Islamabad Live SystemProduction Ready

Pearls AQI Predictor

Air quality forecasting, explained like an intelligent control room.

An advanced air quality forecasting platform for Islamabad that combines automated feature ingestion, daily model retraining, and a cloud model registry to deliver live, 3-day AQI predictions.

AQI72h
PM2.5
NO2
Models
Live AQI IntelligenceAutomated TrainingCloud Model RegistryProduction-Ready ML Pipeline
HourlyFeature Ingestion

Automated collection of weather and pollutant data.

DailyModel Training

Daily retraining pipeline to find the most accurate models.

3 DaysForecast Horizon

Serving Day +1, Day +2, and Day +3 AQI forecasts.

CloudSecure Storage

MongoDB Atlas stores features, metrics, and registered models.

Real-Time Forecast

A fully operational machine learning product.

Observe the entire pipeline: fresh data enters the cloud store, models train and update automatically, predictions are served via an API, and the dashboard presents the results with model explanation.

Feature store17 ActiveSeasonality, lags & trends
Serving speed< 85msFastAPI response time
Pipeline audits100% PassedZero duplicates or nulls
Scikit Models4 EvaluatedRidge, RF, GBDT, MLP
Active Pipeline TelemetryOperational
Ingestion: Hourly Sync
Retraining: Daily Cron
Database: GridFS Storage
API Gateway: FastAPI Host

Live preview

3-day Islamabad AQI forecast

Waking

Loading live forecast from the deployed API...

Day 1--

Day +1

--
Day 2--

Day +2

--
Day 3--

Day +3

--
Open full analytics dashboard

System Map

Four integrated layers powering the platform.

01

Feature Store

Clean hourly weather and pollutant features are validated and stored securely in MongoDB Atlas.

17 ParametersHourly IngestionQuality Audited

Deduplication & boundary checks run 24/7.

02

Model Registry

Ridge, Random Forest, Gradient Boosting, and MLP models compete daily. The best performers are registered.

Ridge, RF, GBDT, MLPGridFS VersioningTemporal Validation

Models validated with honest forward-chaining split.

03

Prediction API

A robust FastAPI backend retrieves the latest registered model artifact and returns 72-hour predictions.

GET /predictLatency < 85msFastAPI + CORS

Serves live coefficients and predicted values instantly.

04

Visual Analytics

A modern analytics interface turns complex model outputs into decision-ready predictions.

Recharts AreaSWR CachingSymmetry Layout

Interactive 72-hour trend and model ranking telemetry.

Ready for review

View live predictions, compare models, and inspect pipeline evidence.

Launch the live dashboard