Skip to main content
Use this tutorial when you want flexible LLM classification without giving up typed downstream behavior.

​
Best fit

  • support intent classification
  • document routing
  • policy or moderation tagging
  • ecommerce or content categorization
  • direct API or batch, depending on volume
  • structured outputs for stable labels
  • evals when the labels matter to the business

​
Workflow

  1. define a tight label set
  2. model the output as a JSON schema
  3. run a few examples synchronously
  4. add an eval before using the classifier for a business-critical workflow
  5. move to batch or training if the workload grows