Skip to main content
Evaluations are not only for testing prompts. You can also use them as batch jobs where each row is an input and each column is an AI-powered computation.

​
Common use cases

  • Data labeling: Run a prompt over production examples to create labeled datasets
  • Research: Process a list of companies, people, or documents
  • Content generation: Generate summaries, replies, emails, or descriptions in bulk
  • Data enrichment: Add company, location, category, or other attributes to a list

​
Create the dataset

Upload a CSV, create rows manually, or build a dataset from request history. Each row should contain the fields your prompt needs as input variables. Learn more in Datasets.

​
Add prompt columns

Create an evaluation and add one or more Prompt Template columns. Map dataset columns to the prompt input variables. You can chain columns together when later prompts depend on earlier outputs.

​
Run and export

Run the full batch, review the results, and export the completed dataset when you are done. You do not need permanent evaluation infrastructure for this workflow. Create a dataset, add prompt columns, run the batch, export the results, and move on.

​
Next steps