OmniJev/awesome-jev-gallery

What it solves

This project compiles the ecosystem around Jev and System One models—AI systems that take typed questions (e.g., choice, score, or boolean) and return calibrated probabilities for each answer option in a single pass. It saves researchers and developers from hunting across blogs, GitHub, and Hugging Face by organizing the key papers, open-source implementations, evaluations, and discussions into one searchable gallery.

How it works

The README organizes 171 entries into thematic sections. The "System One & Jev" section links to TypeSafe's official launch posts, technical docs, and coding adapters. The "Open Source" section lists 49 projects that reproduce or extend the Jev architecture, from small decision models (like OneJev, PlayJev, or sokudan) to inference runtimes and browser demos. Other sections cover tools built with Jev, benchmarks and leaderboards (35 entries), commentary and analysis, and the lineage of ideas that led to Jev. The gallery is actively maintained, with recent additions of a benchmark section and a papers-only list.

Who it’s for

Researchers, machine learning engineers, and hobbyists curious about typed probabilistic decision models. It is especially useful for anyone wanting to understand Jev's design, compare open-source replicas, or apply Jev-style decisions in their own projects.

Highlights

• 171 total entries, including 27 papers, 49 open-source models, 35 benchmarks, and 127 projects with code.
• Includes official TypeSafe materials like the launch post, the typed primitive docs, and the System One Adapter.
• Open-source implementations range from tiny 0.6B models to 27B multimodal systems, many with local runtime demos.
• Covers applications like playing browser games, audio classification, and biomedical evidence decisions.
• Actively updated with news and daily paper additions.