At Sthenos AI, I own production object-detection and multi-object-tracking pipelines for drones and vehicles, working with infrared and RGB imagery from model development through deployment.
I optimize PyTorch models for real-time GPU inference with TensorRT, including FP16 and INT8 quantization under strict latency and memory constraints. I also run training workloads on an on-premise GPU cluster using Ray and reproducible Docker environments.
At Robert Bosch GmbH, I implemented and evaluated computer-vision research for ADAS, including segmentation, scene classification and object detection. I used failure analysis and data-centric iteration to improve model robustness, and trained and deployed quantized models on embedded hardware for real-time inference.
Earlier, as a Research Intern at Maastricht University, I developed and evaluated deep-learning methods for 3D segmentation of subcortical brain structures in structural MRI. My work also includes distributed GPU training, synthetic-data augmentation and performance optimization with CUDA.
