high school senior in san antonio, texas (tmi episcopal, class of 2027). researcher and builder working where computer science, environmental science, and startups meet.
sole-authored deep learning model that predicts wildfire risk from high-resolution satellite imagery. a mobilenet cnn extracts features, which train mlp, knn, and random forest classifiers. independent research, no mentor or co-authors.
presented solo at ieee igarss 2026, washington, d.c. (august 2026). published in ieee xplore: ieeexplore.ieee.org/document/11730559. also accepted as a poster at agu26.
independent study testing whether frozen-encoder geospatial foundation models (prithvi-eo-2.0, clay) beat a fully trained u-net when labels are scarce and the test region is unseen. uses sen1floods11 in a 90-run matrix — 3 models × 5 label budgets × 3 seeds — with held-out events in somalia, sri lanka, and spain.
key finding: at 10% labels, clay beat the u-net on every unseen event across all seeds — foundation models are not uniformly superior. accepted as a poster at agu26 (december 2026).
recognition: selected for agu's outstanding emerging presenter program and agu circles.
a productivity app where an animated cactus mascot celebrates completed tasks and sustainable habits. built solo with react and django.
funded by a $1k starseed grant. teyra.app
an open-source cli on npm that grows a pixelated forest in developers' terminals as they code and funds real trees in kenya and ethiopia. 2,500+ downloads.
a free, open-source cold-email tool that sends personalized daily batches from gmail.
eliee. groundwork — a cli that generates persistent ai context files for coding assistants.
lvlup ventures: source and screen early-stage startups for a ~$250m aum fund; reviewed founder pitches and relayed top deals to the investment team.
samsat: led development of an environmental-science quiz platform; in contention for a $2k grant with potential national rollout.
uiw: built an ai email spam classifier; demoed to 60+ attendees; won best overall.