/ jev lab

Watch Jev make a call.

Jev is TypeSafe's decision model. Instead of generating text, it answers typed questions about whatever you give it: pick one of these labels, rate this on a rubric, yes or no. Every answer comes back with calibrated probabilities. This is exactly how projects on the wall get their category and jury score.

Answers appear here

Four questions in two parallel requests: a category choice, two yes/no checks and a scope score.

The code behind it

Two small systemOne calls in parallel through Netlify AI Gateway. No API key in the code, no prompt parsing, no JSON repair.

import { choice, noul, score, TypeSafeClient } from "@typesafe-ai/sdk";

const typesafe = new TypeSafeClient(); // keys injected by Netlify AI Gateway
const state = { project_name, pitch };

// Two small requests in parallel: one picks the category, one asks the rest.
const [sorted, checks] = await Promise.all([
  typesafe.systemOne({
    model: "jev-latest",
    state,
    questions: {
      category: choice("Which area fits this project best?", {
        moderation: "Trust and safety",
        lead_scoring: "Sales and growth",
        health: "Health and wellbeing",
        playful: "Games and fun",
        // …
        other: null,
      }),
    },
  }),
  typesafe.systemOne({
    model: "jev-latest",
    state,
    questions: {
      decisionPowered: noul("Does this app use AI to make a decision?"),
      needsHuman: noul("Would a human need to review its decisions?"),
      scope: score("How big is this idea for a four-hour hack night?", [
        "Tiny, a toy demo", "Focused, one workflow", "Ambitious, many parts", "Moonshot, weeks of work",
      ]),
    },
  }),
]);

sorted.answers.category.choice;        // "health"
sorted.answers.category.probabilities; // { health: 0.81, playful: 0.09, … }
checks.answers.decisionPowered.noul;   // 0.94