Physics | Data Science | AI

Data-Driven Science

For Social & Environmental Good

Intro to me

Hello there! I am a Lecturer in Physics at Universidad de Costa Rica. My research and teaching focuses on interdisciplinary AI applications to physics, ecology and fishery science. Through my lectures I seek to give students practical experience with tools from data science, classical machine learning and deep learning, which find increasing applications in science.

After working on theoretical quantum computing during my PhD at the University of Cologne, Germany, my research interests shifted towards more applied and interdisciplinary topics. My driving question throughout this process has been how can a mathy guy like me have a positive impact on the world?

This motivating principle led me to work as a postdoc UC Berkeley’s Schmidt Center for Data Science & the Environment, and the Department of Enviornmental Science, Policy and Management. There I led the establishment of a research program at the intersection of data science, machine learning and quantitative ecology.

Open to work. I am now excited to transition from academia to the industry as a data scientist or machine learning engineer. I am interested in on-site, hybrid and remote options in Costa Rica, Spain and Italy.

Get in touch Download CV

Me in a beautiful garden!

Software toolkig

Data toolkit in my work

Through my research projects and teaching I’ve focused on bringing modern software development practices to academic work, promoting the reproducibility and reusability of my work. Here I summarize my software and data-related skills I have gained with this approach.

Reinforcement learning. I’ve extensively worked in developing reinforcement learning approaches in environmental science contexts such as fishery management, invasive species management, caribou conservation, and dynamical multispecies stabilization. Throughout this work I have become experienced with building custom neural network models in PyTorch, training feed-forward and recurrent networks and generating CI/CD pipelines for unit testing. I moreover am leading the creation of an SQL database to enhance the reproducibility of our team’s fishery RL work.

Statistical inference, fullstack development. Through my collaboration with Karuk wildlife managers and researchers, I have become experienced in statistical inference pipelines in R and Python, as well as in the deployment of graphical apps through GitHub Actions, and in setting up computer vision tools for wildlife monitoring. In this project I pivoted from creating code internally within my research team to developing a backend and frontend for an external stakeholder. The code repositories are private to protect indigenous data sovereignty, however our first paper is published here! There, we used Bayesian hierarchical modelling and MCMC to infer accurate estimates of the abundance and homerange of elk in the study area using community observational data.

Cloud-native geospatial data processing. For a few months I contributed to a project developing a quantitative geospatial approach to environmental justice. Geospatial data is typically large and thus hosted in the cloud, usually in AWS S3 buckets, thus quantitative geospatial analyses are more naturally suited to cloud-native approaches even though they are commonly implemented locally in academic work. This project seeked to close this gap, providing tutorials of how to efficiently process geospatial data in environmental research contexts in a cloud-native manner. To maximize the reproducibility of this work, we containerized our pipeline.

ML in teaching. In teaching the Advanced computational physics lecture at UCR, I have become experienced with a variety of ML algorithms which, while they were adjacent to my work, had not shown up in my research projects. For example I have become experienced in setting up practical workshops on using SciKit-Learn, SciKit-Optimize and PyTorch to run ML classification tasks, logistic regression, ridge regression, support vector machine optimization, k-means clustering, and semi-definite programming optimization.

Data visualization. Given that I am a mathy guy who works with both physicists and biological scientists, it has been crucial for me to provide informative and actionable data visualizations every step of the way in my collaborations. I believe that data visualization is the ground on which one can build successful, cross-functional collaborations.

Additional tools. I also use agentic AI tools to support software development: this website was deployed with the Copilot CLI, which I have also used to debug and optimize an ongoing reinforcement learning project.

Mentoring

Mentoring is an important aspect of my professional profile. Knowledge sharing is powerful in industry and academia, and good mentoring can vitalize teams’ productivity, especially in cross-functional and interdisciplinary teams. These are the research projects I have had the priviledge of mentoring:

• Kendall Salazar, U of Costa Rica. Research project on RL for multi-species Lotka-Volterra harvesting problems.
• Aaron Sanabria, U of Costa Rica. Research project on MCMC and statistical estimation for heavy-ion collision data.
• Maxwell Tanguchi-King, UC Berkeley. Internship on software development and indigenous data sovereignty in biodiversity monitoring.
• Nithya Appannagaari, UC Berkeley. Internship on software development and indigenous data sovereignty in biodiversity monitoring.
• Claire Liu, UC Berkeley. B.Sc. thesis on predictability of extreme rain events in the SF Bay Area. Undergrad research assistantship on RL approaches to partially observable population dynamics control problems.
• Maya Nalliah, UC Berkeley. Undergrad research assistantship on RL approaches to highly stochastic and non-stationary population dynamics control problems.
• Felix Hoeddinghaus, U of Cologne. B.Sc. thesis on discrete algebraic groups arising in a quantum computing context.
• Angelos Bampounis, U of Cologne. M.Sc. thesis on representation theory of the single-qubit Clifford group.
• Paula Belzig, U of Cologne & ETH Zurich. M.Sc. thesis on the Stabilizer de Finetti theorem and applications to quantum communication.
• Tobias Hartmann, U of Cologne. B.Sc. thesis on state-independent contextuality for few-qubit systems.
• Stefan Schmit, U of Cologne. B.Sc. thesis on parametric noise on quantum control problems.

Research

My research line uses a variety of deep learning, data science and applied statistics to address hard scientific questions.

I’m also dabbling in other projects like on statistical inference for heavy ion collision experiments with my close friend and collaborator Oscar García. Similarly, I occasionally enjoy contributing to the open source ecosystem on, e.g., MyST or improving the experience of other lecturers in generating useful class list documents.

Publications

Selected publications and works in progress. For a complete list, please visit my Google Scholar profile.

Teaching

Music

Besides my academic work, I love playing music! Here I’ll post songs from several of my musical projects.