The Challenge
A hardware and software development company was pioneering the industry’s first commercial utility Ground Penetrating Radar (GPR) system that provided everyone from utility workers to homeowners the ability to safely verify the presence of utilities before they dig.
The founding team brought deep, first-hand experience from the construction and utility sectors—particularly in excavation, where delays and cost overruns often stem from misidentified underground infrastructure. Despite following standard procedures, crews regularly encountered inaccuracies in site assessments, leading to unplanned disruptions and financial liability for contractors. These repeated challenges inspired the development of a more reliable and accessible solution.
The team had already initiated work on a novel ground-penetrating radar (GPR) device designed to detect underground utilities with greater accuracy and portability. They believed that integrating AI and machine learning could not only enhance the product’s performance beyond current market offerings but also lay the groundwork for patentable intellectual property.
The Solution
The objective was to build an end-to-end prototype capable of analyzing scans from a custom GPR device to accurately detect and locate underground objects.
The architecture leveraged AWS Elastic Kubernetes Service (EKS) for workload management, with Python supporting machine learning, and React/Node powering the frontend and backend. Integrated Inertial Measurement Unit (IMU) data provided additional context on ground conditions, critical for accurate GPR analysis.
End-to-End Workflow
- GPR data is collected and saved in a .gpr B-scan format; IMU-based corrections are applied at this stage.
- Scans are automatically ingested, cleaned (e.g., de-wow, contrast enhanced), and converted to a model-readable format.
- Images are displayed in the UI, where users can trigger model inference locally on their tablet.
- Inference results appear as points or bounding boxes with confidence scores; users can filter results accordingly.
- Scan depth is estimated based on user-specified substrate type (e.g., sand, clay), enabling more accurate interpretation.
How It Works
“This technology is a significant leap forward from what’s currently available—and the best part is that it's built for the people actually doing the work in the field.”
The product is designed as a lightweight, handheld device (approximately 3-4 pounds) featuring a tablet interface connected via Bluetooth to a skid-mounted radar sensor. Earlier concepts explored integrating the system into excavation machinery, but technical limitations led to the current standalone design.
Intended for use by on-site operators such as backhoe drivers or foremen, the device prioritizes durability and ease of use, with an expected learning curve of just five minutes. It uses GPS to map the scan area with gridlines, guiding the operator through a push-and-pull scanning motion. The radar emits pulses up to 5 feet deep, detecting underground objects based on the returned signals. Scan time for a 10’ x 10’ area is about three minutes, with a visual map generated within 30 seconds of completion.
The model, trained to identify characteristic patterns like parabolas in radar reflections, ingests both GPR and IMU data. Penetration depth adjusts dynamically based on substrate type (e.g., sand vs. clay), ensuring consistent results across varied environments. The system includes Bluetooth-enabled, cloud-deployed software with built-in security, while still supporting offline operation for remote worksites.
Tech Stack Details
Full-stack cloud-based application working alongside the existing environment:
- Cloud: Amazon Web Services - EKS
- ML: Python
- Frontend Language: React
- Backend Language: Node
Experience Working with Tribe
A member of the team shared, “Tribe’s ability to quickly and accurately formulate a team of experts based on the client’s needs is astounding.”
With a lean team of around ten people—including a mix of retained staff and consultants—the organization saw partnering with Tribe as the most effective way to scale their technical efforts with precision. The collaboration enabled them to implement state-of-the-art machine learning and AI capabilities from the outset, without the need to hire full-time specialists.
“Tribe’s commitment to meeting objectives and deadlines was apparent and went a long way in earning our trust for the long haul,” the team member added.
Tribe Team Members
- Faheem: Product Lead
- Nathan: AI Advisor
- Marat: Technical Lead
- Orges: Engagement Manager
Impact & The Future
“Using AI to interpret radar data—and to be able to do it almost instantaneously on the job site—is a complete game changer,” said a member of the team. By the end of the initial engagement, Tribe had delivered a highly accurate (~90%+) prototype capable of analyzing scans from a custom GPR device to detect and locate underground infrastructure with confidence. This milestone not only validated the product’s core hypothesis but also attracted investor interest and led to new funding, propelling the product into production.
The prototype incorporated advanced preprocessing methods—such as normalization, denoising, and data translation—which contributed to a successful patent application. “This engagement has demonstrated Tribe’s ability to deploy state-of-the-art AI in an industrial field application, which has earned them an ongoing spot as our technology partner as we move into field testing and production,” noted the same team member.
Looking ahead, the second phase will focus on an enhanced tablet-based user interface and a management portal, allowing job definitions to be pre-set by the home office and accessed seamlessly in the field. Field testing will span several months and include data collection from multiple large contractors, as well as internal testing using a controlled sand bed environment. Findings from this phase will guide further product enhancements.
Pricing and packaging strategies are still being finalized, with both leased and owned models under consideration, along with software licensing options. A soft launch is anticipated for the second half of 2025, with a price point designed to serve both professionals and advanced DIY users. “When contractors are already asking when they can get their hands on the device, we know we’ve applied AI in a way that’s truly impactful to the construction industry,” the team member added.
