Pocket Guardian

• 

  Screen to Add contacts

• 

  Dashboard with SOS button

Inspiration

Many safety apps assume that users can always press a panic button. In reality, emergencies often involve fear, shock, or sudden immobility — moments when asking for help isn’t possible.

Pocket Guardian was inspired by the idea that safety systems should notice danger, not wait for it to be reported.

What it does

Pocket Guardian is an AI-powered personal safety app that helps users during emergencies — even when they can’t actively ask for help.

Key features:

Manual SOS with countdown and cancellation

Safe Walk Mode that monitors inactivity during risky situations

AI threat assessment using short voice input, movement state, location context, and time

Explainable AI decisions, showing why an action was taken

Automatic alerts sent to trusted contacts with relevant context

The app is privacy-aware and only activates AI analysis when risk is likely.

How we built it

Android (Jetpack Compose) for a modern, responsive UI

Gemini API for AI-based threat evaluation and decision-making

Room Database to store emergency contacts and safe zones

Sensors & Location APIs to detect inactivity and context

Speech Recognition for short, event-based ambient voice capture

Coroutines & Flows for safe, asynchronous processing

The AI logic is encapsulated in a dedicated decision engine that returns structured, explainable outputs.

Challenges we ran into

Handling real-time permissions for sensitive features like location, audio, and SMS

Ensuring speech recognition runs safely on the main thread

Making AI responses reliable and structured instead of free-form text

Avoiding false positives while still acting conservatively for safety

Designing a system that is privacy-conscious and not always listening

Each challenge required careful architectural and ethical decisions.

Accomplishments that we're proud of

Built an AI system that explains its decisions, not just acts

Designed Safe Walk Mode to escalate automatically without user input

Created a real-world usable safety flow, not just a demo concept

Balanced automation, transparency, and user control

Delivered a polished product experience within hackathon constraints

What we learned

Safety-focused AI must prioritize trust and explainability

Small design choices can have large ethical implications

Reliable emergency systems require fail-safe thinking

AI is most impactful when it supports humans quietly, not aggressively

What's next for Pocket Guardian

Smart wearable integration (smartwatch triggers)

Offline fallback logic for low-connectivity situations

Enhanced Safe Walk intelligence using historical patterns

Emergency service integration (where legally permitted)

Broader accessibility features for vulnerable users

Built With

• accelerometer
• android
• coroutines
• gemini
• geminiapi
• gradle
• jetpack
• kotlin
• material3
• room
• sensors