SHIELDAI
Enterprise intelligent data anonymization system developed for Scalian that protects personal data sent to LLMs, combining AI, PII detection, and real-time processing for GDPR compliance
The Problem
Organizations need to use LLMs (artificial intelligence) to process information, but this involves sending personal data to external services. They lack tools that automatically protect personal information before sending it to LLMs while complying with GDPR. Manual processes are error-prone and do not scale.
The Solution
Development of a platform that acts as a secure intermediary between companies and LLMs. The system automatically detects personal information using specialized AI models for Spanish, replaces it with synthetic data via Faker before sending queries to LLMs, and restores original data in received responses. All through mappings stored in Redis with TTL. Includes React interface with real-time streaming showing three panels: anonymized data sent, LLM response, and final response with restored data.
Impact
Enables organizations to use LLMs securely while complying with GDPR, automates personal data protection, reduces processing times and provides complete traceability with Prometheus/Grafana monitoring system. Optimized system with ~500 documents per minute processing, ~50ms per document anonymization, <100ms streaming latency and ~10ms per de-anonymization operation.
Category
Artificial Intelligence & Full Stack
Completion date
October 2025
Tech Stack
Highlights
- Automatic protection of personal data sent to LLMs
- PII detection with transformers optimized for Spanish
- Intelligent anonymization with realistic synthetic data using Faker
- De-anonymization system with Redis mappings and configurable TTL
- Multi-format processing: text, PDF, Word, Excel, images
- Image anonymization with facial detection (RetinaFace, MTCNN)
- Real-time streaming with three synchronized panels
- Complete monitoring stack: Prometheus, Grafana, AlertManager
Architecture
Demonstration
Monitoring & Metrics
Key Features
- Secure intermediary between companies and external LLMs (Groq API)
- FastAPI backend with RESTful endpoints and automatic Swagger documentation
- Enhanced regex pattern detection: DNI, NIE, IBAN, phones, emails with validation
- Consistent mapping: same entities generate same synthetic replacements
- Isolated sessions: each process maintains its own context without data mixing
- React frontend with Tailwind CSS, drag & drop for files
- Real-time metrics system: detections, performance, errors
- Multi-format result export with structure preservation
Next project
LOGOTRACKERPRO