AI Tables enable you to extract meaningful insights and make data-driven decisions from unstructured data by providing structured outputs.

​
Why Use AI Tables?

Traditional SQL excels at querying structured data but cannot directly interpret unstructured data (e.g., free text, audio transcripts, unstandardized records). AI Tables bridge this gap by:
  • Transforming Unstructured Data: AI-powered transformations convert unstructured data into a structured schema
  • Enabling SQL Queries: Once structured, the data can be queried with SQL and joined seamlessly with your existing structured datasets

​
UI

​
Create a Table

This step requires that you have a source created in your workspace. Navigate to “Tables” in the left menu bar, and click on ”+”:

ai table creation

Select the source you wish to creat this table from, then press “Continue”. Verify that all the data sample is correct and customize the name as you wish: table Press “Create Table”.

​
Create an AI Column

To create an AI column, there are two options:
  1. Create a custom column
  2. Use a template

​
1. Create a Custom Column

”+ Add column” > Select template: ”+ Blank column” > “Next” In our example of sales call transcripts, we can create a custom column to indicate whether our call was successful (we define a successful call as one that scheduled a follow-up call):

add column


Column name is the name of your column Column type is the data type of your column (we use “Checkbox” in this case since success can be a true or false value) Description is a detailed description of what you wish to extract Default value is what the value defaults to Select related columns are the columns that are relevant to extracting this data
Column Types We offer various column types, ach tailored to match the specific data type of your output:
  • Text generates free-text outputs a text output, ideal for extracting keywords or phrases without restriction to predefined options
  • Checkbox outputs a boolean (true/false) value, suitable for binary indicators like whether a call was successful
  • Number provides an integer output, useful for counts, such as the frequency a feature is mentioned in a call
  • Currency outputs a floating-point number, perfect for capturing precise values, like prices mentioned
  • List returns a list of strings, ideal for tracking non-standardized items, such as a list of competitor features without a predefined set
  • Select outputs a list of predefined strings, making it easier for standardized analysis, such as listing our own features mentioned during a call (this format is recommended for consistent, analyzable data)

​
2. Use a Template

Templates are pre-defined column formats for common use cases, enabling quicker setup and high-quality outputs. Note: This feature is currently in progress and will be available soon

​
Run your AI Table

Upon creating your AI column, hover over the column name and click on . We get the following options:

add column

We recommend starting with “First 10 rows” to verify if the output meets your expectations, allowing you to make iterations as needed before processing the rest of your dataset. Once you’re satisfied with the results, proceed to “All rows”. You also have the option to run individual rows, this is useful for testing multiple AI columns at a time.

​
CLI

​
Edit YAML file

We can create AI tables by simply changing the type and adding the ai_cols parameter:
ai_sales_calls.yaml
description: AI enhanced version of sales_calls
ai_cols:
  - name: successful
    type: Nullable(Bool)
    description: Whether this call was successful (which is defined as successfully
      setting up a follow up call). If the follow up scheduling is successful, the
      call should include an explicit date/time for the call.
    default_value: '""'
    using:
      - Transcript
  - name: competitors_mentioned
    type: Array(String)
    description: Who are the competitors mentioned (if any)?
    default_value: '"N/A"'
    using:
      - Transcript
  - name: features_mentioned
    type: Array(Enum('Customizable Templates', 'Inventory Management', 'Marketing
      Automation', 'Customer Support Integration', 'Social Media Integration', 'None'))
    description: Which features are mentioned in the call?
    default_value: '["None"]'
    using:
      - Transcript
cols:
  ID: UUID
  Transcript: String
We have a new parameter:
​
ai_cols
array
Columns to be filled by the AI source.

​
Push

Push this update by running: 
af push ai_sales_calls.yaml