Designed and CAD modeled pneumatic actuators and limit switch boxes, and created detailed technical drawings in SolidWorks for customer orders and product manuals. Also produced instructional videos and animations for training courses and the company website, simplifying processes like driveshaft disassembly for clients and internal teams.

Gained hands-on manufacturing experience producing pneumatic actuators, and developed technical drawings in SolidWorks to support communication between the manufacturing team and customers.

Completed a Master's Thesis titled "Zero-Shot Model-Based Object Perception in Complex Scenes Using CAD Templates," developing a zero-shot 2D object detection framework using CAD templates instead of labeled training data. Built the Clutter6D synthetic dataset using BlenderProc for sim-to-real transfer. Submitted to ECCV 2026.

Worked as a working student at the DLR Institute of Robotics and Mechatronics, Department of Perception and Cognition, developing a model-based segmentation pipeline by integrating Meta's Segment Anything Model (SAM) with a correspondence model.

Advised on an automation project for a bipedal robot equipped with dual LiDAR sensors.

Led an 8-student team building a ROS2-based quadruped autonomy stack for the DeepRobotics Lynx M20 Pro.

Continued on RoboTUM's Humanoid Team as scope expanded beyond simulation and reinforcement learning into perception and broader control system development.

Joined RoboTUM's Humanoid Team, working toward an energy-efficient, tendon-driven bipedal robot. Ran simulations in NVIDIA Isaac Sim and contributed to the reinforcement learning control pipeline, including a Central Pattern Generator (CPG) in JAX for rhythmic control signals.