Project Tidal has officially spun out from Alphabet's incubator, X the moonshot factory, to become TidalX Al, Inc. We are a new company using Al to transform the ocean economy using underwater computer vision models at the edge!

I work on Project Tidal from Alphabet’s startup incubator, formerly “Google X”, building our machine learning pipelines in the cloud and improving our underwater perception models across domains.

Researched surgical applications of Neural Radiance Fields (NeRF) for 3D Computer Vision Research with Intel Labs and the Advanced Robotics and Control (ARC) Lab

Presented lectures, designed and developed the curriculum and course content, and built auto-grading infrastructure for a Supervised Machine Learning Algorithms course of more than 200 students per quarter

I flew to Indonesia to work with ocean scientists on field deployment of stereo rigs, overhead drones, ROVs, and sonar for carbon sequestration modeling. We presented our initiative at the G20 Ocean conference with the World Economic Forum, and our work was detailed in the MIT Technology Review.

Built out V-Net model architecture for 3D volumetric segmentation of knee pathologies in collaboration with Stanford researchers to accelerate modern MRI diagnosis and analysis

▪ Developed and deployed deep neural networks in a two person ML team for object detection on the Saildrone Fleet of unmanned, solar, ocean drones.▪ Improved detection inference frequency from once every 16 seconds (for 4 images), to once every second (1Hz for 6 images), allowing for real-time detection ▪ Experimented with different backbones (ResNets, VGG); Modified ResNet Block skip connections to facilitate 50% pruning of the ResNet without sacrificing accuracy/F1-score (Recall @ Precision 80 = 91.84)▪ Achieved accelerated detection inference on drones over the previous model by quantizing and pruning (structured) 350 GFLOPs in the ResNet conv layers and FPN to fully optimize for embedded devices▪ Expanded the data augmentation pipeline to include “bag of tricks” from YOLOv4 and YOLOv5 to further improve our RetinaNet

▪ Engineered a Wildfire Smoke Detection Model to be deployed at high-altitude weather stations across California to automatically detect wildfires shortly upon ignition -- an AI/ML solution to the accelerating wildfire crisis for emergency response▪ Developed state-of-the-art Deep Learning models on a cluster for object detection and segmentation, such as Faster R-CNN and Mask R-CNN, using PyTorch and Python.▪ Automated video preprocessing pipelines for training and performance evaluation of the model, frame-by-frame, for hundreds of videos

▪ Facilitated Neuroscience Researchers, as part of Barack Obama’s BRAIN Initiative, towards creating the first taxonomy of every cell in the mammalian brain ▪ Built efficient data pipelines to process RNA-seq, methylation and chromatid-acc data for unsupervised machine learning