OpenDecision
a 400M zero-shot model makes local decisions, plays Doom
Details
- External ID
- 49787404
- Source
- HN
- Company
- —
- Product
- OpenDecision
- Website domain
- github.io
- Launched
- Sept. 21, 2026
- Cohort
- —
- Upvotes
- 6
- Upvotes percentile
- 0.32854864433811803
- Tags
- —
- Fetched at
- Sept. 25, 2026, 5:02 p.m.
- Updated at
- Sept. 25, 2026, 5:02 p.m.
Description
Jev's launch inspired me to work on this project. I had used fine tuned bert models, zero-shot models to achieve jev like functionality in healthcare insurance fraud audits in the last 2-3 years with good results.Jev's launch showed that there is still market and demand for these type of systems. So I wanted to build a FOSS version which would push the research and tinkering (R&T) in this direction.Do not use this where decisions can be costly. There's an accuracy/intelligence ceiling with these models no matter what the api seller tells you.Use this where you can tell the affected party - "oh yeah, sometimes it can mess up" and you both can have a laugh on the expense of machines.On Jev and Typesafe - I think Jev is great, especially the speed, the documentation and the product design itself.Happy to answer any questions.Thanks, D
Enrichment
- Theme
- decision model runtimes and tools
- Vertical
- Horizontal
- Function
- Model & infra
- Audience
- Developer
- AI stance
- AI-native
- Project type
- Hobby / open-source project
- Normalized one-liner
- small zero-shot decision model for local gameplay and tasks
- Manually corrected
- False
Could you build this?
Partial The Python API wrappers and schema primitives are easy to code, but training, calibrating, and distilling a specialized 400M zero-shot NLI model for low-latency decision making requires ML engineering expertise.
What it would actually take: The core engine requires fine-tuning a small transformer cross-encoder (such as DeBERTa-v3) on structured natural language inference, question-answering, and state-evaluation datasets using PyTorch. The inference pipeline requires probability calibration (temperature scaling) to ensure consistent thresholding for choice/relation primitives, along with ONNX/TensorRT quantization to hit sub-100ms inference times on local consumer CPUs.
Discussion
No comments on this launch.
Competitors
Other products that read as similar to this one — 275 launches clear the similarity bar, closest 8 shown.
Attention rank: #232 of 276 (itself plus its competitors, highest first — normalized so YC and Product Hunt are compared fairly).
Launched 319 days after the earliest competitor.
- open-alternative-jev · github · 2026-09-18 · 50 upvotes · similarity 0.56
- open-jev-typed-decision-engine · github · 2026-09-19 · 43 upvotes · similarity 0.56
- JevBench, a reproducible benchmark for typed decision models · hn · 2026-09-22 · 149 upvotes · similarity 0.55
- awesome-jev · github · 2026-09-18 · 106 upvotes · similarity 0.52
- Jev by Typesafe · ph · 2026-09-18 · 10 upvotes · similarity 0.52
- OpenJev · github · 2026-09-20 · 27 upvotes · similarity 0.52
- Jev AI — Jev Model Online · ph · 2026-09-22 · 1 upvotes · similarity 0.50
- openJev-verdict-2.0 · github · 2026-09-19 · 276 upvotes · similarity 0.50
Other launches for this product
- No other launches for this product.
Same idea, different domain
Nobody's really built a model & infra tool for Fintech yet.