Presidio Optimizer

AI-driven PII detection configuration optimizer built around Microsoft Presidio. Upload documents, analyze PII entities, provide feedback on false positives/negatives, and let Claude AI automatically optimize the detection configuration.

Designed specifically for Swedish PII detection with support for personnummer, samordningsnummer, organisationsnummer, and Swedish phone numbers.

Features

• Document Upload — Supports DOCX, XLSX, PDF, and TXT files
• PII Analysis — Detects entities using Presidio with Swedish and English NER models
• Interactive Feedback — Mark false positives by clicking entities, tag false negatives by selecting text
• AI Optimization — Claude analyzes feedback patterns and suggests configuration changes (thresholds, recognizers, entity toggles)
• Version Tracking — Every optimization creates a new versioned configuration
• Session Management — Named persistent sessions for iterative refinement across multiple rounds
• Report Generation — Downloadable Markdown reports with session history and optimization reasoning
• Config Export — Export optimized configurations as JSON or YAML

Architecture

┌─────────────────┐     ┌──────────────────┐
│  Nginx (frontend │────▶│  FastAPI backend  │
│  + reverse proxy)│     │  (Presidio +      │
│  Port 18011      │     │   Claude API)     │
└─────────────────┘     │  Port 18010       │
                         └──────────────────┘
                                  │
                         ┌────────▼────────┐
                         │  File-based DB   │
                         │  (db/sessions/)  │
                         └─────────────────┘

• Frontend: Vanilla HTML/CSS/JavaScript with a glassmorphism dark theme
• Backend: Python 3.11, FastAPI, Microsoft Presidio, spaCy (sv + en models)
• LLM: Claude API (Anthropic) for intelligent configuration optimization
• Storage: File-based JSON — no external database required

Swedish PII Entities

Entity	Description
SWEDISH_PERSONNUMMER	Swedish personal identity numbers (with Luhn validation)
SWEDISH_SAMORDNINGSNUMMER	Coordination numbers (day 61–91)
SWEDISH_ORGANISATIONSNUMMER	Organization numbers
SWEDISH_PHONE_NUMBER	Swedish phone patterns (+46, 07X, 0XX)

Plus all standard Presidio entities: PERSON, EMAIL_ADDRESS, PHONE_NUMBER, LOCATION, DATE_TIME, CREDIT_CARD, IBAN_CODE, IP_ADDRESS, URL, etc.

Quick Start

Prerequisites

• Docker and Docker Compose
• An Anthropic API key

Setup

1. Clone the repository:

   git clone https://github.com/thomaswennersten/presidio-optimizer.git
   cd presidio-optimizer

2. Create your environment file:

   cp .env.example .env
   # Edit .env and add your Anthropic API key

3. Start the application:

   docker-compose up -d --build

4. Access the application:

   • Frontend: http://localhost:18011
   • API: http://localhost:18010

Usage

1. Create a session — Give it a descriptive name
2. Upload a document — Drag & drop or click to upload (DOCX, XLSX, PDF, TXT)
3. Run analysis — Presidio scans the document for PII entities
4. Review results — Entities are highlighted in the text with color-coded categories
5. Provide feedback — Click entities to mark false positives, select text to tag false negatives
6. Optimize — Claude analyzes your feedback and generates an improved configuration
7. Iterate — Re-analyze with the new config and repeat until satisfied
8. Export — Download the optimized configuration as JSON or YAML

Project Structure

presidio-optimizer/
├── docker-compose.yml          # Standalone Docker orchestration
├── Dockerfile                  # Python 3.11 + spaCy models
├── nginx.conf                  # Reverse proxy configuration
├── .env.example                # Environment variable template
├── backend/
│   ├── main.py                 # FastAPI application & endpoints
│   ├── presidio_service.py     # Presidio analyzer engine
│   ├── config_manager.py       # Versioned config persistence
│   ├── session_store.py        # File-based session storage
│   ├── document_processor.py   # Multi-format text extraction
│   ├── feedback_processor.py   # Feedback normalization & aggregation
│   ├── llm_optimizer.py        # Claude API integration
│   ├── custom_recognizer_factory.py  # Dynamic recognizer creation
│   ├── swedish_recognizers.py  # Swedish-specific PII recognizers
│   └── requirements.txt        # Python dependencies
├── frontend/
│   ├── index.html              # Main SPA page
│   ├── css/styles.css          # Dark glassmorphism theme
│   └── js/
│       ├── app.js              # Main application controller
│       ├── api-client.js       # REST API client
│       ├── text-annotator.js   # Interactive text highlighting
│       ├── file-upload.js      # Drag & drop file handling
│       ├── config-panel.js     # Configuration viewer & export
│       ├── iteration-history.js # Version timeline
│       └── session-manager.js  # Session CRUD
└── db/                         # Persistent file-based storage
    └── sessions/               # Per-session data (gitignored)

API Endpoints

Method	Endpoint	Description
POST	/api/sessions	Create a new session
GET	/api/sessions	List all sessions
GET	/api/sessions/{id}	Get session details
DELETE	/api/sessions/{id}	Delete a session
POST	/api/sessions/{id}/upload	Upload a document
POST	/api/sessions/{id}/analyze	Run PII analysis
POST	/api/sessions/{id}/feedback	Submit feedback
POST	/api/sessions/{id}/optimize	Trigger AI optimization
GET	/api/sessions/{id}/config	Get current configuration
GET	/api/sessions/{id}/config/export	Export config (JSON/YAML)
GET	/api/sessions/{id}/report	Download analysis report

License

MIT