Augmented Reality for Utility Projects with Alec Pestov

This newsletter has been prepared for a recent episode recorded for the Digital Construction Podcast, featuring Alec Pestov, Founder and CEO of vGIS.

Every excavation starts with an act of faith. A locator sprays paint on the ground, files a report, and a crew digs on the strength of markings that will fade within weeks and a document nobody reopens unless something goes badly wrong.

Follow that single trench from the moment the paint goes down to the moment the invoice gets paid, and you pass through every problem vGIS was built to solve. It is a more useful way to understand the product than any feature list, because the value only becomes visible in sequence.

AR becomes useful when it deliver value on real projects.

The problem starts before anyone reaches site

Long before the excavator arrives, the information needed to dig safely already exists. It is simply scattered. Municipal GIS databases hold one version of the truth, the design model holds another, and isolated CAD files and PDF sheets hold the rest.

The first thing vSite does is collapse that spread into a single continuously updated cloud model. Not a static digital twin delivered once at handover, but an aggregation of live sources a crew can reach from a phone. This matters because every downstream benefit, from accuracy to evidence, depends on the field working from one dataset rather than four partial ones.

Fragmented information does not become a field problem until someone is standing in a paddock trying to reconcile it.

The next problem is the crew, and most software gets this wrong

Having consolidated the data, the temptation is to deliver it in an impressive new format. vGIS deliberately does the opposite, because most of the workforce has spent entire careers reading paper sheets and measuring with a wheel, and they are effective that way. Re-educating millions of people is not a strategy.

So the entry point is a PDF, just a geolocated one. You see your own arrow moving across the sheet as you walk. Version comparison isolates only what changed between issues, so a manhole that shifted twenty metres into a protected tree surfaces as a permitting cost of six months and a thousand dollars while there is still time to act. That is clash detection performed with your feet, and it requires no training at all.

Adoption follows familiarity. The fastest route to a digital field is a digital version of what the field already trusts.

Where AR earns the step up

The geolocated PDF has a ceiling, though, and that ceiling is where AR begins to justify itself. One button on the same device switches into augmented reality, and three things become possible that no sheet can deliver.

Grading is the clearest. Whether to cut or fill at a given point normally means a surveyor, a total station and hours or days of measurement. A slider that scans the ground and overlays the design grade answers it in seconds, position by position as you walk. The second is clearance between services that overlap at different depths. The third is simply memory, because a crew that has seen the pipe curve retains it in a way no column of offsets achieves.

None of this works without real accuracy, which is why vGIS supports around forty GPS receivers and holds patents on aligning the overlay to the world. On a recent data centre project, utilities had to land inside ten centimetre tolerances on a site where the control points themselves disagreed. Staking out from the model exposed errors of two and three centimetres that no tape measure from a disputed control point would ever have found.

Visualisation without survey-grade positioning is decoration. The accuracy is the product, and the picture is how it gets delivered.

The line vGIS doesn’t cross

That same commitment to accuracy explains a decision that looks like a gap and is actually a discipline. vGIS does not go inside buildings.

Indoors there is no source of truth to align against. Drywall shifts half a metre and the model never records it. A QR code anchors you for an instant, and then drift accumulates: a two percent error across a ten metre walk leaves you twenty centimetres out. Outdoors, tolerances absorb that, since a sewer pipe a few centimetres off line is fine if the slope holds. Indoors, where wires sit centimetres apart, the same error means someone drives a nail through a live cable while trusting the display. GPS also needs sky.

Back outdoors, the record becomes the real asset

Holding that line is what makes the second half of the platform credible. Once you can trust position to the centimetre, everything a crew observes can be captured as evidence rather than as an anecdote.

This is where vGIS quietly does something the locate services cannot. Dial Before You Dig, 811 and their equivalents are dispatchers. They know which utilities operate in an area and who to call, but they own no data and hold no central repository, the Netherlands registry being the one real exception. vGIS does not compete with them. It piggybacks on the work they trigger: scan the paint, scan the exposed utility.

The resulting record is geo-referenced and timestamped, and it pays off in four directions. It settles arbitration about whether the markings were ever in the right place. It protects the locator whose correct work gets quietly repainted after a strike. It means a rotated crew inherits the site instead of redoing it. And it becomes a deliverable the asset owner can reuse the next time anyone opens that street.

Knowledge held only in paint and in people’s heads is not a record. It is an exposure waiting for an incident.

From evidence to profit, and what comes next

Once that documentation exists in structured form, it stops being a compliance artefact and starts being data. This is where the AI sits, and notably it is never presented as AI.

A typical site loses two to three hours a day retyping delivery dockets into a spreadsheet. Photograph one instead and it is digitised and matched to a cost code automatically. In Germany, regulated trench depth and width once required inspectors to drive to every site and manage perhaps two a day, with trenches reopened two or three times to produce evidence before an invoice cleared. A scan takes roughly forty seconds per thirty metres, returns flagged problem areas in about two minutes, and cannot be faked. Crews press a button and get an answer. Nobody mentions the model.

The next step follows directly from all of it. The same captured documentation is being turned into daily financial tracking, so a project manager knows at the end of each day whether the job made or lost money. The honest constraint on the AR side remains hardware, since centimetre accuracy still means carrying a phone plus a receiver, and AR is only now climbing out of the bottom of its hype cycle.

The durable value of field capture is rarely the capture itself. It is what the data becomes once it reaches someone making a commercial decision.

The trench that started with a can of spray paint ends, in this version, as a dataset the owner still holds in five years. That shift, from a mark that disappears to a record that compounds, is the quiet argument for the whole category.