Dunvale

Dunvale started as a border crossing intelligence platform and kept growing. It now ingests dozens of live data streams continuously — border wait times at every US port of entry, weather systems, river gauges, radar, aircraft positions, vessel tracking, rail performance, trade volumes, economic indicators, trade policy documents, and weather alerts — all normalized, cross-referenced, and stored in a single database with anomaly detection running on every poll. On top of the ingestion layer sits a full ML prediction pipeline. Gradient boosting models generate wait time forecasts at 2h, 6h, and 24h horizons across 50+ engineered features. Quantile regression produces confidence intervals on every prediction. A flood disruption model trained on decades of river gauge history runs for high-risk corridors. Weather regime classification translates raw atmospheric data into operationally meaningful signals before hitting the models. Local LLM handles AI inference — no cloud dependency. A 3D command center at dunvale.com/map visualizes all of it in real time over North America. Built in about 3 days by one person using Claude Code. Still in beta. A lot more coming.