As the sole Applied ML Scientist at Liburdi Dimetrics, I own the full AI pipeline end-to-end—from data strategy and model architecture to edge deployment and production delivery—delivering production-grade multi-task models across 15 welding domains for 6 customers. I architect and deploy real-time models for weld inspection, defect detection, and robotic guidance, sustaining around 10 FPS on 2GB VRAM under challenging visual conditions like smoke, glare, motion, occlusion, and dynamic lighting. I built a deep learning video annotation framework using the Segment Anything Model (SAM), reducing manual labeling effort by about 90% and accelerating dataset creation for new domains. I improve cross-domain generalization and adaptation speed through self-supervised pretraining, multi-domain training, and transfer learning, and I enhanced temporal stability of live video segmentation by adding lightweight recurrent modules to transformer single-frame models, reducing frame-to-frame flicker on production streams. Academically, I hold an M.Sc. in Computer Science from the University of Calgary with peer-reviewed publications at ECCV and MICCAI. My research spans semi-supervised segmentation and medical image analysis, including a state-of-the-art wheat head segmentation approach and a kidney tumor segmentation method that featured in MICCAI's KITS23 challenges. I enjoy making ML research usable and impactful in real-world settings, bridging theory and production with careful evaluation and deployment.…As the sole Applied ML Scientist at Liburdi Dimetrics, I own the full AI pipeline end-to-end—from data strategy and model architecture to edge deployment and production delivery—delivering production-grade multi-task models across 15 welding domains for 6 customers. I architect and deploy real-time models for weld inspection, defect detection, and robotic guidance, sustaining around 10 FPS on 2GB VRAM under challenging visual conditions like smoke, glare, motion, occlusion, and dynamic lighting. I built a deep learning video annotation framework using the Segment Anything Model (SAM), reducing manual labeling effort by about 90% and accelerating dataset creation for new domains. I improve cross-domain generalization and adaptation speed through self-supervised pretraining, multi-domain training, and transfer learning, and I enhanced temporal stability of live video segmentation by adding lightweight recurrent modules to transformer single-frame models, reducing frame-to-frame flicker on production streams. Academically, I hold an M.Sc. in Computer Science from the University of Calgary with peer-reviewed publications at ECCV and MICCAI. My research spans semi-supervised segmentation and medical image analysis, including a state-of-the-art wheat head segmentation approach and a kidney tumor segmentation method that featured in MICCAI's KITS23 challenges. I enjoy making ML research usable and impactful in real-world settings, bridging theory and production with careful evaluation and deployment.WWWW…