Score Schemas

Literal AI offers an easy way to manage scores: Score Schemas. With Score Schemas, admin users can control and expose to all users the various types of evaluations allowed in their application. Score Schemas come in two flavors:
  • Categorical — Schemas let you create a set of categories, each tied to a numeric value.
  • Continuous — Schemas with a [min; max] range which users score in.
For Categorical Schemas, each category has a description which is used to guide the scorer (be it Human or AI) in selecting the most relevant category. Here’s how to create a Score Schema:
Create a Score Schema

Create Score Schema

Feel free to browse your steps from the Logs tab and simply add a “Human Review” directly on your Step and Generation entities:
Score a Generation

Score a Generation

To go through a wide range of logs, leverage annotation queues which make it easier to streamline the review process.