
Search for the Embedding Model
In the model catalog search box, type text-embedding-3-small and select it from the results. You should see it listed under Embeddings.
Select the Model
- Select text-embedding-3-small
- Click Use this model
Deploy the Model
Configure the deployment with these settings:
- Deployment name —
text-embedding-3-small - Deployment type — Standard
- Resource location — East US 2 (pre-selected)
- Click Create resource and deploy
Understand What Just Happened
Go to Overview and check the resources. You will likely see two Azure OpenAI resources.
- • Billing
- • Access control (IAM)
- • Monitoring
- • Azure portal management
- • Model deployments
- • Endpoint URL
- • API keys
Go to the second (Foundry-created) resource, navigate to Models & Endpoints, and copy the Endpoint URL and API key. Use these in your application for embedding requests.
Get Your Endpoint and Key
From the Foundry-created resource's deployment page, copy the following three values:
Your Six Credentials
You should now have all six values ready. Drop them into your application — your RAG pipeline can retrieve.
text-embedding-3-small model is deployed and live.
The next step is to start feeding your documents in — chunk, embed, and upload to the index.