Detecting Government-Issued PII

Personally Identifiable Information (PII) issued by government agencies—including Aadhaar numbers, PAN numbers, passport numbers, voter IDs, and driving license numbers—forms the foundation of an individual's identity and security. As organizations increasingly transition from physical records to digital documentation, the volume of sensitive information embedded within documents has increased substantially. With the widening threat landscape involving cyberattacks, data breaches, identity theft, and non-compliance with privacy regulations such as GDPR and India's Digital Personal Data Protection Act (DPDP 2023), there is a pressing need for automated systems capable of identifying and securing government-issued PII. This paper presents a comprehensive implementation of a hybrid PII Detection System that integrates Optical Character Recognition (OCR) for text extraction, rule-based pattern matching for deterministic identifier recognition, transformer-based Named Entity Recognition (NER) for contextual classification, and a secure Spring Boot backend for controlled access and encrypted storage. The system is designed to analyze multiple document formats, including scanned images, PDFs, and text-based files. It extracts embedded text, identifies potential government-issued PII, computes confidence scores using a fusion model, highlights detected information, and stores it securely for audit and compliance processes. The implementation is evaluated across a heterogeneous dataset of structured and unstructured documents, demonstrating high reliability and accuracy. This paper outlines the system architecture, mathematical modeling, implementation procedures, interface modules, testing environment, experimental outcomes, limitations, and future enhancements. The hybrid approach significantly improves detection accuracy compared to standalone methods, enabling its deployment in sensitive environments such as government offices, financial institutions, and corporate data governance frameworks.