Surviv is an offline-first crisis response app for iOS. It lets nearby phones form a local mesh to share warnings, map threats, and coordinate safer movement when cellular and internet access are unreliable or unavailable.

During outages and infrastructure disruption, people lose trusted, timely information first. Without that signal, communities cannot quickly verify threats or coordinate safe decisions.

Surviv delivers three core capabilities on-device:

• Mesh alerts: multi-hop peer-to-peer broadcast with no internet dependency.
• Shared hazard map: danger and safe-route pins synchronized across nearby peers.
• On-device AI detection: microphone classification that can auto-create danger alerts.

• Works offline by default.
• Designed for fast decisions under stress.
• Converts raw events into map-level guidance.
• Uses commodity phones with no extra hardware.

Recent local and global conflicts highlighted a simple need: civilians need a direct way to warn each other when centralized systems are slow, unavailable, or compromised. Surviv was built to support community-to-community protection using the devices people already carry.

• Networking: Multipeer mesh relay with dedup and hop limits.
• Mapping: shared hazard model with timestamped history.
• AI: PyTorch training pipeline exported to Core ML for on-device inference.
• Persistence: SwiftData for continuity when links drop.

• Hardest problem: separating normal city audio from true danger classes such as gunfire and shelling.
• Current model performance: around 81% accuracy and 85% recall.
• We bias toward over-warning: false positives are more acceptable than missed threats in high-risk settings.

• Built a clear interface without sacrificing speed.
• Shipped an end-to-end iOS + mesh + AI pipeline as a student team.
• Focused the product on practical civilian coordination, not demo-only features.

• Swift and SwiftUI application engineering at production depth.
• Low-latency peer relay design using Apple networking primitives.
• Reliable on-device ML deployment from PyTorch to Core ML.

• iOS: SwiftUI, MapKit, SwiftData, CoreLocation, AVFoundation, MultipeerConnectivity, Core ML
• AI: Python, PyTorch, torchaudio, NumPy

1. Open ios-app/surviv.xcodeproj in Xcode.
2. Build and run on simulator or device.
3. Grant microphone and location permissions.

python3 -m venv .venv
source .venv/bin/activate

pip install --upgrade pip

pip install -r requirements.txt

cd ai-engine

python train_mad.py --data-root MAD_dataset --epochs 30 --out-dir mad_runs/default

python test_mad.py --checkpoint mad_runs/default/best.pt --data-root MAD_dataset

python live_mic_mad.py --checkpoint mad_runs/default/best.pt

Optional Core ML export:

pip install coremltools

python export_mad_coreml.py --checkpoint mad_runs/default/best.pt --out MADMelCNN.mlpackage

• Expand map support from a single local region to city-scale partitioned areas.
• Delivery confirmation for critical alerts.
• Report credibility scoring across multiple peer confirmations.
• End-to-end encrypted payload exchange.