AI-Powered Natural Language Querying for Azure SQL Data Warehouse

Question

AI-Powered Natural Language Querying for Azure SQL Data Warehouse

Yernaidu Siraparapu 125

Requirement: AI-Powered Natural Language Querying for Existing Azure SQL Database

We have an existing Azure SQL database containing Fact and Dimension tables that are already structured for reporting and analytics.

We want to add AI capabilities on top of this existing database so that business users can interact with the data using natural language instead of writing SQL queries.

For example, a user should be able to ask questions such as:

"How many sales did we have last month?"

"What were our total sales this year?"

"Which product generated the highest sales?"

"What were the sales for Australia in July?"

"Compare sales between 2025 and 2026."

"Show me the top 10 customers by sales."

The AI agent should understand the user's question, identify the relevant Fact and Dimension tables, generate the appropriate SQL query, execute the query against the Azure SQL database, analyze the returned results, and provide a clear natural-language answer.

Expected Flow

User Question

→ "How many sales did we have last month?"

↓

AI Agent

Understands the question and identifies the relevant tables, relationships, measures, and filters.

↓

SQL Generation

The agent generates an appropriate SQL query, for example:

SELECT SUM(SalesAmount)
FROM FactSales
JOIN DimDate
    ON FactSales.DateKey = DimDate.DateKey
WHERE DimDate.Month = ...

↓

Azure SQL Database

The generated query is executed against the existing Fact and Dimension tables.

↓

Result

Azure SQL returns the actual data.

↓

AI Agent

Interprets the result and responds:

"Total sales last month were $2.4 million."

Key Requirement

We do not want to replace or redesign our existing Azure SQL database.

We want to understand how we can add an AI/Data Agent capability on top of the existing Fact and Dimension model, while ensuring that:

The agent queries the actual Azure SQL data rather than relying on general LLM knowledge.

The agent understands our Fact/Dimension relationships and business definitions.

Natural-language questions can be converted into reliable SQL queries.

The SQL is executed against the live/current database data.

The response is based on the actual query results.

Database security and existing permissions are respected.

We can control which tables and data the AI agent is allowed to access.

The solution can handle business terminology, metrics, filters, and relationships correctly.

The solution provides reliable answers and minimizes hallucinations.

Main Question

What is the recommended Microsoft/Azure architecture and technology for implementing a conversational AI/Data Agent on top of our existing Azure SQL database, where users can ask natural-language questions and the agent generates and executes SQL against our Fact and Dimension tables to return accurate business answers?

We would also like to understand whether we should use Microsoft Fabric Data Agent, Microsoft Foundry Agent Service with SQL/MCP capabilities, or another Azure-native approach, and what changes, if any, are required to our existing Azure SQL database to support this capability.

Requirement: AI-Powered Natural Language Querying for Existing Azure SQL Database

We have an existing Azure SQL database containing Fact and Dimension tables that are already structured for reporting and analytics.

We want to add AI capabilities on top of this existing database so that business users can interact with the data using natural language instead of writing SQL queries.

For example, a user should be able to ask questions such as:

"How many sales did we have last month?"

"What were our total sales this year?"

"Which product generated the highest sales?"

"What were the sales for Australia in July?"

"Compare sales between 2025 and 2026."

"Show me the top 10 customers by sales."

The AI agent should understand the user's question, identify the relevant Fact and Dimension tables, generate the appropriate SQL query, execute the query against the Azure SQL database, analyze the returned results, and provide a clear natural-language answer.

Expected Flow

User Question

→ "How many sales did we have last month?"

↓

AI Agent

Understands the question and identifies the relevant tables, relationships, measures, and filters.

↓

SQL Generation

The agent generates an appropriate SQL query, for example:

SELECT SUM(SalesAmount)
FROM FactSales
JOIN DimDate
    ON FactSales.DateKey = DimDate.DateKey
WHERE DimDate.Month = ...

