Signal, not noise
I'm currently at the Wu Tsai Human Performance Alliance at Stanford, where I build individual-level models of physiological adaptation from longitudinal wearable and training data, and lead a cohort-scale study on menstrual cycle physiology and sleep. I also manage the Alliance's external partnerships, run the Agility Projects program ($1M in annual funding), and keep track of its research portfolio across Stanford.
I studied electrical engineering at the University of Puerto Rico at Mayagüez, then at Stanford, where I completed my PhD with support from NSF, Sloan, and NIH fellowships. My background also includes postdoctoral work on hippocampal and parahippocampal circuits in rodents, supported by an NIH NRSA fellowship and funded in part by the Simons Foundation, and signal processing consulting for FDA-regulated cardiac devices.
First principles, sound methods, the right tool for the problem.
What I work on
Physiological adaptation & performance
Modeling training responses, injury risk, and recovery from longitudinal wearable data; investigating injury mechanisms in elite athletes from biomechanics and historical load data.
Menstrual cycle biology
Cohort-scale normative physiology, sleep, and cycle variability from wearable biometrics (npj Digital Medicine, 2026).
Neural coding & memory
How hippocampal and parahippocampal circuits represent space and experience.
Toolbox
Wearable data is observational, unbalanced, and missing nights. Natural experiments, state-space models, and double-robust ML pull causal answers out of it anyway.
Example: sleep-phase natural experiment
Off-the-shelf architectures rarely match biological questions. I design encoder models and custom architectures where the hypothesis is built into the model itself.
Example: eLife 2024 · encoder-decoder notebook
Biological signals bury structure under artifact. Adaptive filtering, multi-band decomposition, and real-time event detection, from EEG-fMRI to an FDA-regulated wearable defibrillator.
Example: EEG-fMRI preprocessing code
What does a population of neurons know, and when? Representational similarity analyses and population-level decoders, from single units to intracranial EEG.
Example: PNAS 2015 · eLife 2024
AI agents are only as rigorous as the structure around them. I build ADR-governed research repos with codified assumption surfacing and parallel subagent review.
Reproducible by design
I try to make my research auditable end to end: not just sharing code, but sharing it in a form where every figure and statistic can be traced back to the data.
Fully reproducible analysis, from raw biometrics to published figures
The menstrual cycle paper ships with dedicated figure-generation code, every statistic worked through in Jupyter notebooks, and the API built to generate those figures and statistics, alongside a companion website. Browse the repo, or see an example notebook: the sleep-phase natural experiment.
Open code for key papers
Analysis and behavioral control code for the eLife navigation work (TMA, TreeMaze), and MATLAB source for the PNAS ECoG analyses (repo).
Neuromatch Academy
Contributor to Neuromatch's open, global computational neuroscience school: leadership roles, conflict-resolution ombuds team, and mentoring dozens of students.
Publications
Honors & Fellowships
Meanwhile
While my Claude Code sessions run, you might find me on a run of my own, cycling up the Santa Cruz Mountains, homebrewing beer, making pizza, or chasing two little ones.