Daniel Otero Gómez Email & Phone Number

• Led a comprehensive ablation study under the supervision of Professor Randall Balestriero (Brown University) to benchmark self-supervised learning (SSL) methods in computer vision, with a specific focus on anomaly detection in real-world applications.• Conducted 250 experiments using methods like SimCLR, DINO, BYOL, MAE, and Barlow Twins, tested on a non-object-centric sewerage infrastructure dataset with subtle defects, employing ResNet-18 and ViT-Tiny architectures.• The research aims to improve SSL evaluation metrics and create better benchmarks for real-world problems, with results submitted to the NeurIPS workshop "SSL - Theory and Practice."

Applying self-supervised learning to a domain-specific computer vision task• Pioneered the application of self-supervised learning for sewer infrastructure inspection, achieving state-of-the-art performance with a straightforward approach.• Demonstrated strong results on the Sewer-ML dataset using only 10% of the data for fine-tuning, reducing the need for extensive annotations.• Successfully trained a model 5 times smaller than current methods, ideal for deployment on small devices for live detection, enhancing scalability and efficiency in resource-limited settings.Gesture Control on Low Resolution Edge Devices• Developed a low-compute neural network tailored for gesture recognition on edge devices.• Led the data collection and curation process. Implemented a feedback loop that significantly enhanced model performance by incorporating hard samples into the training dataset.• Worked alongside the client to communicate technical limitations and potential solutions. Facilitated informed decision-making and the successful adaptation of the technology to meet specific requirements.Defect Detection Sewer Pipes• Addressed a heavily imbalanced Computer Vision task for defect identification in Sewer maintenance CCTV videos.• Effectively debugged model training, finding critical issues in the data collection strategy.• Trained model on Azure ML Studio and deployed it on AWS SageMaker Batch Transform.

• Collaborated by re-implementing an algorithm proposed to solve heavily constrained problems and integrating it into the library. Reproduced the experiments of the original paper for validation. (Public release soon) • Reinforced knowledge on Lagrangian constrained optimization, advanced linear algebra, and deep learning.

• Created an in-house tool that automates unit test writing for untested code in pull requests, enhancing code reliability and accelerating development timelines.• Used Pytorch, MLflow, and Azure Machine Learning Studio to fine-tune and deploy a LLM that effectively addressed a heavily unbalanced text-classification task.

Colombia

• Built a rule-based monitoring system that alerted the fraud prevention team for suspicious behavior in their web platform and guided technically business projects in Innovation, Customer Support, and Risk Management.• Implemented software tests (Pytest) for the rules and the data pipeline (BigQuery), packaged and containerized the application (Docker), and deployed it in Airflow using GKE. Used CI/CD flow with Jenkins for the development.