Leopoldo Lopez

I’m a Machine Learning Engineer with 6 years of experience spanning applied research and production deployment of machine perception systems on resource-constrained edge hardware. I’ve worked in a formal industry-academia structure (QAISC / CTIM at ULPGC), so my process blends production engineering discipline with peer-reviewed research rigor. My focus is end-to-end CV: from architecture selection and distributed training to evaluation, quantization, and real-time deployment.

I’ve owned the full ML lifecycle for multi-modal perception, defining measurable KPIs like latency, accuracy, and memory—and driving improvements with concrete outcomes. In production, I delivered 10× inference speedup through model optimization and achieved +5 F1 over SOTA (Q1 publication, Automation in Construction 2025). I also deployed real-time multi-modal perception continuously across multiple production environments, monitoring performance to keep systems robust.

On the engineering side, I’ve led multi-GPU distributed training using PyTorch DDP with SLURM, and built evaluation workflows that support ablation studies and threshold tuning. For deployment, I routinely take models through ONNX export, ONNX graph surgery, post-training quantization, and TensorRT conversion to meet edge constraints. I’ve implemented production-grade pipelines for model optimization, streaming inference, and reliability under multi-camera workloads.

I’m equally comfortable with research-driven problem solving—like designing small-object and OCR pipelines, fusing multi-sensor detections to suppress false positives, and using depth/geometry for scene understanding. I’m excited by roles where I can keep pushing real-time perception performance while maintaining observability and deployment reliability end-to-end.