SafeWander

Inspiration

SafeWander was born from a deeply personal need—watching loved ones with dementia struggle with the balance between independence and safety. Traditional GPS trackers only tell you where someone is, but they don't understand why their movement matters. We envisioned a system that could learn individual behavior patterns and provide intelligent, proactive care rather than just reactive alerts.

The name "SafeWander" embodies our core philosophy: everyone deserves the freedom to move while staying protected.

What it Does

SafeWander is an AI-powered GPS tracking and safety monitoring system designed specifically for patients with dementia, Alzheimer's, or other cognitive conditions. It goes beyond simple location tracking by:

Behavioral Learning: Establishes baseline movement patterns using statistical analysis
Intelligent Geofencing: Creates safe, home, and caution zones with automatic anomaly detection
Real-time Risk Assessment: Calculates risk scores based on:
- Distance from safe zones: $ d_{safe} = \min(\text{distance to each safe zone}) $
- Speed anomalies: $ \Delta v = |v_{current} - \mu_{speed}| $
- Time-based patterns: $ \text{risk}_{time} = f(\text{hour}, \text{historical patterns}) $
- Movement entropy: $ H = -\sum p_i \log p_i $ where $ p_i $ is the probability of visiting location cluster $ i $
Multi-level Alert System: Critical, high, medium, and low severity alerts with smart escalation
Caregiver Dashboard: Beautiful, intuitive interface with live maps, activity feeds, and vital statistics
Comprehensive Reporting: Exportable data for medical professionals and family members

The risk engine combines multiple factors into a unified risk score:

$$ \text{Risk}{total} = w_1 \cdot R{distance} + w_2 \cdot R_{speed} + w_3 \cdot R_{time} + w_4 \cdot R_{zone} $$

Where weights $ w_i $ are dynamically adjusted based on patient history and current context.

🛠️ How We Built It

Frontend (Next.js + TypeScript)

Built with Next.js 14 using App Router for optimal performance
Shadcn/ui component library for a polished, accessible interface
Leaflet.js for interactive map visualization with custom markers and zone overlays
Real-time updates using React hooks and automatic polling
Custom demo simulator for realistic movement pattern generation during development

Backend (Python + FastAPI)

FastAPI REST API with automatic OpenAPI documentation
SQLite database with SQLAlchemy ORM for reliable data persistence
Haversine formula for accurate distance calculations: $$ d = 2r \arcsin\left(\sqrt{\sin^2\left(\frac{\phi_2-\phi_1}{2}\right) + \cos(\phi_1)\cos(\phi_2)\sin^2\left(\frac{\lambda_2-\lambda_1}{2}\right)}\right) $$ Where $ r $ is Earth's radius, $ \phi $ is latitude, $ \lambda $ is longitude

AI/ML Components

Baseline Learning Algorithm: Tracks movement patterns over 7-day rolling windows
Anomaly Detection: Statistical deviation analysis using z-scores
State Machine: Manages alert states (pending → acknowledged → resolved)
Risk Engine: Multi-factor scoring with configurable thresholds

Architecture Highlights