↓

Azure SQL Database

The generated query is executed against the existing Fact and Dimension tables.

↓

Result

Azure SQL returns the actual data.

↓

AI Agent

Interprets the result and responds:

"Total sales last month were $2.4 million."

Key Requirement

We do not want to replace or redesign our existing Azure SQL database.

We want to understand how we can add an AI/Data Agent capability on top of the existing Fact and Dimension model, while ensuring that:

The agent queries the actual Azure SQL data rather than relying on general LLM knowledge.

The agent understands our Fact/Dimension relationships and business definitions.

Natural-language questions can be converted into reliable SQL queries.

The SQL is executed against the live/current database data.

The response is based on the actual query results.

Database security and existing permissions are respected.

We can control which tables and data the AI agent is allowed to access.

The solution can handle business terminology, metrics, filters, and relationships correctly.

The solution provides reliable answers and minimizes hallucinations.

Main Question

What is the recommended Microsoft/Azure architecture and technology for implementing a conversational AI/Data Agent on top of our existing Azure SQL database, where users can ask natural-language questions and the agent generates and executes SQL against our Fact and Dimension tables to return accurate business answers?

We would also like to understand whether we should use Microsoft Fabric Data Agent, Microsoft Foundry Agent Service with SQL/MCP capabilities, or another Azure-native approach, and what changes, if any, are required to our existing Azure SQL database to support this capability.

SIVASANKAR YEDDULA 150 Reputation points Microsoft External Staff Moderator

2026-08-24T14:53:04.8466667+00:00
Hi Yernaidu Siraparapu,

Thank you for reaching out to the Microsoft Q&A forum.

Based on the requirements provided, the recommended Microsoft-native architecture is Azure AI Foundry Agent Service integrated with Azure SQL Database.

Recommended Architecture

Business User

→ Natural Language Question

Azure AI Foundry Agent Service

→ Understands the business question

→ Identifies relevant Fact and Dimension tables

→ Generates the required SQL query using schema metadata, relationships, and business definitions

Azure SQL Database

→ Executes the generated SQL against the live/current database

AI Agent

→ Analyzes the query results

→ Returns a natural-language response to the user

Example

User asks:

*"What were the sales for Australia in July?"

The agent:

Identifies the relevant Fact and Dimension tables.

Generates the appropriate SQL query.

Executes the query against Azure SQL Database.

Returns a response such as:

"Total sales for Australia in July were $2.4 million."

Benefits of This Approach

No redesign of the existing Fact/Dimension model is required.

Queries run against the **live Azure SQL Database.

Existing Azure SQL security and permissions are preserved.

Access can be restricted to approved tables, views, and stored procedures.

Responses are based on actual query results, helping minimize hallucinations.

Business terminology, metrics, filters, and relationships can be incorporated through metadata and agent instructions.

Recommended Preparation

To improve accuracy and reliability, consider providing:

Fact and Dimension table descriptions.

Business metric definitions.

Table relationships and join keys.

Reporting views for commonly used measures.

Business terminology and synonyms.

Example natural-language questions and expected outputs.

Restricted access to approved reporting objects.

Microsoft Fabric Data Agent

If your organization is already using Microsoft Fabric, Fabric Data Agent is also a valid option. However, it typically works with Fabric-managed or mirrored data assets.

For your requirement of querying the existing Azure SQL Database directly and using live/current data, Azure AI Foundry Agent Service would be the recommended approach.

Final Recommendation

For a solution that supports:

Natural language → SQL generation

Live Azure SQL data access

Existing Fact/Dimension model preservation

Security and governance controls

Accurate business answers based on actual data

Azure AI Foundry Agent Service + Azure SQL Database is the recommended Microsoft-native architecture.

If this answer helps, please consider clicking Accept Answer so that other community members with similar requirements can benefit from it as well.

