Inspiration
Every woman on our team and every woman we know has experienced that moment of fear: walking home with keys between her fingers, pretending to be on a phone call, or texting a friend, "I'm walking now."
In Beasley and the surrounding neighbourhoods, we see the same fear reflected in our community. Too often, safety and well-being concerns are forced to escalate to police or hospitals when compassionate, local support could have helped much earlier. We wanted to change that.
Because when something feels wrong, waiting 12–19 minutes for help—the average high-priority police response time in major cities like Toronto—can feel like a lifetime. And many people don't call at all, afraid of making things worse.
So we built SafetyNet HER, so no woman has to face those moments alone.
What it does
SafetyNet HER is an AI-powered SMS safety system that:
- Predicts high-risk situations using past incident data
- Allows women to text simple commands like UNSAFE or CALL ME to get immediate help
- Connects users with nearby trained volunteers in minutes
- Provides discreet tools like fake emergency calls, real-time safety tracking, and walk-home escorts
- Prevents escalation by prioritizing community response before emergency services
If a user fails to respond to safety check-ins, the system automatically escalates support—from volunteers → trusted contacts → emergency services if needed.
How we built it
- Twilio – SMS interface & emergency communication
- Node.js + Express – Backend API
- MongoDB Atlas – Incident storage, volunteer network, geospatial matching
- Google Gemini API – Text analysis, urgency scoring, predictive risk modelling
- ElevenLabs – Realistic voice for the emergency "fake call" feature
- React + TypeScript + Vite – Volunteer & dispatcher dashboard
Challenges we ran into
- Designing the system to handle many different crisis scenarios while keeping the experience simple for users
- Coordinating AI analysis, database updates, and live volunteer dispatch with minimal delay
- Ensuring AI reliability—building a comprehensive keyword fallback system when Gemini API was inconsistent
- Balancing speed vs accuracy: achieving 2-second analysis while detecting subtle threats
Accomplishments that we're proud of
- Created a working predictive AI layer that detects high-risk locations and time patterns
- Successfully integrated Google Gemini, MongoDB Atlas, and ElevenLabs into one cohesive platform
- Built a crisis-response model that prioritizes community intervention and prevents unnecessary escalation
- Achieved 96% community resolution rate with only 4% requiring police intervention
- Built real-time tracking with 4.2-minute average response time vs 18 minutes for 911
What we learned
We learned that building safety technology is as much about empathy as it is about engineering. Understanding context—that "I'm going to die" requires different intervention than "someone following me"—requires both technical sophistication and human insight. We also learned the power of intelligent triage: most safety incidents don't need armed police, they need compassionate, trained community support.
What's next for SafetyNet HER
Our next step is building a real volunteer network and implementing a strong pre-screening process to ensure user safety and trust.
We plan to partner with shelters, campuses, and municipalities to expand coverage and deploy our predictive models city-wide. Our goal is to make SafetyNet HER as accessible as possible to the people who need it most.
Future features include:
- Native mobile app with push notifications
- Enhanced fake call feature with realistic conversations
- Integration with existing crisis hotlines
- Machine learning improvements for better pattern detection
- Volunteer credentialing and training modules ```
BUILT WITH:
react
typescript
nodejs
express
mongodb
twilio
google-gemini
elevenlabs
tailwindcss
vite
javascript
Log in or sign up for Devpost to join the conversation.