ML-forecasted fleet fuel delivery

dropt

Proof of concept

A fleet-wide oil delivery scheduler that forecasts per-location fuel consumption and plans a full winter season of deliveries to flatten the peak-day volume — without ever running a tank dry.

The AI angle

A LightGBM forecaster predicts expected and P95 consumption per tank from telemetry, weather, and degree-days; a fleet peak-leveling optimizer plans deliveries inside safety windows, with a hard tank-monitor override guaranteeing zero dry tanks by construction.

45% lower winter peak-day volume vs. naive fill-when-low scheduling
Zero dry tanks — hard safety guarantee, not a target
Forecaster backtested on a season it never trained on (0.68 gal/day MAE)
17/17 automated tests · analytics API + live dashboard

LightGBMPythonFastAPIReact dashboard

Start a build → See more work