A new feature has just been released in Microsoft Fabric that I was so impressed with that I decided to blog about it. Available in public preview is the ability to create a shortcut to a raw text file and pull the info every two minutes into a Delta Lake table. While pulling in the text, you can perform AI transformations. This removes the need for complex data‑integration pipelines and significantly reduces time to insight.
Here are the supported AI transformations:
- Text Summary: Generates concise summaries from long-form text.
- Translate: Translates text between supported languages.
- Sentiment: Labels text sentiment as positive, negative, or neutral.
- PII detection: Finds and redacts personally identifiable information (names, phone numbers, emails).
- Name recognition: Extracts named entities such as people, organizations, or locations.
It’s very easy to setup: While in Fabric, go to a Lakehouse and under the Files section, select + New Shortcut and create a connection that points to a folder of .txt files in ADLS, S3, GCS or another Fabric workspace. Only .txt files are supported now, but soon JSON and Parquet will be supported. Then in the wizard, select the folder and from Transform, select one of the five AI transformation options listed:
Once you complete creating the shortcut, it will take a few minutes for the initial transform to finish. Each table and folder selected will appear as a unique shortcut in the Files section, under which the transformation will show up in Delta format. The columns in the transformed Delta file/table depend on the AI transform you chose. For example, if you selected Text Summary, the columns would be “text” and “TextSummary”:
If you selected Translate, the resulting columns look like:
If you selected Sentiment, the resulting columns look like:
If you selected PII detection, it does not create a Delta table. Instead, a .txt file is created with the PII info replaced with asterisks:
If you selected Name recognition, the resulting columns look like:
The engine checks the source folder every two minutes. New, modified, or deleted files are reflected in the Delta table. Use the resulting table in reports, notebooks, or downstream pipelines (if using it for reports, you would next need to use the Load to Tables option on the resulting folder where the Delta table is located and choose the Parquet file type to copy it to the Tables section). The status of the transformation can be observed under Manage shortcut:
Since a shortcut points to a folder, and you can only select only one transform format for each folder, you will want to create a folder for each type of transform you will do, and put only the files for that particular transform in it. For example, create a folder called TranslateFiles and put all the files to translate text between languages in that folder, then create a shortcut that uses the Translate transform format. Create another folder called SentimentFiles and put all files to determine the sentiment labels in that folder, then create a shortcut that uses the Sentiment transform format. So if you use all the transform formats, you will wind up with five folders and five shortcuts.
Here are some ideas for using this new feature:
- Customer‑feedback sentiment dashboard: Point a shortcut at yesterday’s review exports, choose Sentiment, and publish a Power BI report that surfaces net sentiment by product, region and time period.
- Multilingual support intelligence: Route global support logs through Translate (to English) followed by optional Text Summary (do this by chaining two shortcuts together), giving managers a single, prioritized list of issues across every market.
- Privacy‑first LLM fine‑tuning: Apply PII detection to call center transcripts to create a compliant dataset that is immediately ready for model training.
- Real‑time market signals: Stream a newswire feed through Name recognition to tag companies, people and locations, enrich your knowledge graph and trigger alerts the moment a relevant story breaks.
This new feature allows you to import unstructured data and semi-structured data into Fabric while transforming it on the fly, all without writing any code or building any pipelines!
Also note there is another feature if you need to create shortcuts to csv files to import into a Delta Lake table.
More info:
Accelerating Insights from Unstructured Text with AI Powered OneLake Shortcut Transformations
AI-powered transforms in OneLake shortcut transformations
The post Microsoft Fabric shortcut‑based AI transformations first appeared on James Serra's Blog.