DownloadRampart is a lightweight, on-device privacy protection model developed by the National Design Studio to detect and redact personally identifiable information (PII) before text leaves a user's device. Rather than relying on server-side filtering, Rampart performs token-level PII detection locally, enabling privacy-preserving AI interactions with minimal latency and without exposing sensitive information to external services. The released model is a 14.7 MB ONNX artifact based on a fine-tuned MiniLM-L6-H384 encoder with approximately 18.5 million parameters and a 35-label BIO classification head covering 17 entity types. It works alongside a deterministic recognizer that handles structured identifiers such as phone numbers and IDs, forming a defense-in-depth client-side redaction system. Rampart supports English, Spanish, French, German, Italian, Portuguese, and Dutch, and is designed to run efficiently even on older devices, making it suitable for browsers, mobile and applications.
Features
- On-device PII detection with no server dependency
- 14.7 MB quantized ONNX model for lightweight deployment
- Fine-tuned MiniLM-L6-H384 architecture
- Detects 17 categories of personally identifiable information
- Supports English, Spanish, French, German, Italian, Portuguese, and Dutch
- Works with deterministic recognizers for structured identifiers
- Optimized for browsers, mobile devices, and older hardware
- Designed for privacy-preserving AI and client-side redaction
Project Samples
