JevMade hello@JevMade.com
← Back to guides

JevMade field notes / Python tutorial

Let's look at Jev, the AI that decides without ever writing a sentence

Jean-François Lépine explains the three primitives, measures question batching, and builds a compact semantic grep that combines line ranking with a separate answer-exists check.

Original by Jean-François LépineGetting startedIndependent blogOriginal published Source reviewed

Before you dive in

What you’ll find in the original

  1. A Choice always selects something because its probabilities sum to one; pair it with a Noul when “none of these” must be possible.
  2. Describe Score levels as concrete situations, then inspect the distribution and confidence because the same average score can hide different uncertainty.
  3. The semantic-grep example uses Choice to rank numbered lines and Noul to decide whether the document answers at all, keeping thresholds in code.
Worth knowing

The author’s outputs and 1.53× batching result come from his own run and are not independently reproduced; the article also notes a 255-option limit and no text generation.

More guides →See the experiments →