Overview

Create, manage and analyze datasets for AI model development and evaluation

What it is

Datasets are the core data layer for evaluation and experimentation. Each dataset is a table: columns (e.g. “user query”, “expected answer”, “score”), rows (one row = one example), and cells (the value in each column for each row). Datasets are the single source of truth that prompts, evals, experiments, and optimizations run on.

Purpose

  • Store and manage test/eval data in one place.
  • Run prompts and evals over the same structured data.
  • Compare model or prompt performance across experiments.
  • Support building datasets from product usage (e.g. from observed traces) as well as from uploads, API, or synthetic generation.

Getting Started with Datasets

Create New Dataset

Create datasets using SDK integration, file upload, or synthetic data generation

Add Rows to Dataset

Learn how to add individual records or bulk import data rows

Add Columns to Dataset

Extend your dataset structure with additional data fields

Run Prompts

Test and execute prompts against your dataset entries

Experimentations

Design and conduct controlled experiments to compare approaches

Annotate Dataset

Add metadata and annotations to enrich your dataset
Was this page helpful?

Questions & Discussion