Product Status: Planned / Architecture Phase
DocuMask is an engineering project currently in architectural specification. We do not provide a live or mock file upload widget on this page because secure, irreversible document redaction requires rigorous validation before public release.
The Problem: The "Black Box" Redaction Trap
A widespread security failure in digital document handling occurs when users or basic editing software place visual black rectangles over sensitive text. In many PDF viewers, the underlying vector text layer remains completely intact, readable via simple copy-paste or text extraction utilities.
True redaction requires that sensitive characters, vectors, raster pixels, and embedded metadata be irrevocably excised from the document stream.
Intended 5-Step Redaction Workflow
- Select Document: Open a local PDF or image inside an isolated client-side WebAssembly environment.
- Inspect PII Candidates: Automated pattern heuristics flag candidate phone numbers, email addresses, and identifier strings.
- Review Detected Regions: The user visually inspects highlighted bounding boxes and adds custom manual redaction areas.
- Execute Structural Sanitization: The engine excises underlying text streams, rasters affected image segments, and purges metadata dictionaries.
- Export Clean Artifact: Generate a freshly synthesized PDF or image binary directly in browser memory.
Technical Engineering & Security Considerations
Our architectural research evaluates several critical challenges prior to code freeze:
- Underlying Text Streams: Ensuring that font glyphs, CID maps, and PDF content streams corresponding to redacted text are purged, not merely covered.
- OCR & Scanned Media: For image-only PDFs, the underlying scanned raster pixels must be irreversibly overwritten with solid neutral data, and any embedded OCR text layer synchronized.
- Metadata Dictionaries: Stripping document author metadata, creation timestamps, thumbnail previews, and revision history streams.
- Memory Safety: Ensuring raw document chunks in browser memory are purged upon session termination and not retained in local caches.
- Detection Limits: Pattern heuristics can assist human reviewers but cannot guarantee detection of every idiosyncratic sensitive string. Human review remains mandatory.
Browser-Confined Security Model
DocuMask will adhere strictly to Oompha Labs' privacy baseline: zero transmission of user documents to remote cloud infrastructure. All parsing, redaction synthesis, and file generation will take place in the user's local browser runtime.
Inquiries & Security Discussion
If your organization handles sensitive documents and has specific compliance or redaction requirements you would like to discuss with our engineering team: