Security engineering and machine learning, 2026
ML-Based Data Loss Prevention System
A privacy-conscious DLP platform for small accounting firms that classifies document sensitivity, evaluates user behaviour, and applies an explainable policy response.
Challenge
Smaller organisations need practical protection for sensitive financial documents without sending the raw contents of every file to a central service.
Approach
The architecture separates endpoint scanning, document classification, behaviour analytics, policy evaluation, alerts, incidents, reporting, and model management. The central service stores event metadata and risk signals, not raw client documents.
Result
The two-stage decision model supports proportionate allow, log, warn, alert, or block actions while making privacy a system boundary rather than an afterthought.
Core capabilities
- Document sensitivity classification
- User-behaviour anomaly detection
- Policy engine, incident management, and audit reporting
- Privacy-by-design event architecture
Discuss this work