ML Planner

• Led end-to-end development of Traffic Light Detector System, from model architecture design, distributed training, data workflows to fleet deployment for the 2020 Tokyo Olympics showcase. This system secured 5+ patents.

• Researched end-to-end Semi-Supervised Learning paradigm for Vision Detectors, achieving better performance over supervised baselines using only 1% labeled data on a petabyte-scale dataset.

• Built TRI’s Autonomous MLOps ecosystem, streamlined data workflows, codified and open-sourced Dataset Governance Policy (DGP), and hyperparameters tuning to standardize ML lifecycle across global Toyota teams and external partners.

• Standardized ML development by architecting a generic framework for TensorFlow, PyTorch, and Caffe, significantly reducing time-to-production.

• Improved lane marking detection accuracy by 10% by designing ML models using LiDAR and image sensor fusion, and streamlining data processing to boost ML flywheel.

• Achieved 94% grasping success by developing a 3D object pose estimation algorithm optimized for sub-millimeter precision.

• Reduced robotic arm positional errors 10x by optimizing calibration algorithms.

Maintained Helicopter Sikorsky S-70C(M) and ensured it was safe for flight as part of military service.