At ambic, I built a system that turns room photos into a measured 3D model in about 10 minutes, within about 3% of tape measurements. I judge 3D quality on real data, not just training views.
At Scantinel Photonics GmbH, I worked on perception software for FMCW LiDAR, including point-cloud processing and LiDAR-camera calibration and fusion. I also recorded urban driving datasets with a car-mounted rig.
At Huber Automotive AG, I trained object detection and segmentation models and prepared and checked training data. Earlier, at Deftbox Solutions, I built NLP pipelines and REST APIs for client projects.
For my M.Sc. thesis with Scantinel, I trained models to predict future 3D occupancy from LiDAR and camera without hand labels. I evaluated them using IoU, AP, Brier score, and calibration error.

