Ingest
Smart intake and classification for immigration files.
Client documents arrive as phone photographs, sealed scans and files called doc(3).pdf. The classifier reads all of it, names it, categorises it, and tells you what is missing before a paralegal opens a single PDF.
- 31document classes recognised
- 8scripts supported for OCR
- 4compliance checks per document
Intake queue — Ramírez, EB-2 NIW
45 files received · 10 shown · 312 MB · 6 formats
Diploma — PhD Materials Science, TU Delft
Academic credentialIMG_4821.pdf
Official transcript — TU Delft, sealed
Academic credentialscan0044.pdf
Birth certificate — Bogotá, Colombia
Identity documentSpanish originaldoc(3).pdf
- No translator certificate — 8 CFR 103.2(b)(3)
Expert declaration — Dr. E. Moreau
Expert testimonyletter_final_v2.docx
Expert declaration — Prof. R. Sato
Expert testimonyJapanese originalWhatsApp Image 2026.jpeg
- No translator certificate — 8 CFR 103.2(b)(3)
- Unsigned or undated
IRS Form 1040 — TY2024
Financial recordtaxes 2024.pdf
- PII detected and redacted
Citation report — Scopus, cap. 12 Aug 2026
Citation evidencescholar_export.csv
NSF award notice — CBET-2418803
Funding recordNSF_award.pdf
Passport biographic + entry stamps
Identity documentpassport scans.pdf
- PII detected and redacted
Employment verification — Corvine Systems
Employment recordemployer ltr.pdf
- Unsigned or undated
31 document classes recognised
3 held from the binder
2 redacted
Four checks
What runs on every document, every time.
These are the failures that produce RFEs citing evidence you already filed. All four are mechanical, which is why they should never depend on a person having a good day.
OCR and classification
Multi-pass recognition handles international documents, handwriting on older civil records, and the stamped and embossed surfaces on passports and visas that defeat single-pass OCR. Each document is then classified into one of 31 categories, renamed to something a human can read, and placed in the intake queue with its source filename preserved for audit.
Identity
- Passport biographic pages
- Entry and exit stamps
- National identity cards
- Birth certificates
- Marriage and divorce records
- I-94 arrival records
- Prior I-797 approval notices
Academic
- Diplomas and degree certificates
- Official transcripts
- Credential evaluations
- Enrolment verification
- Thesis and dissertation records
Employment & financial
- Employer support letters
- Employment verification letters
- Pay stubs and W-2s
- Corporate tax filings
- LCA and ETA-9089 certifications
- Offer letters and contracts
Evidentiary
- Expert declarations and affidavits
- Peer review and editorial records
- Citation and impact reports
- Publication lists
- Media coverage
- Award certificates
- Grant and funding notices
- Patent and licensing records
Translation compliance
Flags any non-English document lacking a certified English translation and a translator's competence certificate, as 8 CFR 103.2(b)(3) requires. Held out of the binder until resolved.
Signature and execution audit
Verifies the presence of a signature and a date on employer letters, expert affidavits and declarations, and flags a signatory that does not match the letter body.
Automated PII redaction
Removes Social Security numbers, bank routing and account numbers, and third-party identifiers including children's dates of birth — from the content stream, not with a black box.
The translation rule that produces nine percent of RFEs.
Any document in a foreign language must be accompanied by a full English translation and a certification from the translator stating that the translation is complete and accurate and that they are competent to translate. A translation without the certificate does not satisfy the rule, and it is the second half that gets forgotten.
What the classifier checks
- Document language detected as non-English
- A translation document is present and paired
- A translator certificate accompanies the translation
- The certificate names the translator and asserts competence
- The certificate is signed and dated
Any check that fails holds the document out of the compiled binder and raises a task on the matter.
The pipeline runs multi-pass OCR with deskew, denoise and contrast normalisation before recognition, then scores its own confidence per page. Any page below the confidence threshold is surfaced for human review rather than silently accepted — the failure mode we design against is confident wrong text, not missing text.
Start free trial
Stop losing flat-fee profit to manual PDF formatting.
Upload one messy client folder. Get back an audit-ready, Bates-stamped exhibit packet with a two-tier index and a matched cover letter — in about three minutes.
14-day free production trial · no card · real matters · no watermark
Software for licensed attorneys. Not legal advice.