The Road That Came With Proof
Research
AI in the physical world has to show its evidence.
We study how machines sense roads, mines and water networks, and how that sensing becomes decisions someone can check. We publish the method, the numbers and the limits, so others can test the work.
phyOS Technical Report
Logs versus video
Same road, same model. Pothole orders ran from 0 to 100 on video and 122 to 128 on a detection log.
Read report →
White Paper
Physics AI vs Computer Vision
Why telematics signal verification differs from visual detection, and why infrastructure platforms need both.
Read paper →Selected work from Cognecto
-
Read report →Structured Detection Logs Versus Raw Video as Language Model Input for Road Maintenance Work Orders: A Single-Road StudyOne 682 m road survey, given to one model in two forms. The log gave steadier quantities and located orders with traceable evidence, at about a third of the input.
Cognecto, phyOS team·Nov 2026Technical Report -
Read paper →Physics AI vs Computer Vision in heavy industryWhy telematics signal verification is fundamentally different from visual detection, and why every infrastructure platform needs both.
Cognecto Team·Jan 2026White Paper -
Read paper →BOQ verification at scale: a knowledge graph architectureHow Cortex automates BOQ line-item verification against verified field events. A reference architecture for PMGSY, NHAI and EPC programmes.
Cognecto Team·Dec 2025White Paper -
Read paper →Smart water metering on LoRaWAN: an engineering deep diveNetwork design, gateway placement, message scheduling and battery optimisation for 15-year AMR meter deployments at urban utility scale.
Cognecto Team·Nov 2025White Paper
Core research areas
1
Evidence for language models
What a language model should read to make a field decision. We compare inputs on the same survey and check every claim back to the row or frame behind it.
Logs versus video →2
Road condition sensing
Finding and locating road defects from ordinary survey video, then hand-checking the detections to measure how far they can be trusted.
Logs versus video →3
Physics-based sensing
Reading machines through their signals as well as their pictures, and working out where telematics and computer vision each hold up in heavy industry.
Physics AI vs Computer Vision →4
Verification and audit
Tying billed quantities to verified field events, so that a line item in a BOQ can be traced to what happened on site.
BOQ verification at scale →5
Water network telemetry
Designing low-power metering networks that last for years in the field: gateway placement, message scheduling and battery life at utility scale.
Smart water metering on LoRaWAN →
Research with us
Working on AI for roads, mines or water networks? We would like to compare notes, data and methods.