Ph.D. candidate · Gu Research GroupTake the tour
/ solving polymer problems with AI & robotics
ALS Doctoral Fellow · Berkeley Lab
Berkeley, California
I solve polymer science problems with AI and specialized processing robots, built into self-driving labs.
/ how a self-driving lab works
Each stage hands off to the next, and the model chooses what to try. Watch it run, or click through yourself.
The robot mixes formulations, coats them onto wafers and racks them for measurement, sample after sample.
The formulation and coating robot, sped up 20×. The robotic platform →
Grazing-incidence X-ray scattering turns each film’s nanoscale order into a 2D pattern.
How GISAXS works: a shallow beam probes the film; the detector records how its nanostructure scatters. Beamline campaigns →
NOVA reduces each new detector image, fits its scattering peaks and records the results as measurements come in.
NOVA working through masking, reduction, peak fitting and automated monitoring, sped up 16×. About NOVA →
Each round of results reshapes the model’s picture of the design space and sharpens where to look next.
The model’s predicted q₂/q₁ surface over eleven iterations; stars mark each round’s proposed compositions. Learning from the data →
The updated model proposes the next batch — where the prediction looks promising, or where it is least certain — and sends it to the robot to start the next round.
An illustration of how a Gaussian-process model picks the next batch. Bayesian optimization in practice →
Citation figures from Google Scholar, September 2026.
/ selected work
in progress · ALS beamline 7.3.3
A team-built closed loop in which a Gaussian-process optimizer drives a synchrotron beamline directly, finding the processing route to a named target morphology in 20–30 experiments instead of thousands.
20–30experiments to hit the target, versus a grid scan of thousands
first author · ACS Nano 2026
Hundreds of thin-film samples fabricated in hours, measured by synchrotron scattering, and turned into the first high-throughput map of the order–disorder transition landscape in triblock copolymer–homopolymer blends.
100sof thin-film samples fabricated and characterized per day
co-first author · Macromolecules 2026
A machine learning framework that predicts block copolymer thin-film morphology from processing conditions, trained on high-throughput scattering data across six chemistries.
6block copolymer chemistries in one generalizable model
now building · Berkeley Lab
Current work at Berkeley Lab — a multimodal self-driving lab that uses active learning to map polymer blend morphology and phase behavior, choosing each next sample where the map is least certain.
NowALS Doctoral Fellowship, Berkeley Lab, 2026–27
Illustration/ in print
Want to talk about self-driving labs?Email me, or read more about me.