Smart Columns Overview

Transform and enrich your data using mappings, formulas, or LLM prompts

If you want to get truly valuable insights from your data, step one is making sure it’s clean, structured, and enriched with the right information. That’s exactly why we built Smart Columns.

With smart columns, you can:

• Recode values into cleaner categories
• Apply formulas to transform data
• Extract new attributes like emotions or brand mentions

There are three modes you can choose from:

Mappings

Formulas

LLMs

Let’s go through them step by step.

Quick start

Regardless of the mode you choose, creating your first Smart Column is easy:

1. Pick your source column → the data you want to transform (e.g. Language, Text to analyze, or Country).
2. Choose a mode → Mapping, Formula, or LLM, depending on your goal.
3. Configure the transformation → add logic, mappings, prompts, or formulas.

How to select your source column

Before anything can be transformed, you need to tell Caplena which column to use.

Here’s how:

1. Click into the Expression for “Source Value” field.

2. Click the {ƒx} icon to the right — this opens the variable selector.

3. Choose from available project variables (like Text to analyze, Country, etc.).

4. Caplena will insert the correct variable syntax automatically.

1. Mapping mode

Got inconsistent values that need standardizing? Mapping mode is your best friend. You define a source value and map it to a desired output.

Example

• LV → Latvian
• ❓ Anything else → becomes your fallback value (e.g., “Other”)

You can enter mappings manually or upload them as a JSON dictionary, perfect for large sets.

 Once set up, this column auto-updates whenever you upload new data.

2. Formula mode

Want to reverse a string, calculate something, or clean up a date? Use Formula mode, powered by Jinja syntax.

Not a coding wizard? No problem! Ask ChatGPT to write formulas for you.

Example: Reverse a String

To reverse the content of a column (e.g., Text to analyze), use:

This formula:

• Converts the string into a list of characters
• Reverses that list
• Joins it back into a string

 Output: Caplena is awesome → emosewa si anelpaC

Jinja Flavor & Sandbox

Formula mode runs on Jinja2 using a sandboxed environment (SandboxedEnvironment). This means all standard Jinja2 features are available — conditionals, loops, filters, string/number/date manipulation — while unsafe operations (arbitrary Python execution, file access, etc.) are blocked for security. A few things to keep in mind:

• Variables that are missing for a row (e.g. a column has no value for that row) are treated as falsy in {% if %} checks and as not-equal in comparisons, rather than raising an error.
• Output is always coerced to a string before being parsed into the column’s target type (Text, Number, Boolean, Date).
• Empty string output is treated the same as no value — the cell will be left blank.

Custom Filters

On top of standard Jinja2 filters, Caplena provides the following built-in custom filters:

address_to_coordinates

Geocodes a free-text address or place name to its [longitude, latitude] coordinates using the Mapbox API. Usage:

{{ col_abc123 | address_to_coordinates }}

Returns: A list [longitude, latitude] as floats, e.g. [8.5417, 47.3769]. To extract individual components, use Jinja’s index access:

{{ (col_abc123 | address_to_coordinates)[0] }}  {# longitude #}
{{ (col_abc123 | address_to_coordinates)[1] }}  {# latitude #}

Note: For an accurate coordinate lookup, make sure to always provide full addressees, including the country as input value.

Formulas are incredibly flexible for working with numbers, text, dates, or any custom logic you need.

3. LLM mode

This mode applies AI models (LLMs) to process free-text inputs, enabling more advanced use cases like summarization or classification.

Common use cases:

• Extracting brand names
• Detecting emotion or sentiment
• Removing slurs or sensitive content
• Creating summaries

Templates Available

Caplena provides a growing template library with ready-to-use prompts for common use cases. These templates help you:

• Get started quickly
• Ensure consistent and high-quality results
• Customize further if needed

You can start from a template and then edit the prompt to match your specific use case.

Config history

Every time you change a Smart Column’s configuration, Caplena keeps the previous version. Config history lets you browse those past versions, compare them with your current setup, and roll back if needed.

Where to Find It

Open a Smart Column’s detail page. A Config history button appears in the header once the column has at least one earlier version — for a brand-new column with no changes yet, the button stays hidden.

Compare Versions

Config history shows a timeline that runs from when the column was created through to the Current version. Use the Compare with selector (or the timeline) to pick any earlier version, then read the two panels side by side: the selected historical configuration on the left and your current configuration on the right. This works for all three modes — Mapping, Formula, and LLM — so you can see exactly what changed (timestamps are shown in your local timezone).

Restore a Version

To roll back, select the version you want and click Restore Version. Caplena applies that historical configuration and recomputes the column for new rows.

Restoring is disabled while the column is still computing. Wait for the current run to finish before rolling back.

Best practices

• Start small: Test your logic on a small dataset or with a few rows using the preview.
• Use templates: Especially for LLM tasks, templates can save time and provide a solid base.
• Use fallback values: Always define a fallback in mappings to handle unexpected inputs.
• Name your columns clearly: This helps when chaining multiple Smart Columns together.
• Chain columns: You can use the output of one Smart Column as the input for another (e.g., extract brand → map to numeric code).
• Keep prompts concise: For LLMs, simpler prompts tend to yield more consistent results.