What it is
Document decisions with candidate answers and probabilities.
What it does: Runs multimodal document classification and rule checks as closed questions over one shared image context.
Why it belongs in the showcase: Ships weights, a bench, and an explicit Jev-inspired decision interface for documents.
Caveats: Not hosted Jev; author bench; large download.
Evidence
Public Apache-2.0 repo (2★ at intake) with bilingual README, Hugging Face model SoMarkAI/Docision-1.0 and dataset SoMarkAI/doc-decision-bench (5,234 document records claimed on the badge page). README shows install + demo_docision.py inference and author benchmark SVGs comparing Qwen3-VL-4B, Docision, and JEV-27B-VL accuracies. Labels the work as Jev-inspired multimodal decisions, not hosted TypeSafe Jev.
Caveats
- Community / compatible boundary: local Docision weights, not TypeSafe-hosted Jev
- Benchmark numbers are author-published SVGs; verify the HF dataset yourself
- Inference needs downloading multi-GB model weights
Unofficial case page. Stars / list position observed for heat; do not treat heat as a quality SLA. Review the upstream README before you depend on it.