We built this because nobody else would.
From a frustrated observation in 2024 to 15-layer mathematical certainty today. Walk the engineering journey behind Redactorr's browser-local redaction architecture.
"The moment we ran a real insurance claim, we realised textbook regex was completely blind."
Day One: The Cleartext Flaw
It started with a simple question: what if you could strip sensitive data from a document before it ever left your desk? We picked a country, picked a document type, and started writing the rules. We had no idea how deep this would go.
Not Good Enough: Real Document Reality
Five days in, we ran a real insurance claim through the system. It found the email and phone number. It missed the policy number, the claim reference, and three people’s names. So we tore the entire engine apart and started over.
Original Values Stay Local (WASM)
Most redaction tools send documents to a remote server. We established our non-negotiable boundary: 100% browser-local detection and token replacement in WebAssembly. Original cleartext strings never leave the device.
Inspect Trust Architecture →
50+ Global Identifier Formats
An Australian Tax File Number looks nothing like a US Social Security Number or a UK NHS number. Every country has distinct checksums and statutory edge cases. We engineered recognition across 50+ sovereign formats.
"A random 9-digit string is meaningless. A real ID follows strict checksum mathematics."
Confidence Scoring & Disambiguation
Is "123-456-789" a phone number, a case reference, or a Medicare card? A blunt regex highlights everything. We engineered probabilistic confidence scoring so reviewers see exactly why a string was flagged.
"The Supreme Court of Victoria in Section 123...""Witness [REDACTED: PERSON] residing at [REDACTED: ADDRESS]..."Eliminating False-Positive Noise
The engine was flagging street names in legal citations, and "Victoria" the state instead of Victoria the person. We executed 1,200+ targeted heuristics to separate sensitive entities from ordinary language context.
Statutory Identifier Validation
A random 9-digit string is just a number. A real TFN follows strict national formatting rules. We integrated structural validation across all national IDs, cutting false positives in half overnight.
From 8 to 15 Verification Layers
Tested against 90 million characters and 18,483 real-world documents. We built the 15-layer AEGIS pipeline where 12 specialist agents operate in parallel before the WARDEN consistency gate approves the clean payload.
Explore 15-Layer AEGIS Methodology →3,000+ Recognised Data Types
Medicare cards, BSBs, ABNs, passports, driver licences, AHPRA medical practitioner IDs, and bank account numbers. The pattern library crossed 3,000 verified document types.
"Fifteen verification layers. Twenty-eight industry packs. Zero server uploads."
UK, Canada, New Zealand & Ireland Packs
An Australian medical referral is not a British one. A Canadian tax return has different identifiers than a New Zealand one. We built dedicated country rule packs tested on genuine jurisdictional documents.
AEGIS Core Engine Architecture
We redesigned the entire detection architecture from scratch. Domain-specific rule packs, contextual classification, and zero cloud dependencies. The fourth major version that works the way enterprise documents actually work.
28+ Regulated Industry Engines
An insurance claim carries different risk than a court deposition or a pathology report. We deployed 28+ industry-specific engines for Legal, Healthcare, Banking, HR, DevOps, and Utilities.
Browse Industry Packs →The Sovereign AI Data Pipeline
Fifteen verification layers, each doing one thing ruthlessly well. Sub-millisecond latency in browser WASM. And a guarantee that your data remains sovereign, auditable, and mathematically protected.
Your data is yours. Period.
Thirteen milestones. Four architectural rebuilds. One immutable commitment to local data sovereignty.