HUSSEINE KEITA

– Created an Al system from scratch based on deep learning algorithms in order to automate a manual examination images quality control process for pharmaceutical clinical trials leading to 65% workload decrease – Created a QC classifier for medical device images that removes low-quality inputs, boosting trust and consistency in diagnosis. – Employed AWS Sagemaker to train classification models on images extracted from the company's device – Explored self supervised learning on unlabeld images to build quality representation learning pipeline and reducing labeling costs – MLflow tracking for data and metrics; promoted to Model Registry on target hit; biannual performance stability checks. – Delivered edge inference by reducing CPU latency from 200 ms → 8 ms via ONNX export/OpenVINO INT8 while holding accuracy decisions backed by repeatable benchmarks across multiple architectures. – Collaborated with doctors, clinical and ultrasounds specialists to align with their needs and constraints

– Specialized in object detection and segmentation techniques, these methods were applied to microscopy images to extract different cell features – Employed a GPU cluster to effectively manage the distributed training and optimization of Python-scripted deep learning models, including multimodal and Vision Transformers models, achieving greater efficiency and faster processing times – Merged omics data with phenotypic characteristics derived from cell segmentation, enriching the analysis of cell behaviors – Read research papers focused on computer vision, ensuring continuous alignment with state-of-the-art models and methods