I am a Machine Learning Engineer bridging the gap between deep learning, computer vision, and computational biology. Currently completing my Master's in Bioengineering at Universitat Oberta de Catalunya (UOC), I specialize in building custom, mathematically rigorous ML architectures to solve complex biological bottlenecks, from real-time cytogenetics to transcriptomic classification.
- ChromoSeg-YOLO (Python, PyTorch): Engineered a high-throughput instance segmentation engine for clinical cytogenetics. Overrode standard YOLO loss with a custom GPU-accelerated Dice-Focal and Morphological Boundary Loss to resolve touching chromosome clusters. Achieved a 76% reduction in count error with 3.2 ms latency. Fully tested via Pytest and deployed as an interactive Hugging Face Gradio app.
- Alzheimer's Transcriptomic Classifier (R, Shiny): Developed an end-to-end bioinformatics pipeline analyzing peripheral blood gene expression to detect early-stage Alzheimer's. Conducted EDA, PCA, and differential expression analysis (limma), training multiple ML models. The optimized SVM achieved an AUC of 0.815, packaged into an interactive clinical Shiny application.
- Deep Learning & Vision: PyTorch, Ultralytics YOLO, OpenCV, Custom Loss Functions
- Statistical ML & Bioinformatics: R, Bioconductor, limma, SVMs, PCA, XGBoost/GBM
- MLOps & Engineering: Git, Hugging Face Spaces, Docker, Pytest, CI/CD
- Hardware Acceleration: NVIDIA CUDA, Apple Silicon MPS
- LinkedIn: [https://www.linkedin.com/in/margot-bonilla/]
- Hugging Face: [https://huggingface.co/MargotBonilla]