Panic Proof ✌️

AI-powered multimodal disaster preparedness and emergency response — now with offline Local LLM support and Expo Development Build.

Table of Contents

• Overview
• Features
• Tech Stack
• Architecture Highlights
• Getting Started
• Development Build (Expo Dev Client)
• Offline AI (Local LLM)
• Screenshots
• Acknowledgments

Overview

Panic Proof helps people prepare for and respond to disasters with an AI assistant, location-aware alerts, and gamified checklists. The app works online with Google Gemini and offline using a local model via ExecuTorch.

Features

• AI Assistant: Structured, actionable checklists for emergency prep
• Smart Alerts: Location-based disaster updates and guidance
• Gamified Checklists: Progress, points, and achievements
• Locations & Contacts: Save places, routes, and emergency contacts
• Resource Library: Curated content for major disaster types
• Profiles: Household needs and personalized plans

Tech Stack

• React Native + Expo (SDK 52)
• Expo Router, Expo Modules, Expo Dev Client
• TypeScript, Zustand (state + persistence)
• Supabase (Postgres, RLS, real-time), Clerk (Auth)
• AI: Google Gemini (online) + ExecuTorch (offline local LLM)

Architecture Highlights

• Development Build first-class: We use Expo Dev Client for native modules (image-picker, location, secure-store, executorch).
• LLM Provider Abstraction: online | local | auto with a small service layer and a zustand store to switch providers.
• Model Lifecycle: Download → verify → persist metadata → load for inference. Stored under app documents directory (llm/).
• Consistent Chat Pipeline: Online and offline providers produce identical structured JSON for checklists.

Getting Started

Prerequisites

• Node.js 18+
• Xcode (iOS) and/or Android Studio
• CocoaPods (macOS): sudo gem install cocoapods
• Accounts: Supabase, Clerk, Google AI Studio

Install

1. Clone and install

git clone https://github.com/jamiul-islam/panic-proof.git
cd panic-proof
npm install
# or
bun install

1. Environment variables (create .env.local)

EXPO_PUBLIC_CLERK_PUBLISHABLE_KEY=your_clerk_publishable_key
EXPO_PUBLIC_SUPABASE_URL=your_supabase_project_url
EXPO_PUBLIC_SUPABASE_ANON_KEY=your_supabase_anon_key
EXPO_PUBLIC_GEMINI_API_KEY=your_google_gemini_api_key

1. Supabase setup

• Apply schema and enable RLS as per project docs. Ensure JWT from Clerk maps to your user row.

Development Build (Expo Dev Client)

npx expo prebuild

First run builds a native dev client with required modules. Afterwards, you iterate with fast refresh via expo start.

iOS (Simulator)

npm run run:ios
# or
bun run run:ios

Android (Emulator)

npm run run:android
# or
bun run run:android

Start the bundler (reuse for both platforms)

npm start
# or
bun start

Notes

• We use app.config.js (no app.json) to avoid config drift.
• Hermes/new architecture enabled by default via Expo SDK 52.
• If native dependencies change, rebuild the dev client with the same commands above.

Offline AI (Local LLM)

The app supports full offline inference using ExecuTorch.

What’s included

• Provider toggle (Online/Offline) via the profile modal
• Model download with resumable progress and cancel
• Metadata persistence (metadata.json) and verification on launch
• Consistent structured JSON responses across providers

How to use

1. Open Profile → Offline AI Support
2. Select “Offline” and download the model when prompted
3. After “Ready,” start a chat — checklists are generated locally

Storage & Simulator notes

• Models are stored under the app documents directory at llm/ with a metadata.json file.
• On iOS Simulator, you can inspect/remove files via Xcode → Devices and Simulators → select the app → Download Container.
• Reinstalling the app or clearing data removes the model and metadata.

Fallbacks & errors

• If verification fails or inference errors occur, the app guides you to retry or switch back to Online.

Screenshots

Acknowledgments

Academic Institution

• University of London - Final Project supervision and guidance
• Computer Science Program - Technical framework and requirements

Technologies & Services

• Expo Team - For the exceptional React Native development platform
• Supabase - For providing enterprise-grade PostgreSQL with real-time capabilities
• Clerk - For robust authentication and user management services
• Google AI - For Gemini API enabling intelligent emergency assistance
• Zustand Team - For lightweight and efficient state management
• React Native ExecuTorch team - For blazingly fast local llm support

Open Source Community

• React Native Community - For continuous improvements and extensive ecosystem
• TypeScript Team - For enhanced developer experience and type safety
• Emergency Management Professionals - Whose guidelines informed our preparedness recommendations

Special Recognition

This project demonstrates practical application of:

• Full-stack mobile development with modern tools
• AI integration for social good and emergency preparedness
• Secure database architecture with real-time capabilities
• User experience design for critical emergency applications

---

Built with ❤️ for a safer, more prepared world.

A University of London Computer Science Final Project by Jamiul Islam