Zero-Trust PII Masking & Cryptographic Reconstruction
Sanitize personally identifiable information (nombres, cédulas, correos, tarjetas) before sending prompts to external LLMs. Presidio NER replaces sensitive data with salted surrogate tokens, allowing instantaneous client-side restoration.
Raw Input Text (Original PII)
Sanitized Zero-Trust Output
Stage 3: Downstream LLM Inference & Local Reversible Reconstruction
Haga clic en "Simular Inferencia LLM" para observar el razonamiento del modelo sobre los tokens enmascarados...
Dynamic Client Integration Snippet
Surrogate tokens like [PERSON_3a7f1c08] are generated by taking the first 8 hex characters of a salted SHA-256 hash. Identical entities map to identical tokens across prompts, preserving conversational cohesion for the LLM without leaking raw data.
Conforms to Art. 10 & 37 of Ecuador's Ley Orgánica de Protección de Datos Personales (seguridad y seudonimización) and GDPR Art. 25. Prevents cloud AI providers from logging or training on sensitive citizen and patient identifiers.
Protected by Cloudflare Workers Backend-for-Frontend with strict CORS, rate limiting, and zero exposure of internal microservice secrets in client bundles.