Every tunnel crew has lived some version of this night. The boring machine grinds to a stop at 2 a.m., someone climbs down with a headlamp to see what happened, and within the hour there are three competing theories: bad rock nobody flagged, groundwater nobody predicted, or the crowd favorite, "the geotech report was wrong." Nobody photographed the face before the ground changed underneath it. Nobody can prove, six months later when the claims consultants show up in suits, what was actually down there at 1:45 a.m. versus what someone remembers being down there. The argument becomes two parties litigating a memory.

That argument, over what the ground actually looked like at the exact moment the work happened, is the quiet, expensive engine behind a huge share of underground construction disputes. A Vancouver startup is betting it can replace the memory with a recording.

GroundedAI, a Vancouver based startup, was profiled this week by the Journal of Commerce on the technology it is now pushing hard into the tunnelling sector: a two part system called Lithos and Strata. Lithos is an iPad based LiDAR application that field crews use to capture underground conditions during a shift, scanning the tunnel face, the rock, the water intrusion, whatever is actually there. Strata is the web platform on the other end that georeferences and stitches that data together, then flags the differences between what crews are seeing in real time, what the original geotechnical model predicted, and what the project budget assumed. Put together, the two tools are meant to give owners and contractors a continuously updated, timestamped 3D record of subsurface conditions rather than a memory and a stack of daily reports written after the fact.

GroundedAI's own numbers on why this matters are stark: tunnel projects go over budget by an average of 34 percent, and unaccounted for ground risk is a core driver of that overrun. Founder and CEO Shelby Yee, a geo-engineer with a decade of experience in underground data capture, frames the pitch around exactly the dispute scenario above: instead of two sides arguing about what the rock face looked like months after the fact, GroundedAI's data is meant to give owners and contractors "a definitive 3D picture of what happened at a specific time," turning a disagreement into something closer to a lookup. The company has raised 2 million dollars in a round led by Stand Up Ventures, with SOSV, Accelia, BoxOne, The51, and LOI VC also participating, and plans to use it to grow its team from seven people to eleven while it pushes into transit, energy, water, pipeline, and critical mineral infrastructure work across North America, much of which involves tunnelling or other underground construction. GroundedAI already has consulting relationships with engineering firms including Hatch and Schnabel Engineering, which is a meaningfully different starting position than a startup trying to sell a brand new workflow into an industry that has never used one.

The stakes here are not hypothetical. A four mile tunnel project under Hartford and West Hartford, Connecticut, run by the Metropolitan District Commission, spent years in a differing site conditions fight with its contractor, Kenny Obayashi joint venture, after groundwater far in excess of what the geotechnical reports predicted flooded the dig and blew out the schedule. The two sides eventually settled by amending the pending differing site condition change orders to roughly 51 million dollars, on a project that had been bid at 279.4 million dollars in the first place. That is one project, one dispute, and it still moved nearly a fifth of the original contract value. Multiply that by every tunnel, shaft, and deep foundation job currently being bid against a geotechnical report nobody can fully trust yet, and the appeal of a system that produces a timestamped record instead of a competing narrative becomes obvious.

What makes this worth watching past the funding headline is the direction it points the industry. Differing site conditions claims have always been resolved through negotiation, expert testimony, and whoever kept better paperwork, which is a process built for lawyers, not for the crew that actually has to keep digging. If tools like GroundedAI's actually take hold, the effect is not just fewer disputes after the fact. It changes what a contractor can responsibly price into a bid before the work even starts, because the risk that used to live in "we genuinely don't know what's down there until we hit it" starts to shrink into something closer to a known, trackable variable. That is a small but real shift in how underground risk gets priced, and it is the kind of shift that tends to matter more to the people writing the proposal than to the people reading about the funding round.

Sources