import React from 'react'; import { HelpCircle, Layers, Cpu, Shield, RefreshCw } from 'lucide-react'; import { useNavigate } from 'react-router-dom'; export function HowToUse() { const navigate = useNavigate(); const levels = [ { level: 'Level 0', title: 'No Redaction (Pass-through)', desc: 'Returns the original document without modifications. Used for baseline logging and formatting verification.' }, { level: 'Level 1', title: 'Regex & Basic Rule Scrubbing', desc: 'Redacts structured text identifiers (PAN cards, Aadhaar, phone numbers, emails, bank accounts) across all documents via fast OCR.' }, { level: 'Level 2 - 4', title: 'Gradational spaCy Statistical NER + Visual Biometrics', desc: 'Layers spaCy NLP models for names, organizations, and locations across all files. For PDFs & Images: Level 3 automatically blurs Human Faces, and Level 4 blurs Handwritten Signatures.' }, { level: 'Level 5', title: 'Transformer Deep Learning + Total Biometrics', desc: 'Engages fine-tuned RoBERTa transformers for text/data files. For PDFs & Images: performs Total Biometric De-identification (Faces, Signatures, Official Colored Stamps, Seals, and Fingerprints/Thumbprints).' } ]; return (

How to Use

A step-by-step guide to gradational redaction and synthetic anonymization.

Gradational Redaction Scale

RE-DACT allows users to calibrate redaction intensity on a gradational scale from 0 to 5. The higher the degree set by the user, the deeper the NLP model inspects and obfuscates contextual markers.

{levels.map((item) => (
{item.level}

{item.title}

{item.desc}

))}

Synthetic Name Generation

When using advanced redaction tiers, RE-DACT can replace real personal names, organizations, and identifiers with realistic synthetic equivalents (e.g., replacing real names with synthetic Indian names like "Aarav Sharma" or "Priya Nair"). This produces realistic databases safe for sharing and analysis.

Active Learning Feedback Loop

If you notice a missed entity or make a manual correction in the UI, submit it via the feedback interface. The system stores corrections in an O(1) JSONL pipeline to automatically train and improve future model iterations.

); }