The Problem Space
Sensitive datasets require varying degrees of anonymization depending on viewer clearance, yet standard tools apply blunt all-or-nothing black boxes that destroy context.
Back to all projects
TrueSight
Gradational redaction across media.
NLP and computer vision tool allowing user-defined gradational masking across text, images, PDFs, and video formats.
ModalityText · Image · PDF · Video
RedactionGradational / Tiered
StorageZero Retention
The Architecture & System
Engineered an NLP and computer-vision redaction engine that classifies entity sensitivity and applies gradational masking (blur, pixelate, token replacement, cryptographic hashing) across heterogeneous document types.
Python OCR and named entity recognition (NER) pipelines integrated into an interactive web interface with zero-retention ephemeral processing.
Key Technical Decisions
- Implemented hierarchical sensitivity tiers so public, internal, and confidential viewers receive proportionate views.
- Zero-retention architecture: all input data is processed in ephemeral memory without persistent storage.