06 / Logistics
Weather-Aware ETA Prediction Engine for Indian Logistics
We built an ETA engine that knows Indian roads flood in the monsoon, so instead of one optimistic arrival time it gives a realistic range and tells you why a truck will be late.
- Built by
- 360 Labs Product, Built on a Global Freight Forwarder's Fleet Data
- Stack & approach
- LogisticsETA PredictionMachine LearningOn-Prem DGXPublished Research
The problem
Truck ETAs in India collapse in the monsoon: worst-case journey times differ by 457% between dry and rainy days, and the incumbent telematics estimate assumed a flat ~17 km/h regardless of weather.
Solution & architecture
An ETA engine that predicts a range of arrival times (P10-P90) broken down by road, stops, facilities and weather. It learns real stop behaviour from 8,846 fleet stops and measures waterlogging risk from 248,883 GPS legs joined to rainfall, finding the Mumbai-Thane-Bhiwandi flood corridor on its own. Trained and served on 360 Labs' own NVIDIA DGX.
Why it is built this way
Runs on owned DGX
No per-query cloud cost; Rs. 0/month to run
Fleet data stays private
Client GPS never leaves the environment
Research-grade honesty
Error bars, not a single false-precision number
