Cloudflight

The AI work is genuine engineering, and it is where the client names stop. The Phytec study names its architecture: a custom TensorFlow convolutional network classifies cup type, a segmentation model and a regression model derive volume, part of the training set is generated rather than photographed, and the whole thing runs offline on embedded hardware inside a coffee machine.

The airport baggage system is described as developed, validated, deployed and hosted, with the pipeline set out stage by stage—object detection, instance segmentation, then fingerprinting against a central identification service. No client is named and no accuracy or throughput figure is published.

The five case studies that do name clients are SAP extensions for Brenntag, an employee app for voestalpine, order picking for Schachermayer and a prevention app for AOK Baden-Württemberg, where the page says AI enhancements are still in the pipeline. That is software delivery, not AI delivery.

Best for industrial and public-sector buyers who want a certified European engineering organization and will accept anonymized references on the machine-learning side.