Esteban Rodríguez-Betancourt

Universidad de Costa Rica

San José, Costa Rica

I am both a professional Software Development Engineer and an academic researcher. Currently, I serve as a Software Engineer II at Microsoft, working on Skype’s Experimentation and Configuration Service, where I design and implement high-performance systems. Simultaneously, I am pursuing a PhD in Computer Science at Universidad de Costa Rica, researching natural language processing, machine learning, and distributed systems.

My work at Microsoft involves designing scalable systems, implementing performance optimizations, and contributing to production services used by millions of users. My academic research focuses on semantic shift analysis, word embeddings, and innovative applications of database technologies for vector indexing.

Previously, I served as a Professor at the School of Computer Science and Informatics (ECCI) at Universidad de Costa Rica (2022-2024), where I taught programming courses and introduced students to modern software engineering practices including performance tuning, memory debugging, CI/CD, and collaborative development with Git.

Before Microsoft, I worked as a Software Developer Engineer at Amazon and spent five years at Hewlett Packard Enterprise (Aruba Networks), contributing to networking operating systems and the open source Open vSwitch project.

My research interests span multiple areas including natural language processing, deep learning, database internals, distributed systems, and algorithmic solutions to computational problems. I have published research on topics such as semantic shift in diachronic word embeddings, efficient vector indexing using SQL trigrams, and representation learning in vision systems.

selected publications

Hypersolid: Emergent Vision Representations via Short-Range Repulsion

Esteban Rodríguez-Betancourt and Edgar Casasola-Murillo

2026

Abs arXiv Bib

A recurring challenge in self-supervised learning is preventing representation collapse. Existing solutions typically rely on global regularization, such as maximizing distances, decorrelating dimensions or enforcing certain distributions. We instead reinterpret representation learning as a discrete packing problem, where preserving information simplifies to maintaining injectivity. We operationalize this in Hypersolid, a method using short-range hard-ball repulsion to prevent local collisions. This constraint results in a high-separation geometric regime that preserves augmentation diversity, excelling on fine-grained and low-resolution classification tasks.
@misc{rodríguezbetancourt2026hypersolidemergentvisionrepresentations, title = {Hypersolid: Emergent Vision Representations via Short-Range Repulsion}, author = {Rodríguez-Betancourt, Esteban and Casasola-Murillo, Edgar}, year = {2026}, archiveprefix = {arXiv}, primaryclass = {cs.CV}, url = {https://arxiv.org/abs/2601.21255}, }