• Generate synthetic datasets of digital humans

• Develop & train domain adaptation models to bridge the synthetic-to-real gap (sim2real)

• Train and evaluate ML models (bio-authentification) to measure the impact of synthetic data.

• Built and deployed TCN-based models to predict facial and tongue animation from audio

• Collaborated with 3D artists to generate synthetic animation data

• Improved infrastructure and productivity with CI/CD, automation, and ML tools (Docker, Kubeflow)

• Organized a team of 5 engineers to rewrite and streamline an experimental ML codebase

• Built a modular ML pipeline to support multiple projects and engineers

• Improved data processing speed by 3× for video, audio, and geometry

• Deployed CNN-based models for face reconstruction and gaze estimation from video

• Collaborated with 3D artists to generate synthetic eye images

• Built a Python API to assemble character simulations (cloth, hair, muscle) for Avatar 2

• Extend constraint and collision systems for character simulation

• Developed Maya plugin to set up hair, cloth, and muscle simulations to enable custom tools

• Helped collaboration between technical artists and simulation researcher

More details here: https://dl.acm.org/doi/abs/10.1145/3528223.3530058

Movie: Avatar 2

• Developed a Maya plugin using Bullet for rigid body simulations, powering destruction sequences in Gravity and presented at SIGGRAPH 2014

• Facilitated collaboration between simulation artists and R&D to align tools with production and technical needs

• Improved hair collision handling with a position-based dynamics solver

• Contributed to films including Gravity, Guardians of the Galaxy, and Paddington