Powering Your Data Insights: Integrating OpenAI with Microsoft Fabric

Voiced by Amazon Polly

In today’s data-driven world, organizations are seeking ways to harness the full potential of their data by blending powerful analytics with cutting-edge AI. Microsoft Fabric, when integrated with OpenAI, creates a compelling ecosystem that accelerates insight discovery, automates intelligence, and democratizes AI across business units. Let’s explore how this synergy works and why it matters.

Enhance Your Productivity with Microsoft Copilot

Effortless Integration
AI-Powered Assistance

Get Started Now

What is Microsoft Fabric?

Microsoft Fabric is a unified data platform that brings together data engineering, data science, real-time analytics, and business intelligence in a single SaaS offering. It’s built on top of OneLake, Microsoft’s cloud-scale data lake, and combines the best of Power BI, Azure Synapse, and Data Factory under one roof. Whether you’re an analyst using low-code interfaces or a data engineer writing PySpark code, Fabric offers integrated experiences that streamline data operations and accelerate time to insights. It’s secure, scalable, and deeply connected with the Microsoft ecosystem, making it a go-to platform for enterprises looking to modernize their data stack.

Integrating Microsoft Fabric and Azure OpenAI

By leveraging Azure OpenAI, users can enhance their data workflows with powerful AI capabilities. This integration allows for automated tasks, advanced text generation, and more, making data management and analysis more efficient and insightful.

Microsoft Fabric offers several features that can be enhanced with AI:

Data Cleaning: Automate the process of cleaning and preparing data for analysis.
Text Generation: Create reports, summaries, and content based on data insights.
Data Transformation: Use AI to transform and enrich data for better analysis.
Custom AI Applications: Build applications that leverage AI to provide insights, answer questions, and identify key information.

To integrate OpenAI with Microsoft Fabric, you can use the following libraries:

Azure OpenAI SDK: This library allows you to interact with OpenAI models hosted on Azure.
Microsoft Fabric SDK: This library provides tools to manage and interact with data within Microsoft Fabric.

How to use Azure AI services:

Pre-Built AI Models in Microsoft Fabric (Preview)
Microsoft Fabric offers pre-built AI models currently in preview mode, enabling seamless integration with Fabric authentication and capacity-based billing. These models include capabilities from Azure OpenAI, Text Analytics, and Translator services.

Bring Your Own Key (BYOK) Support
If the required functionality isn’t covered by the pre-built models, you can provision a dedicated AI resource in Azure and integrate it within Fabric. This approach allows for greater customization and control, including BYOK support, where applicable.

Microsoft Fabric integrates Azure OpenAI through the Data Science and Data Engineering experiences, with support for:

PySpark & notebooks using openai, azure-ai-ml, and langchain libraries.
Power BI and Power Query can connect with OpenAI models via Azure ML endpoints.
KQL (Kusto Query Language) in Real-Time Analytics supports calling REST APIs (e.g., OpenAI) using external call functions.
Semantic Link in Fabric simplifies the connection between structured business data (like Power BI datasets) and natural language via LLMs.

Use case using Azure Open AI for Chat completion

The below sample code provides a detailed guide on using the Azure OpenAI service with SynapseML to scale natural language tasks using Apache Spark in Microsoft Fabric

STEP1: Microsoft Fabric subscription (its preinstalled with synapse ML packages for Azure Open AI)

Code to check the packages installed

<em>import synapse.ml.services</em>

<em>print(f"SynapseML cognitive version: {synapse.ml.services.__version__}")</em>

<em> </em>

<em>OUTPUT:</em>

<em>SynapseML cognitive version: 1.0.11.dev1</em>

import synapse.ml.services

print(f"SynapseML cognitive version: {synapse.ml.services.__version__}")

OUTPUT:

SynapseML cognitive version: 1.0.11.dev1

STEP2: Create a Azure Open AI service in your subscription and capture the “service name”, “deployment names” (ex:chat completion , embeddings etc). Make sure the deployment can change time to time on the updates happening.

STEP3: Create a Notebook on the fabric experience and make connections to the Azure OpenAI service created in above step

Code to connect with Azure Open AI. Key is directly hard coded and recommended to use KeyVaults for production environments.

