"PII Redaction with Presidio: Privacy‑Safe Logs, Prompts, and Documents"
Modern systems leak sensitive data in more places than most teams realize: application logs, LLM prompts, retrieval pipelines, uploaded files, screenshots, and scanned documents. This book is written for experienced engineers, architects, privacy specialists, and platform teams who need to build serious de-identification systems with Microsoft Presidio—not toy demos. It treats PII redaction as a production engineering discipline, where architecture, policy, evaluation, and operational safety matter as much as API usage.
Readers will learn how Presidio’s analyzer and anonymizer engines work internally, how recognizers compete and cooperate, how NLP engine choices affect quality and cost, and how to design operator strategies for masking, replacement, irreversible redaction, and controlled reversibility. The book also covers custom recognizers, channel-specific policy design for logs, prompts, and documents, OCR-based image redaction, DICOM scenarios, and rigorous evaluation using precision, recall, regression testing, and risk-based acceptance thresholds.
Rather than repeating basic privacy concepts, the text focuses on implementation boundaries, trade-offs, and failure modes that emerge in real deployments. It is especially suited to readers comfortable with Python services, APIs, ML-adjacent tooling, and distributed systems, and it offers a version-aware, operations-conscious guide to deploying Presidio safely at scale.
