Resume

Institut für Neuro-und Bioinformatik, Lübeck

• Developing and deploying AI solutions for iSCAT microscopy and the KIMEKO AI medical collaboration platform.
• Developed generative diffusion models to synthesize realistic iSCAT nanoparticle images for data augmentation.
• Built LLM-based pipelines for medical causal relation extraction from unstructured clinical text; fine-tuned open-source LLMs (Gemma, LLaMA, Qwen) with LoRA, achieving a 21% improvement in F1 score over the baseline.
• Collaborated with clinical domain experts to curate high-quality causality annotations and built scalable post-processing pipelines for labeled datasets.
• Supervised and provided technical mentorship to a Master's student, guiding experimental design and model development.

Institut für Neuro-und Bioinformatik, Lübeck

• Project: Developed end-to-end deep learning pipelines for nanoparticle segmentation and size estimation using interferometric scattering microscopy (iSCAT) images in collaboration with Harvard University.
• Implemented and benchmarked SAM, Mask R-CNN, CellViT, and UNet architectures; achieved 76% F1-score and 81% Dice on held-out test data.
• Developed ResNet and ViT models for accurate nanoparticle size estimation using physics-informed constraints.

Institut für Neuro-und Bioinformatik, Lübeck

• Implemented and deployed a model for automatic image cropping as a web application using Docker Compose.
• Designed and built an end-to-end framework to train an RL agent for tonal value adjustment (e.g., contrast, saturation).
• Achieved state-of-the-art RL methods in tonal adjustment on the MIT-Adobe FiveK dataset, achieving 26.75 dB PSNR and 0.85 SSIM.
• Deployed a demo app: photo-enhancer featuring manual and RL-based editing with color histograms for photographers.

Institut für Technische Informatik, Lübeck

• Worked on the DFG-funded METERACOM (Metrology for THz Communications) project focused on future 6G/THz communication systems.
• Designed and implemented reinforcement learning algorithms to optimize adaptive beamforming in mmWave and THz MIMO systems, improving beamforming gain under realistic conditions and hardware impairments.
• Evaluated RNN-, CNN-, and transformer-based neural network architectures for modulation classification.

University of Lübeck, Germany

École Nationale Supérieure d'Arts et Métiers Casablanca, Morocco