An app that optimizes routes by pollutant exposure instead of time. NASA TEMPO satellite data + OpenAQ ground stations + AQI-weighted Dijkstra. Continuous Apple Watch monitoring and AR visualization of the 2 000 PM2.5 particles you breathe every second.
Three devices, one source of truth
Same data layer across all three surfaces. The Django backend aggregates NASA TEMPO + OpenAQ + WAQI + OpenMeteo every 15 minutes; the iPhone runs AQI-weighted Dijkstra; the Watch does passive monitoring with haptics; on-device CoreML predicts PM2.5 at 1h, 3h and 6h.
iPhone · Biometric PPI
Personal Pollution Index combines ambient air with live Apple Watch biometrics. Cigarette equivalence, dose during activity, baseline against personal average. Component scores for SpO2/HRV/Heart Rate/Resp Rate.
iPhone · Body Atlas
Visualization of the affected systems — respiratory, nervous, muscular, skeletal — with real-time load indicators driven by ambient AQI. Educational: air isn't just lungs.
Walkthroughs
Three demos recorded during the hackathon — general overview, AQI routing engine and exposure analysis. Click play to watch each one.
Capabilities · Full system
AirWay doesn't just show pollution: it acts on it. Each block tackles one stretch of the flow — from the TEMPO satellite to the haptic on the wrist of an asthmatic runner.
ARKit + RealityKit render the PM2.5 particles you're breathing in real time. Mesh caching pre-caches 4 sphere sizes via NSLock; progressive warmup (300 → 800 → 2000) delivers the first particles in under a second. Stable 60 FPS from iPhone 12 onwards.
Satellite + ground fusion
fusedAQI = (satellite × 0.3) + (groundAvg × 0.7). NO₂, O₃ and HCHO from GES DISC + PM2.5/PM10 from ground stations, recalibrated with OpenMeteo and RAMA CDMX every 15 min.
15 min refreshSmart routing
edgeWeight = distance × (1 + AQI_penalty) + incidents. Sampled every 150 m with a 9-point grid centered on origin, destination and midpoint. Three routes: clean, fast, safe. Customizable 40/30/30 score.
0.8 s per calcApple Watch · PPI
Wrist companion with haptic alert when you enter a polluted zone. Route compressed to key waypoints (<5KB) syncs via WatchConnectivity.
<5KBML prediction
Three PM25Predictor models exported to CoreML from XGBoost (pandas + LightGBM + MAPIE). On-device inference, 30 min heads-up before crossing the threshold.
~85% acc 6hBody Atlas
4 affected systems — respiratory, nervous, muscular, skeletal — with progressive load per system driven by ambient AQI. The educational piece the jury highlighted.
4 systemsVulnerable profiles
Each profile tunes thresholds, recommendations and haptic sensitivity. The app learns when to stay quiet and when to shout — because there's no one-size-fits-all in air.
4 profilesHackathon stack
Bets with a zero-time budget: SwiftUI over UIKit, Django over FastAPI for mature adapters, XGBoost over neural nets to iterate fast. Each call paid for itself.
Closing
AirWay was born at the NASA Space Apps Challenge 2025, challenge Leveraging TEMPO for Predictive Air Quality Forecasting. What we shipped in 48 hours is the foundation of a product that keeps iterating with Render backend and versioned CoreML models.
“The hardest part was convincing people that air quality isn't a stat — it's a decision. Every time you open a routing app, you decide how much you'll breathe. AirWay just makes it explicit.”
— Retro notes, day 2
