Canada is preparing to spend hundreds of billions of dollars on transit, energy, mining, and other major infrastructure projects. But some of the biggest threats to delivering that work on time and on budget lie beneath the surface.
Unlike above-ground construction, where drones and satellite imagery have made detailed data easier to collect, tunnelling projects still rely heavily on limited drill samples, manually prepared inspection notes, technical reports, and models built on assumptions about what lies between known data points.
Vancouver-based GroundedAI wants to give project teams a clearer picture.
The company’s platform allows engineers to capture detailed 3D scans underground and compare actual conditions against project plans within the same construction shift. By identifying unexpected ground conditions earlier, GroundedAI aims to help teams make faster decisions, improve safety, document change orders, and prevent manageable problems from becoming costly delays.
The startup recently raised $2 million in a round led by Stand Up Ventures, with participation from SOSV, Accelia, BoxOne, The51, LOI VC, and other investors. GroundedAI is using the capital to expand into the tunnelling sector and grow its Vancouver-based team, building on early work with engineering firms including Hatch and Delve Underground.
Techcouver spoke with GroundedAI Co-Founder and CEO Shelby Yee about why underground construction has been slow to adopt digital technology, how artificial intelligence can make infrastructure projects more predictable, and the role she hopes the company will play in Canada’s coming construction boom.
Canada is entering what many are calling a once-in-a-generation infrastructure build cycle. Why do you believe the biggest risks to delivering these projects are actually underground, where most people never see them?
SY: Above ground, the cost of getting good data has decreased significantly. More accessible drones and satellites mean surface projects can gather detailed information before anyone breaks ground, so more uncertainty and risk can be reduced upfront.
Underground hasn’t seen this benefit on the same scale. Drilling every hundred metres along a route to see what’s underground isn’t realistic, especially in a city. So projects rely on models, instead, that extrapolate between a handful of data points. The output is a geotechnical baseline report, which is full of assumptions and remains unstandardized.
Bent Flyvbjerg’s research on infrastructure costs shows that when projects go wrong, they really go wrong. On average, tunnel projects go over budget by 34%. And as construction goes on without issues resolved, the cost of resolution and delays only continues to increase.
Your platform aims to give project teams “same-shift” visibility into changing ground conditions. Can you walk us through what happens today on a typical tunnelling project versus what happens when GroundedAI is in place?
SY: Take a daily engineering meeting, where teams make calls on ground support and sign off on what’s been built. Today, someone walks the tunnel, maps what they saw by hand, takes a few photos, and the team reviews it the next day. That meeting is already time compressed, since yesterday’s data barely made it in. Baselines sit in technical PDFs or complex 3D models, so comparing current conditions to plan in any reasonable time is nearly impossible. Bigger picture risks get missed due to the lack of time and poor data fidelity and accessibility.
With GroundedAI, that same engineer captures a detailed 3D scan underground in the time it used to take to walk and take notes. Risks can be identified in that 3D picture the same day. By the time the engineering meeting happens, the team can review actual conditions against the plan together and sign off with greater confidence. Risks surface earlier and get relayed to management to more effectively manage cost escalations and change orders.
We hear a lot about AI in knowledge work, but GroundedAI applies it to physical infrastructure. What does AI actually do in your platform, and where does it deliver the biggest value for engineers and project owners?
SY: Today, our platform captures structured data underground that can be compared against the plan. That alone changes the pace of decision making. Variances get flagged at a speed that keeps up with the construction cycle instead of lagging behind it.
Where we’re headed with AI is the next layer on top of that. We’re building toward a system that learns what normal ground behaviour looks like for a given rock type or ground class from the data we’re collecting, so it can flag the variances that are significant instead of treating every deviation the same. That’s a harder problem than direct comparison, and it’s why we’re building the dataset now that makes it possible.
