Project write-up

VayuLens

A neighbourhood-level pollution command system that combines citizen-uploaded smoke/dust photos, local PM sensors, satellite fire/AOD priors, and weather to detect hidden hotspots, predict AQI spikes, dispatch municipal resources, and prove the intervention worked.

Problem-Solution Fit

City AQI apps are too coarse for dump fires, construction dust, industrial plumes, and smog-trap junctions. VayuLens operates at ward and street-grid level, where residents actually experience exposure.

AI / Technical Execution

AI scores smoke/dust photos, estimates visual haze, detects sensor anomalies, fuses satellite priors, and forecasts 24-hour AQI spikes with explainable features.

Deployability

The prototype deploys through GitHub Actions to GitHub Pages. A city pilot maps to Cloud Run, Pub/Sub, Cloud SQL/PostGIS, Cloud Storage, Vertex AI, BigQuery, and Firebase Cloud Messaging.

Inclusivity

Cheap Android flow, compressed uploads, Hindi/regional labels, offline queue, voice note support, privacy blur, and low-literacy-friendly municipal workflows.

Working Prototype Flow

Citizen photo Sensor spike FIRMS prior AQI forecast Dispatch + proof

Architecture

Evidence CaptureAndroid/web reports, PMS7003/SDS011 sensors, NASA FIRMS, Sentinel/MODIS, IMD/ERA5 weather.
Fusion + ForecastCloud Run ML service, Vertex AI image model, weighted geospatial fusion, XGBoost AQI forecast, PostGIS grid state.
Municipal ActionICCC dashboard, asset assignment, FCM/SMS/WhatsApp alerts, SLA timer, before/after proof in BigQuery.

Impact

  • Detects hidden hotspots before city-level AQI moves.
  • Prioritizes action by exposed population, schools/clinics, and predicted AQI spike.
  • Creates a measurable accountability loop for municipal response.
  • Can start ward-by-ward with CSR-funded sensors and existing ICCC workflows.