AI-powered analysis of civil engineering services in fiber-optic network expansion

Project facts

• Client: OXG Glasfaser GmbH, a provider of telecommunications infrastructure and fiber-optic network expansion
• Project goal: Automate the partial measurement review for civil engineering works using artificial intelligence
• Team: Up to four TNG experts
• Implementation: Proof of concept from May 2025, implementation from July 2025
• Successes: Objective and complete review of partial civil-engineering measurements regarding route lengths and surface classifications instead of spot checks

The starting point

OXG plans to connect around seven million households to a fiber-optic network in the coming years. To do this, fiber-optic cables have to be laid over many tens of thousands of kilometers.

With the start of the civil engineering work, the review of the services provided by the construction service providers increasingly came into focus. Billing is based, among other things, on the length of the route laid and the type of surface, for example asphalt or paving. Until now, OXG employees checked the distances stated in the partial measurements on foot using a measuring wheel. However, such a process does not scale to large-scale expansion.

Our goal

OXG initially commissioned us to evaluate whether the partial measurement review can be automated with the help of artificial intelligence. The goal was to document the actual construction services provided in a comprehensible and complete manner while significantly reducing the manual review effort.

Our approach

In a first step, we demonstrated technical feasibility in a proof of concept. On this basis, we were able to begin developing a productive solution two months later.

To document the civil engineering work, the construction service providers create digital scans of the routes during construction. This image data forms the basis of our solution. For the automatic recognition and classification of different road and sidewalk surfaces, we trained a neural network for image segmentation.

For the creation of the training data, we use LabelMe as an annotation tool. We implement the training pipeline in Python with PyTorch and evaluate the training progress with TensorBoard.

In addition, we developed a web application that visualizes the scans in a map view. For this purpose, we process the geocoordinates with GDAL and provide the image data as Cloud Optimized GeoTIFFs (COG). Construction service providers can:

• View the course of the excavated route
• Adopt the automatically determined route lengths
• Manually correct surface classifications if required, for example if scans are missing, shifted, or obscured

All adjustments flow directly into the basis for later invoice verification.

Technologically, we use Python in the backend as well as TypeScript, React, and OpenLayers in the frontend. The solution is operated in AWS.

The result

With the new solution, construction service providers can view and adopt the determined route courses and route lengths directly when submitting partial measurements. OXG thus has a transparent and easily comprehensible basis for reviewing the services submitted.

A key added value lies in the scalability of the process: instead of complex on-site controls or spot-check reviews, a continuous assessment of the billed services is now possible.

Even after the successful go-live in May 2026, we are continuing to develop the application together with OXG. The focus is on expanding the drawing tools in the web application so that missing route sections can be added directly, as well as evaluating the excavation depth and width for even more accurate billing.

As part of Big Techday 26, our customer gave a talk about this project together with us. You can watch the full recording here.