jev-fill-pdf
Fills Japanese government PDF forms by asking Jev which profile item each unlabeled field is; personal values stay in the browser.
Overview
jev-fill-pdf fills Japanese PDF application forms in one click. These forms often have fields named only by number, scanned page images and 'no changes allowed' encryption, so there is no label text for software to read. The tool decrypts in the browser, runs OCR on image-only forms, and sends Jev only the noisy label fragments around each field plus the names of profile items (name, furigana, postcode and so on). One request holds a Choice per field, with profile keys plus 'none' as options, so all fields of a form are judged in parallel.
Values never leave the browser. Fields at 0.8 or above are filled, 0.5 to 0.8 are filled but flagged for confirmation, and lower ones are left blank for the user to pick. The label list is shown before sending, and confirmed field mappings are stored per form hash, with presets for popular forms that need no OCR or API call. Jev is reached through Vercel AI Gateway with zero-retention settings. The client is a single HTML file with a small relay.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which profile item does this form field correspond to? | — |
How it works
- Drop a PDF; decrypt and OCR in the browser
- Show label fragments that will be sent
- One Jev request with a Choice per field over profile keys plus none
- Fill at 0.8+, flag 0.5 to 0.8, leave blank below; remember mappings
Strengths
- Personal values stay in the browser
- Confidence bands prevent silent wrong fills
- Presets skip the API entirely for known forms
Caveats
- Focused on Japanese forms
- OCR quality affects results
- Version 0.1.1
Getting started
Use the live tool at https://ilove-ai.net/fill.At a glance
- Type
- Service
- Status
- beta
- Language
- JavaScript
- Runtime
- Browser app (pdf-lib, pdf.js, tesseract.js) with Vercel relay
- Has code
- yes
Use cases and integrations
How curated lists describe it
Fill Japanese PDF forms (申請書・届出書) in one click with Jev by TypeSafe AI. Labels go to Jev, your values never leave the browser. OCR for image-only forms, works with 国税庁 forms. Built on pdf-lib / pdf.js / tesseract.js via Vercel AI Gateway.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev decide?
Which profile item each unlabeled form field should be filled with.
Are my personal details sent?
No, only label fragments and profile key names are sent.
What if Jev is unsure?
Fields below 0.5 are left blank for you to choose; 0.5 to 0.8 are flagged to confirm.
Related projects
Docker service that classifies the Paperless-ngx inbox with Jev (or local Clef via Ollama), setting confident fields and queueing the rest.
ProjectbetaDesktop dictation and meeting-notes app with optional Jev fast judgments for citation checks, history search and live meeting cues.
ModelproductionPythonmacOS menu-bar app that splits copied text into fields and offers Tab-to-paste hints in matching input fields, ranked by recency.
ProjectbetaSwiftWindows office workbench with search, OCR and batch renaming, where Jev optionally helps pick extracted contract fields.
ProjectbetaRustStreaming PII redaction library for TypeScript with 155 regex detectors, optional NER, and opt-in Jev confirmation before redacting names.
ProjectbetaTypeScriptLocal-first desktop and Android attention workbench that extracts deadlines from WeChat and campus notices; Jev optional.
Projectbeta