Creating and Running Jobs

The core of Row Sherpa is the Job. A job represents a single batch process where an AI model runs across every row of your uploaded CSV file.

Step 1: Upload Data

Start by clicking New Job from your dashboard.

Upload CSV: Drag and drop your file or select it from your computer.

> Note: If you're looking to process an Excel file or Google Sheet, simply save it as a CSV first before uploading it to a Row Sherpa job.

Step 2: Configure Engine

This is where you tell the AI what to do with your data.

1. Select Model

Select the LLM that will process your data. If you need a web search to be made when analyzing the data, select Sonar (by Perplexity).

2. Define Your Prompt

You can define your prompt in one of the following ways:

  • By choosing one of the pre-defined templates,
  • By writing a prompt form scratch (choosing 'Custom (start from scratch)'),
  • By using a previously saved prompt.

Note: your prompt always need to include the {csv_data}  placeholder to instruct the AI model what to do with your CSV data. At execution, {csv_data} will be replaced by the content of the row which is being iterated upon from your CSV.

See Managing Prompts for details.

3. Define Outputs

If you don't specify anything in the Structured Output section, the model will simply follow your instructions and return Free Text. While this can work well for some use cases, there are many where returning a Structured Output will help you a lot.

To ask the model to return a Structured Output, simply define specific fields (e.g., "Category", "Sentiment Score") in the Structured Output section as well as the type of variable they should be (string, number, boolean or array). You can also click on Advanced mode to input a JSON file that defines these fields and their variable type.

Don't hesitate to check the templates to see how they use Structured Outputs.

  1. Live Preview

Before running your job on your entire CSV, it's a good idea to run a Live Preview. Live Preview performs what you instructed Row Sherpa to do on your data but only on the first row of your CSV. This enables you to validate that your prompt perfoms as it should or that the output is structured the way you want.

Step 3: Review and Launch

Before running, you will see a summary:

  • Dataset, including the number of rows to process,
  • Model chosen,
  • Prompt.

Click Launch Job to begin processing. You can close the tab and come back later; the job runs on our secure cloud servers.

Monitoring and Results

You can track progress on the Dashboard page that you will be automatically redirected to as you launch your job. Once the status shows as Completed, a Download button will appear, allowing you to get your original CSV enriched with the new AI-generated columns.

Related Articles