<em>import os</em>

<em>from pyspark.sql import SparkSession</em>

<em>from synapse.ml.core.platform import running_on_synapse, find_secret</em>

<em> </em>

<em># Bootstrap Spark Session</em>

<em>spark = SparkSession.builder.getOrCreate()</em>

<em> </em>

<em>if running_on_synapse():</em>

<em>    from notebookutils.visualization import display</em>

<em> </em>

<em># Fill in the following lines with your service information</em>

<em># Learn more about selecting which embedding model to choose: https://openai.com/blog/new-and-improved-embedding-model</em>

<em>service_name = "</em><em>openaiXXXXX</em><em>"</em>

<em>deployment_name = </em><em>"gpt-35-turbo-instruct</em><em>"</em>

<em>deployment_name_embeddings = "</em><em>text-embedding-ada-002</em><em>"</em>

<em> </em>

<em>key = "</em><em>AGU0zJxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxv</em><em>"</em>

<em> </em>

import os

from pyspark.sql import SparkSession

from synapse.ml.core.platform import running_on_synapse, find_secret

# Bootstrap Spark Session

spark = SparkSession.builder.getOrCreate()

if running_on_synapse():

from notebookutils.visualization import display

# Fill in the following lines with your service information

# Learn more about selecting which embedding model to choose: https://openai.com/blog/new-and-improved-embedding-model

service_name = "openaiXXXXX"

deployment_name = "gpt-35-turbo-instruct"

deployment_name_embeddings = "text-embedding-ada-002"

key = "AGU0zJxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxv"

STEP4: Creating Prompts: We will use the integrated environment to get the datasets to be used for chat completion . But for demonstration we are creating a dataset of prompts and use the OpenAICompletion object to process these prompts.

<em>df = spark.createDataFrame(</em>

<em>    [</em>

<em>        ("Hello my name is",),</em>

<em>        ("The best code is code thats",),</em>

<em>        ("SynapseML is ",),</em>

<em>    ]</em>

<em>).toDF("prompt")</em>

<em> </em>

df = spark.createDataFrame(

[

("Hello my name is",),

("The best code is code thats",),

("SynapseML is ",),

]

).toDF("prompt")

STEP5: Create an OpenAICompletion object with max tokens set to 200 and creating 3 columns with prompt, error and completion.

<em>from synapse.ml.cognitive import OpenAICompletion</em>

<em> </em>

<em>completion = (</em>

<em>    OpenAICompletion()</em>

<em>    .setSubscriptionKey(key)</em>

<em>    .setDeploymentName(deployment_name)</em>

<em>    .setCustomServiceName(service_name)</em>

<em>    .setMaxTokens(200)</em>

<em>    .setPromptCol("prompt")</em>

<em>    .setErrorCol("error")</em>

<em>    .setOutputCol("completions")</em>

<em>)</em>

<em> </em>

from synapse.ml.cognitive import OpenAICompletion

completion = (

OpenAICompletion()

.setSubscriptionKey(key)

.setDeploymentName(deployment_name)

.setCustomServiceName(service_name)

.setMaxTokens(200)

.setPromptCol("prompt")

.setErrorCol("error")

.setOutputCol("completions")

)

STEP6: Transform the dataframe with the OpenAICompletion Client

<em>from pyspark.sql.functions import col</em>

<em> </em>

<em>completed_df = completion.transform(df).cache()</em>

<em>display(</em>

<em>    completed_df.select(</em>

<em>        col("prompt"),</em>

<em>        col("error"),</em>

<em>        col("completions.choices.text").getItem(0).alias("text"),</em>

<em>    )</em>

<em>)</em>

<em> </em>

from pyspark.sql.functions import col

completed_df = completion.transform(df).cache()

display(

completed_df.select(

col("prompt"),

col("error"),

col("completions.choices.text").getItem(0).alias("text"),

)

)

OUTPUT SAMPLE:

Similarly you can apply the Azure OpenAI for embeddings, prompt engineering for Translation or Question Answering .

Conclusion

Whether it’s summarizing massive datasets, generating real-time narratives in Power BI, or building intelligent pipelines that understand context, Azure OpenAI in Fabric empowers data teams to do more with less code and faster deployment. As Microsoft continues to expand native support and tooling, the synergy between generative AI and modern data platforms is poised to redefine how businesses derive value from data.

Become an Azure Expert in Just 2 Months with Industry-Certified Trainers

Career-Boosting Skills
Hands-on Labs
Flexible Learning

Enroll Now

About CloudThat

Established in 2012, CloudThat is an award-winning company and the first in India to offer cloud training and consulting services for individuals and enterprises worldwide. Recently, it won Google Cloud’s New Training Partner of the Year Award for 2025, becoming the first company in the world in 2025 to hold awards from all three major cloud giants: AWS, Microsoft, and Google. CloudThat notably won consecutive AWS Training Partner of the Year (APJ) awards in 2023 and 2024 and the Microsoft Training Services Partner of the Year Award in 2024, bringing its total award count to an impressive 12 awards in the last 8 years. In addition to this, 20 trainers from CloudThat are ranked among Microsoft’s Top 100 MCTs globally for 2025, demonstrating its exceptional trainer quality on the global stage.

As a Microsoft Solutions Partner, AWS Advanced Tier Training Partner, Google Cloud Platform Partner, and collaborator with leading organizations like HPE and Databricks, CloudThat has trained over 850,000 professionals across 600+ cloud certifications, empowering students and professionals worldwide to advance their skills and careers.