Leading research on multimodal AI systems with focus on video understanding, video generation, and autonomous systems.

• Video Generation: Built and shipped production-scale video generation models using VLMs and diffusion architectures on ModelArts MLOps platform
• Training Infrastructure: Designed end-to-end training pipelines with containerized environments and reproducible experimentation playbooks
• ReWind (CVPR 2025): Co-developed a large language model for long video understanding with instructed learnable memory
• OCR & Multimodal Perception: Spearheaded OCR capabilities, deploying multimodal perception models across multiple languages
• 4D LiDAR Generation: Working on OmniLiDAR for controllable and multi-sensor 4D LiDAR generation (submitted to ECCV 2026)

Deep learning research for kinematic time series analysis and robotics applications.

• Designed deep learning pipelines for kinematic time series analysis, reconstructing pen trajectories from motion sensor data using TCNs
• Published in International Journal on Document Analysis and Recognition (IJDAR)
• Optimized DNNs on resource-constrained edge devices for real-time sensor processing
• Collaborated in Franco-German initiative, translating perception models into engineering roadmaps

Combined research and teaching responsibilities with focus on deep learning and 3D modeling.

• Led R&D on UNet-based architectures for visual perception and 3D modeling of ancient maps
• Versailles-FP dataset: Wall Detection in Ancient Floor Plans (ICDAR 2021)
• Supervised Master's research on neural network optimization and multimodal learning

Research and development in historical document analysis and information extraction (EURHISFIRM project).

• Developed production DNN pipelines for named entity extraction from historical documents
• Implemented scalable data preprocessing and model training workflows with automated evaluation
• Published in Document Analysis Systems (DAS 2020)