If you have any further questions, please let us know. We will be happy to assist.
SIVASANKAR YEDDULA 150 Reputation points Microsoft External Staff Moderator

2026-08-25T17:23:44.55+00:00

Hi @Yernaidu Siraparapu ,

I hope the information provided has been helpful. If you have any further questions or need additional assistance, please let us know in the comments, and we will be glad to help. If this answer helps, please consider clicking Accept Answer so that other community members with similar questions can benefit as well.
SIVASANKAR YEDDULA 150 Reputation points Microsoft External Staff Moderator

2026-08-27T16:53:38.3233333+00:00

Hi @Yernaidu Siraparapu ,

I hope the information provided has been helpful. If you have any further questions or need additional assistance, please let us know in the comments, and we will be glad to help. If this answer helps, please consider clicking Accept Answer so that other community members with similar questions can benefit as well.

1 answer

Your answer

SIVASANKAR YEDDULA 150 Reputation points Microsoft External Staff Moderator

2026-08-25T17:23:44.55+00:00

Hi @Yernaidu Siraparapu ,

I hope the information provided has been helpful. If you have any further questions or need additional assistance, please let us know in the comments, and we will be glad to help. If this answer helps, please consider clicking Accept Answer so that other community members with similar questions can benefit as well.
SIVASANKAR YEDDULA 150 Reputation points Microsoft External Staff Moderator

2026-08-27T16:53:38.3233333+00:00

Hi @Yernaidu Siraparapu ,

I hope the information provided has been helpful. If you have any further questions or need additional assistance, please let us know in the comments, and we will be glad to help. If this answer helps, please consider clicking Accept Answer so that other community members with similar questions can benefit as well.

Answer 1

Based on the requirements you described, your existing Fact and Dimension model in Azure SQL Database is already a good foundation and does not need to be redesigned.

From a Microsoft architecture perspective, I would consider the following options:

Option 1: Microsoft Fabric Data Agent (Recommended for BI and Analytics Scenarios)

If your primary use case is business users asking analytical questions such as sales, revenue, top customers, trends, and comparisons, Fabric Data Agent is likely the most straightforward Microsoft-native option.

In this approach:

Expose your Azure SQL data through Fabric.
Define the business model, relationships, measures, and semantics.
Configure the Data Agent to use those business definitions.
Users ask questions in natural language.
The agent translates the request into the appropriate data query and returns a business-oriented response.

This approach is particularly effective when:

The data model already follows a dimensional design.
Business metrics require consistent definitions.
You want to reduce the risk of hallucinated SQL and business logic.
Users primarily need reporting and analytical insights.

Option 2: Azure AI Foundry Agent Service + Azure SQL (Recommended for Custom Conversational Experiences)

If you need a fully customized conversational agent that:

Generates SQL dynamically
Executes queries directly against Azure SQL
Applies custom business rules
Integrates with Teams, web applications, or internal systems
Uses additional enterprise knowledge sources

then Azure AI Foundry Agent Service is generally the more flexible architecture.

Typyical flow would be:

User Question -> Azure AI Foundry Agent -> Schema/Business Context -> SQL Generation -> Azure SQL Database -> Query Results -> AI Reasoning & Response -> User

In this model, the agent should be provided with:

Table metadata
Fact-to-dimension relationships
Business glossary and metric definitions
Approved schemas/views
Access control rules

This helps ensure the generated SQL aligns with business semantics rather than relying solely on LLM reasoning.

To determine whether Fabric Data Agent or Azure AI Foundry Agent Service is the better fit, it would help to understand:

Is the source Azure SQL db or Synapse Dedicated SQL Pool?
Do you require row-level security (RLS) for different users?
Are business metrics such as Revenue, Sales, and Margin already formally defined?
Will users interact through Teams, PBI, a web application, or another interface?
Do you already have Fabric capacity/licensing, or must the solution operate directly against Azure SQL?

Help make this community better for everyone: if this answer resolved your issue, please accept it or leave an upvote. If not, share more details in a comment so we can continue the discussion and find the right solution.