{"slug": "open-sourced-jev-architecture-last-year-with-model-paper-and-dataset", "title": "Open-sourced jev architecture last year with model,paper and dataset", "summary": "An open-source developer published an arXiv paper (2503.23303), a Hugging Face model (DeepMostInnovations/sales-conversion-model-reinf-learning), and a training dataset (DeepMostInnovations/saas-sales-conversations) in March 2025 describing a non-autoregressive architecture that predicts probabilities against a JSON schema, and now says a frontier lab has proposed the same idea as a breakthrough without releasing technical papers, open weights, or an open dataset. The developer says their model uses PPO over sequence embeddings to output turn-by-turn conversion trajectories with probabilities from 0.0 to 1.0, while the frontier lab's approach uses parallel sampling trained via RLCD to output confidence distributions and schema choices. A second paper (2510.01237) published in September 2025 is described as matching the same idea.", "body_md": "Everyone now talks about the architecture that's not auto regressive and does lightning fast probability prediction with a json schema. I worked on this literally one year back in March 2025, published an arxiv paper, pushed the model to huggingface along with the pypi package and training dataset. And then one year later, a frontier lab came, proposing the same idea like literal breakthrough without technical papers, open weights and no open dataset. For anyones information the main guiding model is RL not embedding model or LLM\n\nPaper: https://arxiv.org/abs/2503.23303\n\nModel: https://huggingface.co/DeepMostInnovations/sales-conversion-model-reinf-learning\n\nDataset: https://huggingface.co/datasets/DeepMostInnovations/saas-sales-conversations\n\nAlso the second work published in September 2025 was exactly the same one jev proposed now\n\nPaper: https://arxiv.org/abs/2510.01237\n\nMy model uses PPO over sequence embeddings to output turn-by-turn conversion trajectories (probabilities from 0.0 to 1.0).\n\nJev uses parallel sampling (trained via RLCD) to output confidence distributions and schema choices.\n\nIt's incredibly frustrating that the thing that you made with months of hard work, sweat and sleepless night is architecturally similar with the vertical use case and don't get the support you deserve because frontier lab build something horizontal. The open-source story in general\n\nComments URL: [https://news.ycombinator.com/item?id=49736660](https://news.ycombinator.com/item?id=49736660)\n\nPoints: 1\n\n# Comments: 1", "url": "https://wpnews.pro/news/open-sourced-jev-architecture-last-year-with-model-paper-and-dataset", "canonical_source": "https://news.ycombinator.com/item?id=49736660", "published_at": "2026-09-17 05:13:00+00:00", "updated_at": "2026-09-17 05:25:12.962383+00:00", "lang": "en", "topics": ["ai-research", "machine-learning", "ai-startups", "ai-tools"], "entities": ["arXiv", "Hugging Face", "DeepMostInnovations", "PPO", "RLCD"], "alternates": {"html": "https://wpnews.pro/news/open-sourced-jev-architecture-last-year-with-model-paper-and-dataset", "markdown": "https://wpnews.pro/news/open-sourced-jev-architecture-last-year-with-model-paper-and-dataset.md", "text": "https://wpnews.pro/news/open-sourced-jev-architecture-last-year-with-model-paper-and-dataset.txt", "jsonld": "https://wpnews.pro/news/open-sourced-jev-architecture-last-year-with-model-paper-and-dataset.jsonld"}}