Jonathan He Email & Phone Number

United States

United States

San Diego, California, United States

• Worked in a team of 5 for Acoustic Bird Species Identification Software for the San Diego Zoo
• Evaluated model performance across over 130 bird species on over 100K recordings against the ground truth; identified the model’s weakness in detecting temporal patterns in bird calls
• Fine-tuned the Audio Spectrum Transformer on a 200GB dataset of over 400K recordings; implemented methods such as DeepSpeed integration and mixed-precision training to speed up training by 3x
• Conducted hyperparameter sweep for optimal batch size and learning rate, achieving AUC of 0.956
• Collaborated on developing a Windows-based software for visualizing, labeling, and verifying audio recordings

• Developed air vehicles algorithms to film fast-moving objects through object detection and pose reconstruction
• Implemented OpenPose network for monocular RGB pose estimation, combined with LSTM for trail prediction, achieving 5ms reaction time under parallel processing

• Developed a deep-learning feature detection algorithm in LC-MS proteomics
• Integrated custom Faster-RCNN object detection network with in-house sliding window technique to systemize feature detection with visual approach, increasing detection rate by 35% and precision by 16%

• Built ROS multithreaded framework for drones to follow trails of aruco tags in a team of 4
• Designed aruco-based localization algorithm through matrix computations for six dimensions of movement
• Developed a real-time stepwise motion control algorithm through PID and coordinate tracking

Bachelor'S Degree, Computer Science

Early College, Computer Engineering