Nirbhay

Inspiration

Every few months, we read another heartbreaking news story about a woman in danger who couldn't reach her phone in time. Existing safety apps require manual SOS activation—useless when someone is in panic, being coerced, or has had their phone snatched. We realized technology should work for the victim, not wait for them. That's why we built Nirbhay—to make women truly fearless.

What it does

Nirbhay is an autonomous AI-powered women's safety app that:

Detects danger automatically using phone sensors (accelerometer + gyroscope) to identify struggle patterns, sudden jerks, and abnormal movement

Triggers SOS without user input when a threat is verified

Shares live location instantly with emergency contacts via SMS

Suggests safer routes using crime hotspot data from OpenStreetMap

Works offline via cellular triangulation when internet is unavailable

Prevents false alarms through a 3-5 second re-verification layer

How we built it

Frontend: React Native with Expo + TypeScript for cross-platform compatibility, Tailwind for emergency-focused UI

Backend: Python FastAPI for high-performance async operations, WebSocket for real-time tracking, REST APIs for secure communication

AI/ML: GRU model for threat classification, NLP for distress keyword monitoring, signal processing (RMS, variance, jerk calculation) for motion analysis

Infrastructure: MongoDB for logs and user data, JWT authentication, Fast2SMS for alerts, Unwired Labs for cellular triangulation

AMD Integration: Leveraging EPYC processors for cloud hosting, Radeon GPUs with ROCm for model training, and Ryzen for on-device edge inference

Challenges we ran into

False alarm prevention: Balancing sensitivity to detect real danger while ignoring everyday movements like running or sudden braking

Offline functionality: Ensuring reliable location tracking in low-connectivity areas without draining battery

Sensor data accuracy: Interpreting raw accelerometer/gyroscope data to distinguish struggle from normal activity

Real-time performance: Maintaining low latency for SOS triggers while running silently in the background

Accomplishments that we're proud of

Building a fully autonomous SOS system—something no competitor offers

Creating a solution that works even when the user can't act

Successfully integrating crime hotspot mapping for proactive safety

Designing a lightweight model that runs efficiently on edge devices

Completing a working prototype that could save lives

What we learned

Human safety requires zero dependency on user action during panic

Sensor fusion + AI can reliably detect physical distress patterns

Offline-first design is critical for real-world safety scenarios

AMD's heterogeneous computing (CPU+GPU) accelerates both training and inference

The most impactful AI solves problems people face but can't articulate in the moment

What's next for Nirbhay

Smartwatch integration for wearables with dedicated motion sensors

Community-sourced safe zones with crowd-sourced incident reporting

Police integration for direct emergency dispatch

Audio analysis to detect screams or threatening sounds

Expansion to elder care, child safety, and lone worker protection

Production deployment with government partnerships and NGO collaborations

Built With

• api
• authentication
• cellular
• encryption
• expogo
• fast2sms
• https
• infrastructure
• jwt
• limiting
• location-data
• natural-language-processing
• rate
• route
• safe-zones
• tailwind-css
• triangulation
• typescript
• websocket