Skylens Aperture Logo
Skylens
Try Demo
person
WeMakeDevs Hackathon

Read the air from a photo.

Real-time hyperlocal air quality estimates powered by everyday smartphone photos.

Optical Vision Core
Active Model v2.4
HORIZON REF
AZ 142° • ELEV 38° 42,000 LUX
Optical Depth (AOD): 0.14 LIVE FRAME
Estimated AQI
32 Good Air Quality
eco
Coverage Gap

The Air Monitoring Blind Spot

Official stations cost $20k+ and sit kilometers apart, missing local smoke, highway exhaust, and neighborhood microclimates.

Standard Monitoring radar

Official Sensor Stations

Highly accurate but spaced 10–30 km apart. They miss street canyons, school zones, and transient pollution bursts.

Grid Coverage 16% Urban Span
Skylens Vision smartphone

Crowdsourced Street Optics

Every photo measures sunlight attenuation and atmospheric haze to deliver high-resolution street-level estimates.

Micro-Grid Resolution Sub-250m Street Precision
Process Pipeline

How Skylens Works

1
photo_camera

Snap

Frame any portion of the daytime horizon. Our live alignment guide secures sun azimuth and optical depth.

check_circle Automatic GPS & sun-angle lock
2
filter_drama

Estimate

Optical AI models infer aerosol optical depth and particulate density calibrated against official stations.

check_circle Sub-second neural inference
3
route

Share & Navigate

Readings populate live community maps, helping pedestrians choose cleaner, low-exposure routes.

check_circle Anonymized spatial mesh update
Live Demo

Try it right here

This is the real app on the real AWS backend. Tap Demo scan, pick a sample sky, and watch the pipeline run — no camera or GPS permission needed.

Demo mode uses bundled sample photos and a fixed Bengaluru location, so the live scan exercises the real upload → model → map pipeline without permission prompts.

How it is built

Architecture

A single Flutter client talks to a serverless AWS backend. Every photo is scored by a vision model, fused with official stations, and written to a geohash grid that powers the map and routing.

smartphone

Flutter app

Scan, Map, Route, Coach. Web build in the phone frame above.

hub

API Gateway

HTTP API with a Cognito JWT authorizer and CORS for the web origin.

bolt

Lambda functions

upload-url, scan, map, route, coach — plus a Cognito guest trigger.

storage

Data & model

S3 images, DynamoDB grid, SageMaker vision model, Bedrock coach.

Built with AWS

Seven services, one job each

Everything runs serverless, so there is nothing to keep alive between demos.

Cognito Build it

Guest sign-in so judges can try instantly without an account.

API Gateway Build it

Single HTTPS entry point that checks JWTs and applies CORS.

Lambda Build it

One function per endpoint; scales to zero when idle.

DynamoDB Ship it

Scans and geohash cells; the map and routing read straight from it.

S3 Ship it

Private bucket for photos and the model artifact, with a 30-day expiry.

SageMaker Ship it

Serverless endpoint running the sky-photo AQI classifier.

Bedrock Ship it

Powers the Air Coach, with a canned reply if the model is unavailable.

Team & links

Built for the WeMakeDevs Environmental Hack