Small decisions. Interesting possibilities.Submit contentSubmit
Jimothy Jev Distillation preview

Jimothy Jev Distillation

A distillation CLI uses hosted Jev teacher judgments to label data for a smaller local classifier.

Added to Jevfast

How it uses Jev

Jimothy can use Jev-compatible typed examples as teacher labels for task-specific classifier training. For the included support-routing task, each input state is a support message and the teacher request carries one Choice question named route; Jimothy validates the returned label and caches completed teacher responses for reuse.

What Jev decides

Support request routing (illustrative request shape)Example answers · not a recorded Jev response · Source ↗
Question 1 · route
YOUR APP
INSTRUCTION

Which team should handle this support request? Choose the primary issue.

STATE

A support-ticket text, such as the included example about a duplicate subscription charge. This is an example task/input, not a captured Jev call.

JEV · CHOICE
  1. billing
  2. shipping
  3. technical

App workflow

  1. Define the task and inputs

    Provide a typed task definition and unlabeled input rows. Jimothy validates the question contract and each input before making paid teacher requests.

  2. Label unique inputs with Jev

    For each uncached unique state, send the chosen task question to the configured Jev-compatible teacher endpoint and validate the typed answer.

  3. Cache and train

    Persist validated teacher responses so a rerun can reuse completed labels, then use the labeled examples in the task-specific classifier training flow.

Why it is interesting

It uses a structured decision model to produce labels for a smaller local classifier, separating the teacher's typed classification contract from the runtime classifier that can run locally.