A Carnegie Mellon-trained roboticist who learned surveying in a coal mine and mapped uranium deposits inside decommissioning pipes is now building what he calls a new category of software — one that sits at the intersection of CAD, GIS, and the physical world.
Alex Baikovitz did not set out to build CAD software. He set out to build robots. The ones he worked on early in his career were not the kind that show up at trade shows: they went into contaminated nuclear tunnels in Hanford, Washington, and inside uranium-laced process piping at Department of Energy decommissioning sites in Portsmouth and Paducah. The work was precise, constrained, and deeply spatial — not unlike surveying, as it turned out. That realization, and the winding path from Carnegie Mellon’s robotics program through the DOE’s brownfield remediation world to a land surveying firm’s overheating back-office computer in Boston, is the origin story of Mach9 and its Digital Surveyor platform.
Baikovitz co-founded Mach9 in 2021 and serves as CEO. Digital Surveyor 2, the company’s second-generation web-based feature extraction and CAD platform, launched publicly this month following a preview at Geo Week in Denver. The platform is built for surveying and engineering firms, as well as the geospatial organizations within large infrastructure owners — DOTs, utility companies, and their contractors — that are producing point cloud-derived deliverables at project scale.
The Robot That Taught Him to Survey
Baikovitz’s introduction to geospatial came through Red Whitaker, the Carnegie Mellon roboticist whose credits include some of the first robots to clean up Three Mile Island and early self-driving car development. Whitaker was Baikovitz’s advisor, and the work they did together for the DOE’s Office of Environmental Management was, in retrospect, a crash course in the fundamentals of spatial measurement.
The first project was a borehole inspection system for a contaminated storage tunnel at Hanford — an inverted periscope-style robot that descended into the opening, spun a single-line laser scanner, and generated a 3D model of the tunnel’s interior. RGB and thermal imagery were overlaid into a single model. The application was structural: mapping beam deformation in a wood-timbered tunnel to assess its survivability. “That was a really awesome introduction to the space,” Baikovitz said, “back in 2016, 2017, where basically it was a miniaturized terrestrial laser scanner that was providing insight into an environment that we didn’t have much exposure to.”
A follow-on project went further into the physical infrastructure of the nuclear weapons complex — a robot inserted into the gaseous diffusion process piping at facilities in Portsmouth, Ohio, and Paducah, Kentucky, tasked with localizing itself inside the pipe and quantifying how much uranium had deposited on the interior walls. The output was an inventory for decommissioning contractors. “It was around quantities and measurement of the physical world in these very challenging, complex situations,” Baikovitz said. None of it was called surveying. All of it was.
Close collaboration with Red’s brother Chuck, a surveyor working in Pittsburgh’s coal mines, grounded that technical experience in the professional discipline. By the time Baikovitz left that world for other projects — including a period at SpaceX — he had concluded that the robotics problems he had been solving were, at their core, hardcore surveying and mapping problems. That understanding is what eventually produced Mach9.
The Boston Back Office and the Pivot
Mach9 launched as a hardware company, building reality capture devices — a vehicle-mounted system called the Cube — for collecting laser scan data. Within the first year, the company found itself at a crossroads that a lot of hardware startups in the geospatial space have faced: the hardware market is competitive, well-capitalized, and dominated by established manufacturers.
The pivot came from watching what happened after the data was collected. At a land surveying firm in Boston — one of Mach9’s early hardware customers — Baikovitz and his team watched the firm’s head of geospatial take their point cloud, drop it into AutoCAD, and start manually drafting features one by one. Tree. Sidewalk. Curb. The workstation was struggling under the file size. The process was slow, unnatural, and clearly not scaling. “The computer is overheating because there’s this giant point cloud in here,” Baikovitz recalled. “You’re trying to take a bunch of cuts and cross-sections, and it just wasn’t clicking.”
The same pattern showed up again with Michael Baker International, where a VP of geospatial was direct about the problem. He had no interest in Mach9’s hardware — he already owned four mobile laser scanners from Optech. What he needed was a more automated production solution that would let his team meet tight turnarounds, compete for larger projects, and deliver better outputs to the DOTs that were his primary clients. “If you had a more automated solution, that’s going to really help us be more productive, be more competitive,” they told Baikovitz.
The message was consistent enough that by late 2022, Mach9 had committed to building Digital Surveyor — a hardware-agnostic, cloud-based platform for feature extraction and geospatial production. The pivot was demand-driven, not technology-driven: the market told Mach9 what to build before the company had decided to build it.
Digital Surveyor 2: AI That Learns the Job
Digital Surveyor 2 is a web-based CAD environment designed specifically for point cloud workflows. The architecture is deliberately different from the desktop-based production tools that most surveying and engineering firms currently use: processing moves to the cloud, collaboration is native to the platform, and the automation layer is embedded in the workflow rather than bolted on.
