TECHNOLOGY
A pipeline of three specialized systems, each doing one thing well, chained together so compliance happens automatically.
STAGE 01
A vision model trained on stormwater BMPs classifies what's in every photo an inspector or site manager captures: silt fence, inlet protection, stabilized construction entrances, concrete washouts, sediment tracking, and failure states like undercutting or overtopping. Each detection carries a confidence score and bounding region.
STAGE 02
The rain-event engine polls NOAA — on a 15-minute cadence for every active site. When a forecast or rainfall crosses a permit threshold, it opens a rain event, starts the inspection clock, and notifies the right people before the deadline is even close.
STAGE 03
A language model assembles detections, rainfall records, and site history into an inspection-ready narrative: findings and corrective actions with deadlines — formatted the way regulators expect.
Rainfall comes from NOAA, not a consumer weather API. When a regulator asks where a number came from, you have a defensible answer.
The engine removes busywork, not accountability.
Each finding links back to its photo, timestamp, and rainfall record — an audit trail built in from the start.