I work on visual understanding, multimodal models and agentic systems. My experience spans research and production, from problem formulation and model development to evaluation and deployment. I prefer roles with substantial technical ownership and hands-on individual-contributor scope.
Developed VLM-based monitoring for content relevance on search pages. The approach uses structured probes, adversarial evaluation and prompt calibration to identify visual failures that aggregate behavioural metrics may miss.
Built a detector across 25 visual defect categories. Diffusion-based augmentation supplied targeted examples for six data-scarce categories, and LoRA adaptation of Qwen-VL improved F1 from 0.62 to 0.71.
Developed agentic navigation for Amazon Retail pages, combining visual understanding, interactive-element discovery, action selection and trajectory analysis. The system outperformed state-of-the-art web-navigation methods by nine percentage points. Co-authored benchmark work evaluates agents on realistic multi-step shopping missions.
Led a visual-relevance system for search suggestions, combining ViT and CLIP representations with clutter moderation and depth-guided segmentation. The work generated an estimated $74M in added revenue. Also built graph models for related-query recommendations.
Co-authored MDG-DIS, which uses depth information during training to improve foreground separation in overlapping scenes, then runs inference without a depth input.
Multimodal evaluation, synthetic-data defect detection and agentic navigation.
Visual search, image quality, segmentation and recommendation systems.
Full-body reconstruction from sparse IMU data, sensor fusion and state estimation. Invented and patented a method for imputing missing sensor data.
PhD research in spatiotemporal feature learning, hand-pose classification and gesture recognition. Built a Time-of-Flight depth-camera system for in-vehicle gesture interaction.
Submitted; not accepted · Co-authored benchmark work
ECCV Workshop · with Oleg Smirnov
IEEE IVMSP
IEEE ISMAR · with Jason Rambach, Alain Pagani and Didier Stricker
Journal of WSCG
I’m a computer vision and machine-learning scientist. My work has moved from spatiotemporal features and inertial sensing to visual discovery, multimodal models and agentic evaluation.
I value rigorous experiments, careful failure analysis and direct technical ownership across the research-to-production path. Based in Luxembourg.