Computer vision · Foundation models · Reinforcement learning
I work on learning systems, from 3D perception to foundation models.
I completed my Ph.D. at Carnegie Mellon University, advised by Kris Kitani, with a focus on multi-view geometry and 3D human pose estimation. In December 2022, I joined Waymo to work on 3D perception for autonomous driving.
I’m now exploring foundation models, post-training, and reinforcement learning through hands-on implementations and experiments.
Joined as a Software Engineer
3D object detection for autonomous driving, spanning model development, training, evaluation, and deployment.
Computer Vision · Advisor: Kris Kitani
Multi-view geometry and multi-person 3D human pose estimation beyond calibrated laboratory settings.
Research · BMVC 2022
Multi-View Multi-Person 3D Pose Estimation with Uncalibrated Camera Networks
Reconstructing multiple people in 3D beyond a calibrated camera setup. A study at the intersection of human pose estimation and multi-view geometry.
Research · CVPR 2021
Wide-Baseline Multi-Camera Calibration using Person Re-Identification
Using person re-identification to connect camera views and estimate their relative geometry, even across wide baselines.
A small experiment
Temperature changes how a language model turns scores into next-token probabilities. Try it.
p = softmax(logits / T)
Fixed example logits: [2.0, 1.2, 0.4, −0.1]. Change T to compare the same four candidates.
Writing
Long-form notes on foundation models and post-training.
Currently taking shape, one experiment at a time.
