Switzerland
I am a Senior Machine Learning Engineer at Logitech focused on the intersection of Applied Research and Product. My work covers the full Computer Vision lifecycle, ranging from high-level research in 3D reconstruction and Generative AI to the technical constraints of on-device, real-time inference. Whether designing 3D CV prototypes or optimizing for low-power hardware, my goal is to build robust AI architectures that power the next generation of consumer devices.
Machine Learning & AI Engineering - On-Device ML: NPU Optimization, Real-time Inference, Model Quantization, Low-power Deployment. - Deep Learning Frameworks: PyTorch, MLOps, Neural Network Architecture Design. - Generative AI: Stable Diffusion, ControlNet, Vision Language Models (VLMs), Rapid Prototyping. - Computer Vision: Object Tracking, Scene Understanding, Depth Estimation, Point Cloud Alignment, Camera Calibration. - 3D Vision & Reconstruction: RGB-D Integration, Novel View Synthesis (NVS), 3D Telepresence. Data & Infrastructure - Data Strategy: Dataset Curation, Synthetic Data Generation (SDG), Data Acquisition. - Software Engineering: GitHub Workflow, CI/CD, Code Quality/Maintainability, Python, C++. Leadership & Strategy - Research Leadership: Literature Synthesis (CVPR, ICCV, ECCV), Technical Mentorship, Intern Supervision. - Cross-Functional Advisory: AI Ethics, Data Compliance, Legal/Business Teams Technical Guidance.
Thesis Title: Volumetric scenes and people: capture, segmentation and rendering improvement Problem: Addressed rendering defects and distortions in 3D point clouds that hinder the quality of human holograms in consumer-grade hardware. Calibration & Optimization: Developed a complete calibration pipeline for the Microsoft Azure Kinect to correct depth-color mismatches and factory-calibration errors, significantly improving visual fidelity. Artifact Removal: Engineered an algorithm to identify and prune deformed regions in 3D point clouds, optimizing data for multi-camera placement strategies. Background Segmentation: Benchmarked and implemented offline segmentation methods for cluttered, real-world indoor environments, reducing the reliance on green screens or controlled studio spaces. Vision: Focused on the distinction between "functional 3D for machines" and "visually pleasant 3D for humans" to define the next generation of immersive media.