Core Feature

Five-layer AI redaction
that understands context

Most redaction tools find patterns and black them out. Dezcry runs a multi-layer pipeline that understands why something should be redacted — then applies that understanding consistently across your entire document set.

Book a Demo

The pipeline

Each layer adds a different kind of intelligence. Enable all five for maximum accuracy, or configure the layers that match your workflow.

NER Detection

Pattern recognition at scale

A named entity recognition engine identifies names, addresses, phone numbers, emails, financial data, and dozens of other entity types using statistical models and custom pattern matchers. This layer catches the obvious — fast, reliable, and comprehensive.

LLM Analysis

Context-aware decisions

A large language model evaluates each detection in context. A name mentioned in a news article is different from a name in a complaint. This layer understands intent, distinguishing data subjects from bystanders, personal data from public information.

Verification

Independent cross-check

A separate LLM deployment cross-checks every decision from L2. This adversarial pass catches false positives and surfaces missed entities — the same principle as dual-reviewer workflows, automated.

Entity Resolution

Consistency across documents

The same person, address, or account number should be redacted the same way everywhere. Entity resolution ensures that "John Smith" in document 47 is handled identically to "J. Smith" in document 3,200.

Review Routing

Humans where it matters

High-confidence redactions are applied automatically. Low-confidence ones are flagged for human review with the context needed to make a fast decision. Your team spends time on edge cases, not routine work.

Not just for DSARs

Automated redaction is useful wherever sensitive information needs to be removed before documents change hands.

Data Subject Access Requests

Redact third-party personal data from disclosure packages. Meet regulatory deadlines without throwing bodies at the problem.

Privilege Review

Identify and redact legally privileged content before production. Consistent application across the entire document set.

Ad Hoc Redaction

Sensitive board materials, HR investigations, M&A due diligence — any scenario where information needs to be removed before sharing.

Regulatory Investigations

Produce redacted document sets for regulators while protecting information that falls outside the scope of the request.

What makes this different

Context-aware: understands the difference between a data subject and a person mentioned in passing

Configurable confidence thresholds — control how much automation to apply per entity type

Sensitive categories (health, political opinions, sexual orientation) default to stricter thresholds

Checkpoint resume: if a job crashes mid-way, it picks up where it left off

Full audit trail on every redaction decision — who, what, when, and why

Deploy into your own Azure tenancy or a dedicated Dezcry environment — all AI processing stays with your data, no third-party APIs

featureClassification.dsarCalloutLabel

featureClassification.dsarCallout

Next feature

AI Classification