Decisions you can defend
Every finding is anchored to a page, a region, a producer version and a confidence score. Reviewers see why, not just a verdict — and can defend the decision months later.
Local-first · Now in early access
Catch document defects before they reach the closing table.
Docutise reviews every page of a loan file for signatures, dates, notary marks, barcodes, PDF container evidence and visual defects — backed by evidence, tuned for near-zero false pass, and processed entirely on your own hardware.
No cloud model calls
Evidence for every finding
Full audit trail
Wet signature · 0.94
Date
Notary seal
Needs review
72
Signatures2 found
Notary seal0.88
Date evidence1 stale
Built for the realities of lending operations
Prefunding QC
Post-closing QC
Trailing documents
API-driven audits
The problem
A single missed signature can unwind a funded loan.
Document defects surface late, cost days at closing, and expose lenders to repurchase risk. Manual stare-and-compare doesn't scale, and generic file scanners give a silent pass with nothing to defend it. Docutise treats every page as evidence to be examined — and says so clearly when it isn't sure.
Catch defects before they cost you
The system is tuned to fail closed. When required evidence is missing or uncertain, the file routes to a person instead of quietly stamping accepted.
Your data never leaves
All analysis runs on local hardware. No cloud model calls, no alternate inference path. Borrower NPI stays inside your walls — by architecture, not by policy.
Evidence lanes
Every page, checked across eight lanes of evidence.
Docutise inspects each page independently and cross-references the results. Each lane produces its own defensible evidence with coordinates you can point to on the document.
Page quality
Blur, low resolution, contrast, skew, crop, exposure, noise and compression — flagged before they hide a defect.
PDF structure
Page count, encrypted or corrupt containers, AcroForm indicators, unsafe features, annotations and metadata.
OCR & layout
Extractable text, page-level OCR blocks, layout regions and reusable evidence coordinates for every finding.
Barcodes
Detection, decoding and masking policy — with required-barcode exceptions when one must be present and readable.
Signatures
Wet signatures, initials and electronic signature blocks — kept distinct from separate PDF digital-signature evidence.
Dates
Missing, invalid, future, stale or unparseable dates — evaluated against the active audit profile's policy.
Notary indicators
Notary language, seal and stamp indicators, commission language, and clearly-flagged notary uncertainty.
Alterations
Strikeouts, whiteout, void or cancel marks, obscured text and suspicious blank regions that hint at tampering.
How it works
From upload to an audit you can defend.
A predictable pipeline runs on every document. Deterministic checks first, local models where certified, and a person on anything uncertain.
Intake
Validate MIME, magic bytes, size, page count and encryption. Malware-scan locally before anything is stored.
Render
Rasterize each page at 300 DPI, normalize and thumbnail — the visual substrate for every downstream check.
Examine
Run quality, PDF, OCR/layout, barcode, signature, date, notary and alteration lanes across every page.
Score
Apply the audit profile's rules, cap scores on blocking findings, and assemble evidence-backed exceptions.
Review
Reviewers triage exceptions; an admin makes the final call. Export a durable, defensible audit package.
Review console
A review console for QC professionals — not a generic file scanner.
Docutise speaks the language of lending operations: loan file, document, exception, condition, cure, re-upload, reviewer, admin final review.
- Evidence panels & page overlaysClick a finding, land on the exact region of the exact page that produced it.
- Human-in-the-loop by designReviewers triage and recommend; only an admin can accept, reject, or require a re-upload.
- Cure & re-upload workflowRoute defective documents back for a replacement and track them to resolution.
Uniform Residential Loan Application
URLA · 9 pages
Accepted
URLA · 9 pages
Closing Disclosure
2 exceptions · stale date, notary uncertain
Needs review
2 exceptions · stale date, notary uncertain
Notary Acknowledgment
Missing seal · re-upload requested
Rejected
Missing seal · re-upload requested
Borrower Authorization
Wet signature 0.94 · barcode ok
Accepted
Wet signature 0.94 · barcode ok
Egress boundary
GPU / model workers · no internet during processing
Cloud model calls
BlockedDocument egress
BlockedBorrower content in logs
RedactedTenant & object scoping
Enforced
Privacy by architecture
Lending documents contain NPI. Docutise is built like they do.
Processing is local-first with no cloud path to leak to. That's not a setting you can flip off — it's the shape of the system.
- Zero document egressOriginals, OCR text, barcode values and model outputs are treated as sensitive and stay local.
- Tenant-, object- and role-scopedCross-tenant access is prevented at the data layer, and admin-only actions gate every final decision.
- Signed, short-lived, audit-logged linksExports and evidence bundles are sensitive artifacts — never casual downloads.
8
Independent evidence lanes on every page
0
Cloud model calls — processing is local by design
5
Document classes in the frozen v1 audit scope
100%
Findings backed by a page, region & confidence
Built API-first
Drive audits from your LOS, POS or document platform.
A stable, documented API lets external systems upload documents, start audits, retrieve findings, resolve exceptions and download reports — without touching UI internals.
- Retries won't create duplicate workUploads and audit starts are idempotent — a flaky connection can't spawn duplicate documents or audits.
- Signed download pathsRetrieve originals, page artifacts and export bundles through short-lived, signed links.
audit.sh
# 1 — upload a document (stays on your hardware) curl -X POST https://your-host/documents \ -H "Authorization: Bearer $KEY" \ -H "Idempotency-Key: $(uuidgen)" \ -F "file=@closing_disclosure.pdf" # 2 — start the audit curl -X POST https://your-host/documents/doc_9f2a/audit # 3 — read the evidence-backed scorecard GET /audits/aud_71c/scorecard → { "decision": "needs_review", "overall_score": 72, "findings": [ … ] }
Questions
What teams ask first.
Does anything leave our environment?
No. Docutise runs local-first: there are no cloud model calls and no alternate cloud inference path. Original documents, rendered pages, OCR text, barcode values and model outputs are all treated as sensitive and stay on your hardware. Model and GPU workers are intended to run without normal internet egress during processing.
What document types are in scope today?
The frozen v1 audit scope covers Loan Estimates, Closing Disclosures, the Uniform Residential Loan Application (URLA), Borrower Authorizations and Notary Acknowledgments — each with its own required checks, date policy, signature evidence and notary indicators.
How does it avoid false passes?
Docutise is tuned for a near-zero false-pass posture. Deterministic checks run first; local detector and semantic lanes are used only where certified. When required evidence is unavailable or uncertain, the file is routed to manual review rather than passed — and an admin makes the final acceptance or rejection call.
Is this a finished production system?
Docutise is a local-first prototype today, with a working API, review console, evidence model and local processing stack. We're in early access while we complete detector certification, signed-PDF fixtures and the production hardening described on our company page. We'd rather be honest about that than oversell it.
Can we integrate it with our existing systems?
Yes. Docutise is API-first, so a LOS, POS, document platform or internal loan-file system can upload documents, start audits, retrieve findings, resolve exceptions and download durable reports programmatically. See the developer guide.
Bring evidence-grade QC to your loan files.
Request early access and we'll help you stand up a local pilot on your own hardware.