Closing India's Infrastructure Inspection Gap with Brahmaksh

Every steel asset India builds — a warship under construction at a Kolkata shipyard, roads, a fuel tank, a defense platform in the field — is protected by a coating a few hundred microns thick. That coating is the only thing standing between the steel and the slow chemistry of corrosion. When it holds, the asset lasts decades. When it fails, and it always begins to fail somewhere first, oxidation starts eating load-bearing steel long before anyone sees a problem. On a bridge that means creeping structural risk. On a naval vessel it means a readiness problem measured in dry-dock days and crores.

The entire discipline of coating inspection exists to catch that failure at the earliest possible stage — a blister, a hairline crack, the first bloom of rust — while a repair is still a touch-up and not a structural job. Catch it early and you spend a little. Miss it and you spend a great deal, later, under worse conditions.

Why Coating Inspection Is So Hard Today

For all its consequence, coating inspection is still done largely by eye. A skilled surveyor grades what they can reach against standards like ISO 4628 — rust grade, degree of blistering, cracking, flaking — and writes it down. The judgement lives in the surveyor’s head, which means two competent people can grade the same surface differently, and the record is only as good as the notes and photos taken by hand afterward.

Brahmaksh’s Infrastructure Platform for Automated Inspection

The approach is the one already running on road corridors, adapted to a steel surface. An edge device — a camera paired with an onboard NVIDIA Jetson — moves across the surface, whether handheld, on a pole, or mounted to a crawler. On the device itself, a detection model flags candidate defects frame by frame: rust bloom, blistering, coating breakdown, cracking, edge delamination. This runs entirely on-device, in milliseconds, with no network — which matters, because you cannot assume a signal inside a steel ballast tank.

When a finding crosses a severity threshold, a small structured event — not video — is sent to the cloud, where a vision-language model grades the severity, estimates the extent, and places the finding into a prioritised repair list a coatings engineer can act on. The device sees; the cloud reasons. The only thing that ever leaves the asset is structured findings and low-resolution thumbnails, processed on Indian, MeitY-empanelled infrastructure. The raw imagery stays with the asset.

Where This Stands — and Where It’s Going

We’ll be straight about status. The engine that finds, classifies, and costs defects is proven on road corridors; pointing it at coatings is a matter of training the detector on coating-failure imagery and tuning the grading rubric to coating standards — the architecture doesn’t change. We’re building toward field validation on real steel assets, and we’re clear-eyed that a controlled demonstration is not a deployed system.

The goal is worth building toward: coating inspection that’s faster, safer for the people who do it, and consistent enough to trust — so the first bloom of rust gets caught while it’s still a touch-up, not a structural job.