Start Here: Learn Jev Step by Step

Stage 1 — Understand

Beginner · 15 minutes · No prerequisites. Learn the shape of a decision model before installing anything. Read What is Jev?, How Jev works, then Choice, Score, and Noul.

You should be able to write a question with a bounded answer space. Try this checkpoint: a department is a Choice, an ordered urgency rubric is a Score, and “is a refund explicitly requested?” is a Noul.

Stage 2 — Build

Beginner · 25 minutes · An API key and basic Python or JavaScript. Make your first API call. Pick the Python SDK or JavaScript SDK, then improve the input with the state guide.

Checkpoint: print the selected team, full probability distribution, model ID, and token usage. Do not connect a live refund action yet. The first result should be easy to inspect without changing a customer account.

Stage 3 — Control

Intermediate · 30 minutes · A working request. Read confidence and multiple questions. Then add confidence routing and explore composite scoring.

Checkpoint: an ambiguous ticket must go to a review queue. A high score is not permission to move money. Keep authentication, policy, and confirmation in code.

Stage 4 — Ship

Intermediate · 30 minutes · A small labeled evaluation set. Read the limitations, error guide, rate-limit guide, and version-pinning advice.

Checkpoint: a timeout, invalid answer, or unavailable provider does not execute an action. Record the actual model version and compare a new version against the same labeled examples before switching.

Stage 5 — Real projects

Beginner to advanced · 30–60 minutes each. Start with support triage, then try a model router, an agent risk gate, RAG filtering, or LLM guardrails.

Each tutorial has a downloadable Python program, a no-network fixture mode, and a live mode requiring your own key. Compare the fixture output with a live response; the fixture illustrates the data contract and is not a measured model result.