Promoted 4× in 18 months (Software Engineer → Perception Lead → Perception & Autonomy Lead → Software Lead); now direct 5–10 engineers across the AUV's vision, SLAM, planning, and mission-control stack (ROS 2, NVIDIA Jetson). Migrated the inherited 10,000+ line ROS 1 codebase to ROS 2, more than doubled underwater detection accuracy (mAP 0.40 → 0.90) by building the dataset and training YOLO + RT-DETR compiled to TensorRT (FP16/INT8) for real-time on-vehicle inference, delivered GTSAM-based SLAM fusing stereo, IMU, DVL, and depth (every landmark under 0.3 m error on rosbag replay), and root-caused the AUV's sluggish surge from 75 GB of wet-test data to a thrust clamp set at 12 N instead of 30 N — then retuned the GPU-parallel MPPI controller (CUDA/C++).
PythonC++CUDAROS 2PyTorchTensorRTGTSAM