Post
CommunityJev Cookbooks
Document Review With Unsiloed and TypeSafe Jev
A walkthrough that pairs Unsiloed document extraction with Jev: extracted medication rows and their citations become Jev's state, one Noul question per row flags implausible or obscured doses, and the results decide which fields need review.
- Format
- Post
- Reading time
- 3 min
- Published by
- Community
- Author
- Unsiloed AI
- Last verified
- Oct 2026
- Languages
- JavaScript
- Platforms
- API
Highlights
- Unsiloed extracts structured fields and locates them on the page; Jev then judges each extracted value against focused questions.
- Each field carries extraction_score and grounding_score, plus a citation with page and bounding box on the original scan.
- All six medication rows share one request; each question is evaluated independently, with row IDs matching answers back to rows.
- A Noul of 0.04 is a 4% probability of yes — a strong no — not a statement that Jev is 4% confident.
- Requests post to https://api.typesafe.ai/v1/systemone with model jev-1.13.0 and a TypeSafe API key in the Authorization header.
Quickstart
ts
const request = {
model: "jev-1.13.0",
state: {
context: "Check for obvious quantity, unit or frequency errors. Dose means amount taken each time, not supply quantity. Do not assess patient-specific safety or propose corrections.",
medications: [{ id: "row-5", medicine: "Cetirizine 10 mg tablet", dose: "1 tablet", frequency: "Once daily as needed" }]
},
questions: {
"row-5": { type: "noul", instructions: { question: "Is the extracted dose and frequency plausible on its face for this medicine and strength in the stated adult outpatient context, without an obvious quantity or frequency transcription error?", row_id: "row-5" } }
}
};Watch out
Grounded in the truncated Unsiloed blog post read on 2026-10-15; the excerpt stops at the API description, so the routing code, costs and interactive prescription example are not covered here. All figures come from a single six-row medication example, and the post notes Jev judges the extracted text without seeing the image or Unsiloed's scores.
Questions about this entry
Short answers drawn from the linked project or page.
It checks documents for OCR failures and general mistakes by combining Unsiloed and Jev: Unsiloed extracts structured fields and finds them on the page, and Jev judges the extracted information against focused questions so the pipeline can decide which fields need review.
Extraction plus source evidence. Each field comes back with two confidence scores — extraction_score for the value and grounding_score for its location — and a citation with page and bounding box. Enable citations with enable_citations=true and wait for the job to finish before building Jev's state.
A plausibility judgment on the extracted text. Unsiloed preserves what is printed and scores ambiguous readings, while Jev answers whether the dose and frequency are plausible on their face for the medicine, catching clearly printed but wrong values that OCR handles perfectly.
Noul, which returns a yes-probability between 0 and 1. The post notes a Noul of 0.04 means a 4% probability of yes to the question — a strong no — and that it is not a confidence value. Values near the middle deserve caution.
Each medication row gets an ID such as row-5, used both as the question ID and as row_id inside the instructions, so code can match Jev's answer to the same extracted row. The post notes question IDs themselves are not sent to the model.
The example sends all six rows in a single request, each with its own question, and the post says each question is evaluated independently. It does not document a limit on how many questions one request can carry.
Post the request as JSON to https://api.typesafe.ai/v1/systemone, authenticated with your TypeSafe API key in the Authorization: Bearer header. The example pins model jev-1.13.0 and builds state from the extracted records plus context.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
X: Finding a signature on a scanned form
X: A data analyst agent with Jev as the review layer
X: A 12-page PDF on building a Jev harness
X: A review of Jev for AI workflows
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.