Alexis Alva

- Analyzed ganglion cell data using unsupervised learning algorithms to identify biological activity patterns, contributing to the discovery of new insights in retinal signal processing. - Built and maintained machine learning pipelines using DVC (Data Version Control), enabling reproducible experiments and efficient model training and version tracking. - Collaborated closely with neuroscience researchers to interpret model outputs and iteratively refine models based on scientific findings, aligning ML outcomes with experimental objectives. - Delivered an oral presentation of project progress and results at the Society for Neuroscience Annual Meeting, effectively communicating complex ML methodologies and discoveries to a specialized academic audience.

- Implemented and customized open-source platforms such as Open Journal Systems, Open Monograph Press, and Data Management Plan, aligning them with institutional requirements and improving research workflow efficiency. - Administered cloud infrastructure on DigitalOcean, ensuring reliable deployment, maintenance, and performance of research data platforms. - Collaborated with a multidisciplinary team through regular meetings to iteratively adjust platforms based on user feedback and evolving university priorities. - Supported research on research data management practices, contributing to the development of open science initiatives at UTFSM and advancing institutional data governance efforts.

- Trained and optimized YOLO-based object detection models to identify weeds in drone imagery, improving accuracy through experimentation, hyperparameter tuning, and dataset refinement. - Labeled high-resolution aerial images to build a reliable training dataset, enabling accurate model learning in agricultural settings. - Used ClearML to track experiments and monitor performance in real time, streamlining the model development process.

- Improved the accuracy of a geolocation algorithm for traffic sign detection by refining the alignment between video frame data and GPS signals, increasing positional precision for sign mapping. - Trained autoencoder neural networks to detect degraded traffic signs (e.g., faded or damaged) along mining routes, enabling timely identification and replacement to enhance road safety and operational compliance. - Developed and optimized data pipelines to improve the solution’s workflow, facilitating efficient information processing and integration of system components.

- Designed and implemented a full machine learning pipeline for automatic number plate recognition, covering data labeling, model training, and deployment. - Developed and annotated a custom dataset to support accurate object detection and character recognition. - Deployed the optimized model on NVIDIA Jetson Nano, achieving real-time inference performance on edge hardware.

Este formato es el más profesional y fácil de leer gracias a las viñetas. International Intensive Training Program in AI Co-organized by Universidad Nacional de Hurlingham (UNAHUR) & Universidad Nacional de San Martín (UNSAM) Addressed the technical and ethical challenges of AI within the framework of the "Santiago Declaration" for regional technology governance. Specialization Certifications (23.5 academic hours) Successfully completed modules by solving technical challenges in working groups: Advanced Computer Vision: "The Convolutional Neocognitron" Course (8 hours). Natural Language Processing (NLP): "Artificial Neural Networks and Natural Language" Course (8 hours). AI applied to Healthcare: "Artificial Intelligence and Degenerative Diseases" Course (7.5 hours).

The Latin American School on Computational Neuroscience (LACONEU) is an intensive program that combines theory and practice in computational modeling of neural systems. I worked on implementing biologically plausible neural network models, explored simulation techniques and neural data analysis, and collaborated with researchers from diverse fields on interdisciplinary projects.