# Unstructured Data Extraction Using AI Functions in Dataflow Gen2

Three and a half years ago, in pre-AI age, I wrote a blog post titled “**Extracting Matching Words Using A Pre-Defined List in Power Query/M”.** I helped a colleague extract certain error codes/phrases from sentences using M. It felt great to wield the M sword. Fast forward to today, you can use the newly announced [AI Functions feature in Dataflow Gen2](https://learn.microsoft.com/en-us/fabric/data-factory/dataflow-gen2-ai-functions) to achieve the same result using natural language.

Blog: [Extracting Matching Words Using A Pre-Defined List in Power Query/M | Sandeep Pawar](https://pawarbi.github.io/blog/powerquery/m/list/2022/03/11/powerquery-M-extracting-words.html)

Following the same example I shared in the blog above, I wanted to extract certain keywords (programming languages) from text. In M, I split the text and did list lookup to find the matching words.

![](https://raw.githubusercontent.com/pawarbi/blog/master/images/word13.png align="left")

## Fabric AI Prompt:

This shouldn’t need much explanation. Similar to [AI Functions in Fabric](https://learn.microsoft.com/en-us/fabric/data-science/ai-functions/overview?tabs=pandas-pyspark%2Cpandas), you specify the column, enter the prompt and let the LLM do the work.

Here is the prompt I used : `Extract any programming, query, or scripting languages mentioned (including those used in BI tools such as M, LOD ). Comma-separate if multiple. NA if none or unsure`

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1764971226777/b5d14a3e-2db9-459a-8672-31eaa36f0da9.png align="center")

**Result:**

![](https://cdn.hashnode.com/res/hashnode/image/upload/v1764971675031/1ea4d254-5703-4bfa-a177-f46360b27c6b.png align="center")

Give it a try:

```python
let
    Source = #table(
        type table [Respondent = text, Text = text],
        {
            {"Person1", "I like Python"},
            {"Person2", "We use Python and R"},
            {"Person3", "SQL"},
            {"Person4", "My team uses Spark , SQL"},
            {"Person5", "Excel all the way"},
            {"Person6", "I hate DAX"},
            {"Person7", "M is magic"},
            {"Person8", "I don't use anything"},
            {"Person8", "I learned MDX before DAX and still use MDX for our SSAS cubes "}
        }
    ),
  #"Renamed columns" = Table.RenameColumns(Source, {{"Text", "quote"}}),
  #"Added AI prompt column" = Table.AddColumn(#"Renamed columns", "language", each FabricAI.Prompt("Extract any programming, query, or scripting languages mentioned (including those used in BI tools such as M, LOD ). Comma-separate if multiple. NA if none or unsure", Record.SelectFields(_, {"quote"})), type text)
in
    #"Added AI prompt column"
```

Instead of the UI, you can also use [FabricAI.Prompt](https://learn.microsoft.com/en-us/powerquery-m/fabricai-prompt) M function. Note that this function is not available in Power BI Desktop.

Read the documentation for more details: [Fabric AI Prompt in Dataflow Gen2 (Preview) - Microsoft Fabric | Microsoft Learn](https://learn.microsoft.com/en-us/fabric/data-factory/dataflow-gen2-ai-functions)