| Contract type: | Service contract |
|---|---|
| Client: | GEO DATA GmbH |
| Task: | AI consulting, development, and training of artificial neural networks, as well as an end-to-end AI prediction tool for segmenting construction trenches and trench centre lines using Python, Azure Machine Learning, Azure Data Factory, and Azure DevOps. Development of a snapping tool and setup of a test scenario with Azure Container Apps. |
| Time period: | 05/2024 to 11/2025 |
Project Overview
In addition to my consulting work in AI, I developed and trained neural networks using the Python library PyTorch. Specifically, convolutional neural networks (CNNs) were used to segment construction trenches, and the resulting segmentations were used to derive the centre lines of the trenches. For this purpose, in a downstream processing step, I implemented various geometry functions and machine learning algorithms in several Python scripts that calculate the centre lines and generate the final result.
For visualisation, inference (prediction), and dissemination of results, I built end-to-end pipelines in Python using Azure Machine Learning and Azure Data Factory, and applied MLOps practices aligned with the requirements of the EU AI Regulation (AI Act). By integrating and dynamically utilising computing clusters with NVIDIA GPUs, the pipelines can be scaled as needed.
Complementing the overall design of the AI architecture, I developed a Python snapping tool for line output that includes geometry functions and a machine learning algorithm (LightGBM). In addition, I deployed several Docker containers in the Azure cloud for a test scenario and integrated them into Azure Container Apps. Standardised interfaces between the containers ensure data exchange.
All results, tools, and pipelines are versioned and documented in Azure and Azure DevOps.