Evaluations are a key technique for improving your application — whether you’re working on a single prompt or a complex agent. Iterating on evaluators has often involved a lot of guesswork.

With Align Evals you get:

• A playground-like interface to iterate on your evaluator prompt and see the evaluator’s “alignment score”

• Side-by-side comparison of human-graded data and LLM-generated scores, with sorting to identify “unaligned” cases

• A saved baseline alignment score in order to compare your latest changes to the previous version of your prompt

Read more about it and learn how to get started in our blog post.