Palo Alto, California, United States
Multimodal LLM product development
• Implement the pipeline for model deployment on Qualcomm NPU
• Design a data generation pipeline for visual understanding task, improving data quality and model training.
• Train and evaluate the accuracy of VLM models for the visual understanding task
XR product development: Hand and upper body tracking system
• Enhanced the accuracy of hand detection (DetNet) and keypoint extraction (KeyNet) in XR system.
• Trained models using a combination of synthetic and real datasets for improved performance.
• Designed and implemented the real data and synthetic data generation pipeline for upper body datasets.
XR product development: SLAM system development
• Develop SLAM mapping module to achieve accurate localization for MR effects while ensuring low power/memory/cpu usage on mobile devices (including system architecture and comprehensive algorithms development/optimization, etc.)
• Design and implement relocalization evaluation tools to validate recall rates and accuracy, both with and without ground truth data
• Collaborated closely with academic institutions, including Tsinghua University and Carnegie Mellon University, on collaborative research projects
