shadcn.io is not affiliated with official shadcn/ui
PDF Processing Dashboard Block for Solid
A queue of four PDFs beside a drawn page with numbered field boxes and a table of extracted fields with confidence words, where Low fields can be corrected and Accept field confirms one and updates the accuracy.
Installation
Overview
Check what the extractor was unsure of before it reaches your records, with this document extraction panel for Solid with Ark UI. A hairline frame with a plus mark on each corner puts a queue of four Scanlark PDFs at the start, each with its file name, page count, stage (Extraction, Validation, Complete or Failed) and accuracy. For the file you choose, a drawn page shows numbered boxes where each field was read, beside a table of the fields with their value and a confidence word: High, Medium or Low. A Low field's value is an Input you can correct, choosing a field outlines its box on the page, and the Shine Button Accept field marks it Confirmed. Accuracy is the mean of the confidences with confirmed fields at 100, and a file with every field confirmed reads Complete. It's built from Input, Native Select, Badge, Shine Button and Guides, with a drawn page. For document automation and data entry review.
FAQ
How do I fill it with my documents?
Edit DOCS: each has a file name, pages, a stage and its fields. A field has a name, the extracted value, a score from 0 to 100 and a box on the page given as start, top, width and height in percent. The confidence words, the accuracy and the stage shown in the queue are worked out from the scores and from what you have confirmed.
What do High, Medium and Low mean?
A score of 95 or more reads High, 85 to 94 Medium and below that Low. Only Low fields show an input, because those are the ones worth correcting. Once you accept a field it reads Confirmed and counts as 100 in the accuracy, and the selection moves to the open field with the lowest score.
What does Accept field do?
It confirms the chosen field with whatever its value now says, refuses an empty value and tells you in the status line, which also shows the accuracy before and after. A confirmed field, or the failed scan, answers with a short note instead of changing anything.
What stays local, and how do I wire it for real?
Everything: nothing is uploaded, no file is read and the drawn page is a picture, not your PDF. To make it real, render the page image from your extraction service, place each box from the coordinates it returns, and send confirmed values to your own API from accept.