The value either way shows up in the moments where speed and clarity change the outcome: catching a variance in time to adjust ground support before it becomes a safety or maintenance issue, or documenting a difference clearly enough that a change order can be filed and accepted with confidence instead of a reconstructed memory.
You’ve said many tunnelling projects still rely on paper logs and spreadsheets. Why has underground construction been slower to adopt digital tools than other industries, and do you think we’re reaching a tipping point?
SY: The environment itself is hard to build technology for. You often don’t know which way is north. Water’s dripping on you. It’s dusty. You’ve got gloves on, safety glasses, steel-toed boots, ear protection, and only a small light from your hard hat.
Then there’s the pace. An engineer might get fifteen minutes to inspect a section before an impatient crew with heavy equipment needs to get back in. A manager might have twenty-four to forty-eight hours to identify an issue or defend a claim, and often doesn’t have the data to do it properly.
It takes a specific kind of team to build for that. My co-founder Stuart and I have spent a decade working underground with these constraints, and we’ve built a team that’s tech forward enough to work in an AI native environment.
What sets us apart from a lot of the consultancies in this space is that we also know how to commercialize. We’ve brought a product to market, listened to problems on site, made sure people know the solution exists, and taught an industry how to use something new.
The tipping point is a few things converging. On the macro level, project costs are climbing and the country’s need to build new infrastructure fast keeps growing. On the adoption side, a new generation is moving into decision-making roles, and they’re saying yes to new technology and pushing it into new projects.
You’ve raised $2 million from investors including Stand Up Ventures at a moment when governments are committing unprecedented infrastructure spending. What milestones are you hoping this funding helps you achieve over the next 18 to 24 months?
SY: We want to have helped defend a cost overrun claim that’s saved Canadian taxpayers tens or hundreds of millions of dollars. We want to keep growing a Vancouver based team that attracts and keeps strong engineering and business talent. And we want to keep expanding into markets facing the same problem: the US, Australia, and Europe.
You’ve already attracted customers including Hatch and Delve Underground while the company is still early in its journey. What have those early deployments taught you about how engineering firms are embracing AI and digital workflows?
Consultants juggle a lot of software day to day, so they tend to be fast learners with new tools. A lot of the engineers we work with are early to mid-career. This gives them a tool that helps them do their job well, and something they can share, in the company and in their professional circles. We’ve watched that happen at the firms we work with.
It’s a competitive industry, and firms want the edge that comes from adopting new technology early. What we really enable is for these consulting companies to own more of their own data. That data becomes an asset. It helps them do better work on the current project, and it helps them win the next inspection or maintenance contract on that job, or a similar one in the region.
Looking ahead five years, if Canada succeeds in building the transit systems, mines, and energy infrastructure it has planned, what role do you hope GroundedAI will have played in making those projects more predictable, affordable, and successful?
SY: Infrastructure doesn’t really work on a five-year clock. We build houses and buildings much faster than we build large infrastructure, so let’s look ten to twenty years out instead.
I want a fairer and faster construction process. Cities should have the confidence to stay on budget, backed by data instead of a model’s interpolation. Contractors should be able to handle a cost overrun with a straightforward change order instead of a drawn-out dispute.
Success looks like mines in Ontario, BC, and Saskatchewan and other provinces coming online. Cities with refreshed drinking water and wastewater systems. Pipelines moving energy safely to market. Transit hubs in Vancouver, Toronto, and Montreal delivered on time. We’ve reduced the uncertainty enough on all of it to save billions in cost overruns, and that’s the part I want GroundedAI to have had a hand in.
The post How GroundedAI Is Bringing AI Below the Surface appeared first on Techcouver.com.
How GroundedAI Is Bringing AI Below the Surface was first posted on July 16, 2026 at 7:00 am.
©2022 “Techcouver.com“. Use of this feed is for personal non-commercial use only. If you are not reading this article in your feed reader, then the site is guilty of copyright infringement. Please contact me at rob@kitsilano.ca
