Full end-to-end solution

VayuLens detects, predicts, dispatches, and proves.

The challenge asks for a neighbourhood pollution map using citizen photos, local sensors, and satellite imagery. VayuLens implements that core flow and adds the missing civic-tech layer: municipal dispatch and proof that action worked.

20%Problem-solution fit
25%AI execution
25%Deployability
15%Inclusivity
10%Impact
5%Clarity

End-to-end working flow

1Citizen evidenceSmoke/dust photo with GPS, timestamp, Hindi note, offline queue.
2Sensor anomalyPMS7003/SDS011 PM2.5 spike corrected for humidity and local baseline.
3Satellite priorNASA FIRMS fire signal plus Sentinel/MODIS background pollution context.
4Forecast24-hour AQI peak and spike window using weather, lags, land-use, and evidence scores.
5Action proofDispatch mist cannon/crew and record before/after PM and haze reduction.

What is novel

  • Citizen photos act as soft visual PM sensors through smoke, dust, and haze estimation.
  • Hotspots require cross-source evidence instead of trusting one viral complaint.
  • The map predicts plume movement and AQI spike timing, not just current heat.
  • Municipal action is closed with before/after proof and SLA history.

Why it wins

  • Directly addresses the problem statement.
  • AI is visible in the workflow and explains each decision.
  • Runs publicly on GitHub Pages through CI/CD.
  • Has a credible Google Cloud path for real deployment.

Technology choices

Layer
Hackathon prototype
City pilot
Frontend
HTML, CSS, JavaScript, GitHub Pages
React/Next.js dashboard embedded in ICCC
Civic API
Simulated state machine in browser
Spring Boot on Cloud Run
ML
Deterministic image/sensor/satellite scenario + charted forecast
Vertex AI image model + XGBoost/LightGBM + PostGIS fusion
Data
Seeded demo scenario for reliability
OpenAQ/CPCB, NASA FIRMS, Sentinel-5P, MODIS AOD, IMD/ERA5
Deployment
GitHub Actions CI/CD to GitHub Pages
Cloud Run, Pub/Sub, Cloud SQL/PostGIS, BigQuery, FCM

Demo script

  1. Open the prototype and show the ward grid is initially Clear.
  2. Click Run dump-fire scenario.
  3. Point to photo evidence, PM2.5 spike, FIRMS prior, and wind consistency.
  4. Show AQI peak prediction and exposed population.
  5. Show mist cannon + cleanup crew recommendation.
  6. Click Record after-reading to close the proof loop.

Submission checklist

  • Working prototype: live GitHub Pages dashboard.
  • Phase 1 pages: report intake, evidence graph, forecast lab, dispatch operations.
  • Phase 2 page: AI copilot with municipal brief, citizen alert, and crew checklist.
  • Enterprise pages: ops, integrations, admin, analytics, and research basis.
  • Real OpenStreetMap base layer on the command dashboard.
  • First-time guided tour for evaluators.
  • Code repository: public GitHub repo with README.
  • Pitch deck: browser deck and Markdown source.
  • Project write-up: web page and Markdown source.
  • Architecture: rendered web diagram plus technical Markdown.