AI Functions in Microsoft Fabric
Microsoft Fabric is an all-in-one analytics platform that helps teams work with their data across various tools and workflows. With AI being built into every aspect of technology, it’s fascinating to see how AI is being built into Microsoft Fabric, too! With the new AI Functions in Microsoft Fabric, you can bring the power of large language models (LLM) directly into your data pipelines, notebooks, and analytics workflows. Microsoft announced this capability in Public Preview in March 2025. In this post, we’ll breakdown how Microsoft Fabric is elevating its analytics platform with this suite of AI Functions.
What are AI Functions in Microsoft Fabric?
With AI Functions, you can invoke the LLM with just a single line of code. This eliminates the need for complex AI model deployments and enables users with the following capabilities:
- Summarize large amounts of text
- Classify text values based on custom labels
- Translate content into different languages
- Extract specific information from text (like name, etc.,)
In short, AI Functions are like shortcuts to powerful AI tasks without having to build or train any models. You don’t have to be a machine learning expert in order to use this capability. Sounds interesting, right? Let’s get into what AI Functions in Microsoft Fabric are actually useful for.
What AI Functions Can Do?
Say, you have hundreds of support tickets or internal emails in your mailbox. Reading through each one of them is practically impossible. That’s where these functions come in. They help you break down, organize, and make sense of large volumes of text quickly. Below is the table listing the different AI Functions and how you can use them in pandas and PySpark.
| Function | Description | Use Cases |
|---|---|---|
| Summarize | Shortens long text into short summaries | Lengthy company internal emails into short, concise summary Short discharge summary for patients upon checkout with critical information |
| Classify | Categorizes text based on custom labels or tags that you define | Classify support tickets based on severity (urgent, critical, etc.,) Sort customer feedback based on specific tags (feature request, inquiry, etc.,) |
| Extract | Retrieve specific information from input text | Extract name, location from a customer email database List patient name, appointment dates from patient registration database |
| Translate | Convert text from one language to another | Translate customer emails or customer feedback from Spanish to English Translate doctor notes in English into patient’s native language to improve care |
| Similarity | Check two different text and tells you how similar the text is | Find similar customer support tickets that highlight the same problem Detect plagiarism (similar content) in student assignments |
| Sentiment Analysis | Identifies the tone of text – positive, negative, or neutral | Flag customer reviews having words like “unacceptable”, “bad” and address them before they escalate |
| Fix Grammar | Clean up grammar, sentence structure, and punctuations | Clean up customer support tickets before they are assigned to agents |
| Generate Response | Generate custom replies based on the text | Auto-generate custom, professional responses to customer feedback Curate personalized responses to frequently asked questions |
Code Syntax
| Function | pandas Syntax | PySpark Syntax |
|---|---|---|
| Summarize | df[“summaries”] = df[“text”].ai.summarize() | df.ai.summarize(input_col=”text”, output_col=”summaries”) |
| Classify | df[“classification”] = df[“text”].ai.classify(“category1”, “category2”, “category3”) | df.ai.classify(labels=[“category1”, “category2”, “category3″], input_col=”text”, output_col=”classification”) |
| Extract | df_entities = df[“text”].ai.extract(“entity1”, “entity2”, “entity3”) | df.ai.extract(labels=[“entity1”, “entity2”, “entity3″], input_col=”text”) |
| Translate | df[“translations”] = df[“text”].ai.translate(“target_language”) | df.ai.translate(to_lang=”spanish”, input_col=”text”, output_col=”translations”) |
| Similarity | df[“similarity”] = df[“col1”].ai.similarity(“value”) | df.ai.similarity(input_col=”col1″, other=”value”, output_col=”similarity”) |
| Sentiment Analysis | df[“sentiment”] = df[“text”].ai.analyze_sentiment() | df.ai.analyze_sentiment(input_col=”text”, output_col=”sentiment”) |
| Fix Grammar | df[“corrections”] = df[“text”].ai.fix_grammar() | df.ai.fix_grammar(input_col=”text”, output_col=”corrections”) |
| Generate Response | df[“response”] = df.ai.generate_response(prompt=”Instructions for a custom response based on all column values”) | df.ai.generate_response(prompt=”Instructions for a custom response based on all column values”, output_col=”response”) |
How to Get Started?
AI Functions are available in Spark and pandas DataFrames within Fabric.
If you’re working with pandas, you need to install two packages before using AI Functions (one-time process):
- OpenAI package to connect to the model
- AI Functions package to run the AI tasks
If you’re using PySpark, you don’t have to install anything. AI Functions are already pre-installed in Fabric PySpark environment. Refer the Microsoft documentation for the necessary installation commands.
Advantages of AI Functions
- These AI Functions are built directly into Fabric (Notebooks and Pipelines) without the need to deploy any additional AI services
- Makes developers more productive with just a single line of code. You don’t need any ML background to be able to use this.
- AI Functions run within Fabric. You don’t have to leave the Fabric environment to be able to run them. Therefore, you get better governance and security for your data.
- No additional cost for Fabric Users (if you’re already on F64 and higher or P SKU subscriptions)
Prerequisites
- If you are on F64 or higher SKU (or even P SKU), you will be able to run AI Functions. You can use this with smaller capacity, but you must provide your own Azure OpenAPI resources through custom settings as explained in the Microsoft Documentation.
- You need to be on Fabric 1.3 runtime or higher. The library which includes AI functions is preinstalled in the runtime.
- Ensure your administrator enables the tenant switch for Copilot and other features powered by OpenAI
Overall, AI Functions in Fabric is a great example of how AI can save you time and effort. The silver lining is that you don’t need to set up or train AI models or write complex lines of code. If you’re spending time manually sorting through messy text or jumping between different tools to analyze feedback, these functions bring everything into one place (into a single notebook). It’s a more practical way to bring AI into your daily work without overcomplicating things.
If you’re planning to use Microsoft Fabric in your organization and want help getting stated, our team can help you. We help businesses set up Fabric, build interactive reports and manage their data. Learn more about our Microsoft Fabric Consulting Services here.
