Roblox Engine, Avatar Heads & Bodies

• —Establishing the team's ML foundation: machine learning for evaluating avatar quality at platform scale

Meta Reality Labs Codec Avatars, DataGen

• —Used DiT- and U-Net-based diffusion models to generate 8m+ frames of HMC synthetic datasets, powering downstream avatar models for '26+ ARVR devices
• —Published: GenHMC (arXiv 2025) ↗ — novel method to improve codec avatar encoder's SoTA accuracy + data efficiency via diffusion-generated HMC images

Cruise Perception, Detection Core

• —Boosted multi-task LiDAR perception model training efficiency 5x, driving an estimated $3.8M annual savings by optimizing GT target generation, GPU memory/backprop flow, and hyperparameters
• —Independently led migration of two critical-path LiDAR perception models to a structured, reproducible PyTorch Lightning framework
• —Diagnosed bottlenecks in a multi-platform modeling library developed by 20+ engineers

Meta Reality Labs Holograms

• —Designed and prototyped a real-time neural re-rendering model for AR glasses, refining imperfect 2D inputs due to occlusion and sensor sparsity for remote calling features
• —Led cross-org collaboration (8 engineers) to unify two ML training stacks, cutting research-to-deployment lead time by 4 months

Meta Reality Labs AR Authentic Presence

• —Delivered 3 CV ML models (hand detection/keypoints/gestures, person segmentation, foot keypoints) powering 100k daily AR effects on Messenger, Facebook, and Instagram
• —Optimized inference from 20-25s native to 20-40ms on int8-quantized CPU hardware
• —Owned creation of a 400k+ sample real-world dataset

Stanford University M.S. & B.S. Computer Science