From complex data to simple solutions
Professional Experience
Projects I've been involved in throughout my professional career
Architected and scaled machine learning solutions that revolutionize biotech research and clinical applications. Transformed complex proprietary workflows into standardized, accessible libraries while implementing robust testing frameworks across cloud infrastructure.
Created AI solutions that not only perform technically but genuinely advance healthcare outcomes through improved *omics-based target identification and analysis of medical cohort data at scale.
Coordinated initiatives across five specialized teams to enhance system reliability for distributed ML pipelines, serving as a technical connector to bridge the gap between cutting-edge AI technology and practical applications in medical research.
Developed innovative bias mitigation techniques for medical imaging that brought AI-powered analysis closer to clinical trial readiness, making precision medicine more accessible and reliable.
Pioneered AI technologies for super-resolution microscopes that transformed how biological research is conducted. Engineered an active learning system that automatically identified critical biological features, empowering scientists by reducing data analysis time by over 10X.
Enhanced microscopy research capabilities by implementing cutting-edge instance segmentation and federated learning techniques that significantly improved our smart microscope's functionality.
Strengthened the technical foundation of our AI initiatives by implementing robust CI/CD pipelines and expanding test coverage to 90%, creating a sustainable platform for ongoing innovation that continues to solve common microscopy research challenges and empower biologists with powerful, accessible machine learning tools.
📍🇩🇪 Berlin (Germany)
📍🇬🇧 Oxford (United Kingdom)
📍🇷🇴 Cluj-Napoca (Romania)
Pioneered innovative computer vision solutions by engineering two novel CNN-based systems for 3D scene reconstruction from stereo images. Designed custom neural network architectures and loss functions that pushed the boundaries of what was previously possible with traditional approaches.
Created solutions that not only outperformed the department's analytical benchmarks in both accuracy and smoothness but also achieved performance comparable to top-10 submissions in the prestigious KITTI benchmark – a significant achievement for research-stage work.
Academic Career
Highlights of my academic ventures
This research project presents an innovative approach to human re-identification in surveillance systems by developing a feature selection model that identifies subjects from low-resolution CCTV images based on soft biometric descriptions.
The work addresses a critical challenge in modern security applications where traditional biometric methods fail due to poor image quality and distance constraints. By leveraging semantic queries that describe human-observable traits like clothing color, gender, and pose, the system bridges the gap between human descriptions and machine understanding, enabling automated analysis of surveillance footage without manual inspection.
(Click on the title of this section to access the original paper)
Feature selection for subject ranking using Soft Biometric Queries
📍🇬🇧 University of Southampton (United Kingdom)
(MSc Artificial Intelligence Dissertation Project)
Stereo Disparity estimation using Deep Neural Networks
📍🇷🇴 Technical University of Cluj-Napoca (Romania)
(Bachelors Degree Research Project)
This project tackles the fundamental computer vision problem of stereo disparity estimation, which involves computing depth information from pairs of stereo images by finding corresponding pixel matches between left and right camera views.
The challenge lies in developing robust algorithms that can accurately match pixels across stereo image pairs while handling occlusions, texture-less regions, and illumination variations that commonly occur in real-world scenarios.
The project makes significant personal contributions to the field, particularly through the development of an entirely novel approach that was designed, implemented, and trained from scratch using an original training procedure not previously documented in the literature.