presidio-pii-specialist

PII detection in dual-language text uses spaCy NER models combined with Presidio's analyzer to identify sensitive data like names, addresses, and identifiers. This Skill applies Polish and English models, custom recognizers for Polish identifiers (PESEL, NIP, REGON), regex patterns, and entity deduplication to redact detected PII in the final output.

Yes. Presidio integrates spaCy NER models with regex fallbacks, allowing both machine-learning-based entity recognition and pattern-based detection to work together. This Skill uses regex to augment ML detection and handle edge cases where patterns are more reliable than trained models.

This Skill includes custom recognizers for PESEL (national ID number), NIP (tax ID), and REGON (business registration number), the primary Polish identifiers. These recognizers detect and redact Polish-specific PII that standard NER models alone may miss.

Entity deduplication removes overlapping or duplicate PII detections across the 3-branch parallel detection pipeline, ensuring each piece of sensitive data is flagged and redacted only once. This prevents redundant redactions and conflicting sanitization decisions.

Detection uses spaCy's en_core_web_lg for English and pl_core_news_lg for Polish NER. These models identify entity types like persons, organizations, and locations, which are then cross-referenced with custom recognizers and regex patterns for comprehensive PII coverage.

Yes. The Skill integrates with Arbiter v2 for SANITIZE and ALLOW decision paths, supporting Vigil Guard v2.0.0 compliance pipelines. Invoke it after sanitization to redact detected PII in final response outputs.