AI-powered hyperlocal air quality forecasting and personalized outdoor safety recommendations using satellite data, spatial interpolation, and machine learning!
Live Demo • Features • Architecture • Tech Stack
HawaGuide is an AI-powered environmental intelligence platform designed to help users identify the safest outdoor locations in Delhi based on predicted air quality and personalized health risk.
Unlike conventional AQI applications that rely solely on nearby monitoring stations, HawaGuide generates hyperlocal pollution estimates for 50+ Delhi green spaces using satellite-derived environmental data, spatial interpolation techniques, and machine learning forecasting models.
The platform introduces PEVI (Personalized Exposure Vulnerability Index), a health-aware metric that combines multiple pollutants with demographic and medical factors to recommend safer outdoor destinations.
- Hyperlocal AQI prediction across 50+ Delhi green spaces
- Hourly pollution estimation using Kriging and Gaussian Process Regression (GPR)
- 6-hour air quality forecasting using an ensemble of:
- XGBoost
- LSTM
- Prophet
- Personalized exposure assessment using:
- Age group
- Health condition
- Outdoor duration
- Multi-objective optimization balancing:
- Air quality
- Travel distance
- Interactive cyber-terminal inspired dashboard
- Real-time recommendation engine
The application provides:
- Personalized environmental exposure analysis
- Forecasted pollution trends
- Interactive optimization slider
- Green-space ranking
- Health-aware recommendations
- System monitoring interface
Traditional AQI represents environmental conditions but does not account for individual health differences.
HawaGuide introduces PEVI, a weighted health-aware exposure metric that combines:
- PM2.5
- PM10
- NO₂
- O₃
- CO
The score is adjusted based on:
- Age group
- Health condition
- Outdoor exposure duration
This enables personalized outdoor recommendations instead of generic AQI values.
- Kriging Interpolation
- Gaussian Process Regression (GPR)
Used to estimate pollution levels across regions without monitoring stations.
An ensemble model combines:
- XGBoost
- LSTM
- Prophet
to generate 6-hour forecasts with <8% Mean Absolute Percentage Error (MAPE).
Recommendations are generated using:
Score = α × Risk + (1 − α) × Distance
where
- α = 0 → Prioritize shortest travel distance
- α = 1 → Prioritize safest air quality
Users can dynamically adjust this trade-off through the interface.
Satellite Data
│
▼
Spatial Interpolation
(Kriging + GPR)
│
▼
Hyperlocal AQI Maps
│
▼
Feature Engineering
│
▼
Forecasting Ensemble
(XGBoost + LSTM + Prophet)
│
▼
PEVI Calculation
│
▼
Multi-Objective Optimization
│
▼
Interactive Dashboard
- React
- Tailwind CSS
- Framer Motion
- Recharts
- FastAPI
- Python
- XGBoost
- TensorFlow / Keras
- Prophet
- Scikit-learn
- PyKrige
- Gaussian Process Regression
- GeoPandas
- Rasterio
- Pandas
- NumPy
- Vercel
- GitHub
| Metric | Value |
|---|---|
| Coverage | 50+ Delhi Green Spaces |
| Refresh Rate | Hourly |
| Forecast Horizon | 6 Hours |
| Forecast Accuracy | <8% MAPE |
| Pollutants | PM2.5, PM10, NO₂, O₃, CO |
| Personalization | 4 Age Groups × 3 Health Conditions |
https://hawa-guide-hyperlocal-aqi-safety-pl.vercel.app
- Multi-city support
- IoT sensor integration
- Route-level exposure optimization
- Mobile application
- Real-time traffic-aware forecasting
- Satellite data automation
- Wearable device integration
PEVI – Personalized Exposure Vulnerability Index
Patent filed in April 2026 for a health-aware multi-pollutant environmental exposure metric designed to personalize outdoor safety recommendations.
This project is intended for research, academic, and demonstration purposes.