What's new and planned for Fabric Data Science in Microsoft Fabric
Important
The release plans describe functionality that may or may not have been released yet. The delivery timelines and projected functionality may change or may not ship. Refer to Microsoft policy for more information.
Fabric Data Science provides data scientists with an end-to-end workflow for building their machine learning models, from exploration to model scoring. From a data exploration perspective, data scientists can use R and Python in notebooks, and built-in tools like Data Wrangler for easy analysis. Users can track and compare their model experiments and runs with MLFlow. They can save the best performing model in the workspace as a new model item and easily use Predict for batch scoring at scale. Data science in Fabric is deeply integrated with the rest of the stack, meaning it's seamless to score data in a lakehouse, write back the predictions to OneLake, and visualize the data in reports using Direct Lake mode.
Investment areas
AI Functions for LLM-Powered Text Enrichment and Transformation [Public Preview]
Estimated release timeline: Q4 2024
Release Type: Public preview
AI functions in Fabric will allow notebook users to seamlessly perform tasks like text summarization, translation, classification, sentiment analysis, grammar correction, and more, providing a simplified API for common enrichments and making it easier for users to apply them with fewer lines of code. The functions will be initially available on top of pandas DataFrames and ultimately available via Spark, SQL, and other programming surfaces across Fabric.
Low Code AutoML
Estimated release timeline: Q4 2024
Release Type: Public preview
Our Low-Code AutoML tool empowers data scientists and analysts to easily create machine learning models without the need for extensive coding. Through an intuitive, step-by-step wizard, users can configure and launch AutoML trials directly from the user interface.
AI skill integration with Azure AI Foundry
Estimated release timeline: Q1 2025
Release Type: Public preview
With the Fabric AI skill integration in Azure AI Foundry, Fabric AI Skill will serve as a knowledge source for Agent Service in Microsoft Azure AI Foundry. This enables the agent to use Fabric as a data hub, tapping into the insights available within Fabric to answer user queries accurately and efficiently. By connecting to Fabric AI Skill, the agent can retrieve data insights directly from Fabric, allowing consumers to interact with and analyze their Fabric data seamlessly through the AI applications in Azure AI Foundry.
Semantic Models as new data source for AI Skill
Estimated release timeline: Q1 2025
Release Type: Public preview
This feature allows users to query their Power BI Semantic Models in Fabric using natural language, receiving both a concise answer and the corresponding DAX query. Users can ask questions like “What were the total sales over the last 12 months?” and get not only the result but also the underlying DAX query for transparency and reuse. In future, user should also be able to provide few-shot examples—sample questions- to guide the AI Skill that semantic model is the best tool to answer those questions. This approach makes data insights more accessible to all users while providing advanced users with greater control and transparency over the analysis.
KQL database as new data source in AI Skill
Estimated release timeline: Q1 2025
Release Type: Public preview
This feature allows users to query their Kusto databases in Fabric using natural language, receiving both a concise answer and the corresponding KQL (Kusto Query Language) query. Users can ask questions like “What was the total number of logins last week?” and get not only the result but also the underlying KQL query for transparency and reuse. To enhance accuracy, users can provide few-shot examples—sample questions with expected answers. The system supports iterative queries, enabling users to refine their questions or update notes for more precise outputs, making data analysis more accessible while empowering advanced users with greater control.
AI Skill becomes a conversational AI agent
Estimated release timeline: Q1 2025
Release Type: Public preview
The AI Skill is now conversational, enabling users to engage in natural, back-and-forth dialogue to explore and understand their data with ease. This enhancement allows users to ask follow-up questions, refine queries, and receive dynamic insights, making data exploration more intuitive and interactive.
Low-Code AI-Powered Operations in Data Wrangler [Public Preview]
Estimated release timeline: Q1 2025
Release Type: Public preview
A new suite of AI-powered operations in Data Wrangler will allow users to describe code transformations with natural language and generate the corresponding Python; translate custom Python code into PySpark code; and apply SynapseML transformations like text translation and sentiment analysis in a matter of clicks.
Copilot for Data Science/Data Engineering references Fabric Documentation
Estimated release timeline: Q1 2025
Release Type: Public preview
We are excited to announce a new feature in Fabric Copilot for Data Science and Data Engineering. The Copilot can now access Fabric documentation and reference it in its replies, providing users with relevant information directly within their workflow.
Key Highlights:
- Seamless Integration: Copilot in DS/DE now integrates with Fabric documentation, offering contextual assistance and detailed information without leaving the workspace.
- Enhanced Productivity: By referencing Fabric documentation, Copilot in DS/DE helps users quickly find answers, reducing search time and increasing productivity.
- Contextual Assistance: Copilot in DS/DE provides precise documentation references to support data analysis, visualization, and engineering tasks.
The new feature in Fabric Copilot for Data Science and Data Engineering empowers users with the information they need, right when they need it.
Real-Time Endpoints for Machine Learning Models [Public Preview]
Estimated release timeline: Q2 2025
Release Type: Public preview
Alongside the existing functionality for batch scoring with PREDICT, Fabric will allow data scientists to serve real-time predictions from any registered ML Model using secure, scalable online endpoints that are automatically configured. These endpoints can be called from other Fabric engines or from external apps, allowing users to deploy their models for wide, reliable consumption.