Built like a forecasting lab.

SkyPulse isn't another weather app. It's a small intelligence pipeline running entirely on the edge.

Triple-source aggregation

Every reading queries OpenWeatherMap, WeatherAPI, and Tomorrow.io concurrently. Disparate JSON shapes are normalized into one unified schema.

Confidence scoring

A weighted blend of inter-provider variance (temperature, humidity, wind) yields a 0–100% confidence score. More agreement → higher confidence.

AI next-hour forecast

Aggregated state plus temporal features are fed to Lovable AI. The model returns a typed prediction via tool-calling for next-hour temperature, trend, and rationale.

Threshold alerts engine

Cross-validated metrics are evaluated against safety thresholds — extreme heat, severe wind, low visibility, thunderstorms, UV — to surface emergency warnings.

Interactive heatmaps

Leaflet renders global tile layers for temperature, precipitation, clouds, wind, and pressure. Click anywhere to pull a live aggregated reading for that point.

Resilient by default

If a provider degrades, the aggregator continues with the survivors and lowers confidence accordingly. No single point of failure.

Stack

· TanStack Start (React 19, Vite 7)
· Edge server functions (Cloudflare Workers)
· Tailwind v4 design tokens (oklch)
· React-Leaflet 2D mapping
· Lovable AI Gateway (Gemini 2.5 Flash)
· Lovable Cloud secrets management