When building a dataset iteratively for an LLM app, having a defined schema for testing across examples lets you avoid broken code and keep your data clean and consistent.

With LangSmith, you can now define and flexibly manage dataset schemas. LangSmith validates your examples against the defined schema, raising error messages when detected. Auto-complete in the UI also helps you conform to the schema.

Updating your schema is also simplified in LangSmith — you can go through a queue of existing datapoints that no longer fit the desired schema, so that you can reformat them to the new schema and fix them directly in the UI.

• Read more about dataset schemas in our blog post.

• Check out the docs for more examples.