The centerpiece of the second-generation release is what Mach9 calls the suggestions workflow. As a user extracts features from a point cloud, the system observes the decisions being made and begins predicting the next move — effectively learning the extraction schema in real time and surfacing AI-generated completions that the user can accept with a tab keystroke or override. QA/QC happens inline, at the pace of production, rather than as a separate downstream review step.
“It’s not pushing AI onto the team,” Baikovitz said. “It’s providing an experience that’s very similar to how people would want to be doing extraction, but really speeding it up.” The design intent is that automation handles the low-judgment, repetitive extraction work — tracing curb lines, populating utility poles, extending linear features across consistent geometry — while the professional’s attention is reserved for decisions that require domain knowledge and contextual judgment.
The other axis of Digital Surveyor 2’s design philosophy is onboarding speed. Traditional CAD software has long ramp times; learning Microstation or Bentley’s suite to production proficiency takes months. Baikovitz’s team set an explicit target of getting a new user to productive output in hours, not weeks. Langan Engineering tested that claim directly: after seeing Digital Surveyor 2 at Geo Week, the firm put a production technician on a 30-minute onboarding call and had a completed first project delivered the same day. “Being able to really get someone right off the ground and hit the ground running, and have meaningfully fast turnaround without a substantial amount of training — I think that goes tremendously far,” Baikovitz said.
The platform is cloud-only, consumption-based — priced per mile or per area with rates tied to commitment levels. Mach9 works with government agencies and large infrastructure owners for whom data security is a non-negotiable procurement requirement. On-premises installation is not offered.
A New Category, or a New Layer?
Baikovitz argues that Digital Surveyor is filling a gap between the reality capture world and the CAD/GIS world that none of the incumbents have adequately addressed.
His framing draws a historical arc: in the 1980s and 1990s, CAD digitized the drafting table. GIS then digitized the paper map. Today, the convergence of abundant, affordable LiDAR data, cloud computing, and AI creates a third inflection — software that can take the physical world as input and produce the design-grade or analysis-grade digital maps that infrastructure owners need. “Data collection isn’t the problem,” Baikovitz said. “The problem is, how do I turn that data into usable information and digital maps?”
The interoperability commitment is real: Digital Surveyor outputs are designed to land cleanly in ArcGIS, AutoCAD, and the downstream systems where engineering and asset management work happens. Mach9 positions itself as a production layer upstream of those platforms, not a replacement for them.
The physical AI framing Baikovitz uses is worth noting. As AI systems increasingly need to reason about the physical world — infrastructure layout, asset condition, network topology — the structured, georeferenced feature data that Digital Surveyor produces becomes more valuable as a training and inference substrate. Mach9 is building, in his framing, a system that understands the physical world and encodes that understanding in formats that both humans and downstream AI systems can use.
Transportation, Utilities, and the Labor Problem
Mach9’s customer base has been built primarily in transportation — DOT-funded linear projects, highway surveying, and network-scale asset mapping. The company holds an active contract with the Maryland State Highway Administration for safety-critical asset mapping across the state road network, and describes similar work underway in other states.
The near-term growth focus is utilities and telecommunications. The drivers are structural: BEAD funding is pushing fiber deployment at a scale the industry’s existing production capacity cannot easily absorb. Data center construction is accelerating. Electric grid hardening and expansion — driven by both aging infrastructure and new load from electrification — is generating mapping and design work faster than firms can hire and train the CAD technicians to execute it.
That last point is the commercial lever Mach9 is pulling. The surveying and engineering industry has a well-documented talent supply problem: skilled CAD technicians and surveyors are difficult to find and retain. Digital Surveyor is, among other things, a productivity multiplier for the workforce that does exist — enabling teams to scale output without proportionally scaling headcount.
On the electric utility side specifically, Mach9 is working with major investor-owned utilities on automating the extraction of pole and tower locations and specific asset measurements for delivery into PLS-CADD, the transmission and distribution simulation platform that utilities depend on for system modeling. That workflow — from aerial or mobile LiDAR capture to PLS-CADD-ready deliverables — has historically required labor-intensive manual extraction. Mach9’s automation targets that bottleneck directly. The telecom parallel is similar: tower and pole location data, underground conduit mapping, and network documentation at the scale that BEAD-funded deployment demands.
Rail is another vertical where Baikovitz sees natural extension of the same core capability, with the same types of linear projects and asset inventory needs.
The through line from Hanford’s contaminated tunnels to a cloud-based CAD platform for highway surveyors is not obvious, but Baikovitz draws it without strain. The measurement problems were always the same; only the environment changed. What Mach9 is building now is, in his telling, the logical endpoint of that trajectory — software that takes the physical world as input and encodes it in forms that both human professionals and downstream AI systems can reason about. Whether that constitutes a new category or a new layer in an existing stack may be a question the market answers before the industry finishes debating it. The production bottleneck is real, the workforce constraints are structural, and the data is already there. Mach9’s argument is that the missing piece was always the software in between.