SoloSentry

Inspiration

Solo travelers, journalists, and at-risk individuals face threats when no help is nearby. Cloud-based safety apps fail when networks are unavailable or devices are compromised.

We asked: what if personal safety intelligence lived entirely on the device?

SoloSentry was inspired by the need for instant, private, offline protection.

What It Does

SoloSentry is a mobile app that continuously monitors threats:

Uses ambient audio, location, and optional visual cues

Detects suspicious behavior or danger in real-time

Sends offline alerts and guidance

Encrypts evidence locally, includes panic-wipe mode

Works entirely offline—no cloud, no data exposure

How We Built It

RunAnywhere SDK orchestrates models efficiently on-device

Whisper (Quantized) captures audio for threat detection

DeepSeek R1 (Distilled) analyzes behavior and generates guidance

Llama-3 Vision (Optional) detects visual anomalies

Local TTS delivers silent voice instructions

All processing happens entirely on-device, ensuring zero latency and privacy.

Challenges

Running multi-modal reasoning offline with minimal latency

Designing for high-stress scenarios and noisy environments

Encrypting sensitive evidence and supporting a panic-wipe

Balancing threat detection accuracy with false positive safety

What We Learned

Privacy-first AI requires offline-first design

Multi-modal SLMs can deliver life-critical insights on-device

Latency matters more than perfection in emergency scenarios

Ethical AI is more than privacy—it’s real-world survival

What’s Next

Enhance multi-language audio and dialect support

Improve threat scoring using behavioral personalization

Integrate offline community safety alerts

Expand visual threat detection while maintaining privacy

SoloSentry demonstrates that offline AI is essential for real-time, life-saving protection.

Built With

• kotlin
• python
• runanywhere
• whisper