repositories

Squishy Robotics grew out of UC Berkeley and NASA work on survivable planetary probes; today it commercializes drop-resilient tensegrity robots that can be air-deployed with customizable sensors and mesh networking to stream real-time ground data from hazardous environments for disaster response, industry, and defense.

Sep 2025–Feb 2026, I worked as an undergraduate researcher with the Berkeley Expert System Technologies Lab on Squishy Robotics methane sensing. I built a large-scale semi-synthetic labeling pipeline (~100,000 hyperspectral training images) that modeled raw robot sensor output from the hardware team to create ML-ready datasets for the ML team. I trained and benchmarked CNN and vision-transformer architectures adapted from recent methane-quantification literature, evaluating them against deployment constraints, diagnosing failure modes, and iterating on data quality and model design for measurable gains.

The Molecular Foundry is a Department of Energy nanoscience user facility at Lawrence Berkeley National Laboratory; the National Center for Electron Microscopy within it provides aberration-corrected electron microscopes and deep expertise for atomic-resolution imaging and advanced characterization workflows.

Jun–Aug 2025, I worked as a research intern with the Molecular Foundry's electron-microscopy scientists and the Crucible scientific data / ML platform. I performed user research to learn what tools the scientists used to analyze their data, then translated findings into concrete product requirements. I designed and shipped a full-stack Data Portal (React, FastAPI, Docker on GCP Cloud Run) so researchers could upload, query, and explore electron microscopy datasets through a single web interface.