Let Jev paint a sketch region by region

01 · Role in the system

What Jev does here

Each selected region is measured and turned into a typed state; code builds a candidate field of paint materials and treatments and asks Jev a single choice question. Jev returns a full probability distribution - the choice is the argmax and the confidence is a function of sample count and top probability - and eSketcher normalises, ranks and bands that certainty before painting. The chosen material leaves the stream at the bottom of the screen, flies across the desk and paints the region, and the painted colours feed into the next decision, so the picture emerges from a chain of calibrated judgments. A FastAPI backend isolates every Jev call behind a provider interface with a real provider for api.typesafe.ai and an offline mock, and the repo ships architecture diagrams, a decision log, tests including a real-provider test, a fifty-sketch demo film and a live gallery.

02 · Control boundary

Where Jev sits

Jev as the per-region material judge in a generative art pipeline: code measures and enumerates candidates, one choice question decides the material, probabilities are read back with certainty bands, and the painted result joins the state for the next decision.

Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.

03 · Known limits

What this evidence does not prove

• Art outcomes are aesthetic judgments; the docs describe internals but no external quality benchmark exists.
• Backend calls need a TypeSafe API key; an offline mock provider stands in for development.
• One decision per region; no long-horizon composition planning.

04 · Attribution

Public sources

This is a Community record: the project was published by a third-party community author.

• Faizullah9181 ↗Community · github · public · checked 2026-09-28