Jev + Nutrient 文档抽取：把非结构化文档变成路由决策
作者认为 Jev 的威力取决于答案形状是否已知，并用 Nutrient 抽取文档结构后让 Jev 判断文档类型、路由去向、是否继续和是否需要人工复核。
The more I think about Jev, the more I believe its power depends on one thing: knowing the shape of the answer. If the answer is yes/no, one of a defined set of choices, or a score, Jev can give software a fast, inexpensive, probabilistic answer it can act on. If you don’t know the answer’s shape, you haven’t eliminated much. You still need an LLM to discover the relevant information and turn an open-ended problem into something constrained. Documents are where this becomes much more interesting. Nutrient’s data extraction can preprocess a document into structured, grounded content. Jev can then make decisions over that content: • What type of document is this? • Where should it be routed? • Does it meet the conditions to continue? • Is confidence high enough to automate, or should a human review it? That pairing turns unstructured documents into deterministic classifications, routing rules, and workflow decisions. That’s what we’re exploring with docsignals. Documents in. Signals out.