Juan Manuel

• End-to-end computer vision pipeline for tire defect detection, processing over 10,000 high-resolution images/hour (768×768 px) across linear RGB and laser channels.
• Training and fine-tuning of models with over 300M parameters (Transformer and CNN architectures), achieving over 90% accuracy in multiclass classification across more than 10 categories with complex features.
• Systematic research of training and preprocessing configurations to improve production performance: class imbalance analysis, data augmentation and hyperparameter optimization.
• Development of ETL pipelines for building and versioning visual datasets at industrial scale.
• Definition and implementation of evaluation metrics (Precision, Recall, F1-Score, mAP, IoU) and tracking with MLflow and TensorBoard.
• Production model deployment with Docker and AWS.
• Writing technical documentation for architectures, datasets and pipelines.

Skills: PyTorch | TensorFlow | Computer Vision | Deep Learning | CNN | Transformers | Object Detection | OpenCV | MLflow | Docker | AWS

• Analyze and define project requirements, designing and developing machine learning solutions using state-of-the-art algorithms.
• Design and implement custom machine learning algorithms for a variety of projects, including natural language processing, neural networks and computer vision, among others.
• Stay up to date with the latest research advances in various fields of artificial intelligence.
• Write and review technical documentation.
• Meet with clients to understand their requirements and tailor solutions to their needs.
• Collaborate with other team members to achieve project objectives.

• Development and research of Lagrangian techniques for automatic identification of atmospheric blocking situations.
• Collaboration on other group projects related to atmospheric physics.