Production, implementation, & acceleration of Pixel face authentication.

• Specialized lab for research in robotics, automation, AI, and deep learning.• Worked on revolutionizing the capabilities of surgical robots through combining deep reinforcement learning, novel perception pipelines, and temporally-aware neural networks.• Publications in ICRA, ISMR, and RA-L.

• Lead design and integration of Silicon Edge TPU into third-party pipelines. • Allows automatic delegation of ML inference in existing pipelines to on-device accelerators, providing better performance and higher energy efficiency.• Increased frame-rate of popular MediaPipe face-tracking algorithm by 35x by utilizing on-device Edge TPU.• Opens door for more complex and computationally intensive MediaPipe models to run in realtime on mobile devices. • Profiled and proposed opportunities for optimization to reduce overall latency in ML pipelines mixed with GPU, CPU, and TPU.

• 76 out of 36,828 applicants chosen for this program.• Lead research, development, and implementation of a protocol that allows for local file creation of cloud-based documents via third party applications.• Drove feature rollout to hundreds of millions of Dropbox users.• New protocol improved local cloud-doc creation time by up to 3600x, reducing creation and accessibility latency from up to one hour to less than a second.• This protocol is patent pending: creation of instantly accessible cloud documents from a user's local filesystem was previously impossible. • Enhanced robustness of our asynchronous file handler to reliably create and repair documents from a wider range of file states.• Winner of company culture award for building and installing new safety mechanism into the staircases as a side project.

• Lead developer for professor-led research team attempting to reveal new information about the location, size, and surfaces of small Uranian satellites.• Supervised by Imke de Pater, current UC Berkeley professor and former head of UC Berkeley Department of Astronomy, as well as Mark Showalter, Principal Investigator of NASA's Planetary Data System.• Designed algorithms and augmented state-of-the-art photometry techniques. • Lead author on upcoming publication on significant astronomical discoveries, namely field-leading limits and the most precise photometry of Mab to date.

• Lab assistant and office hour tutor. • Catalyze student problem-solving intuitions in one-on-one interactions.• Assist students in project development.• Class focused on principles of software engineering, such as storage management, abstract data types, asymptotic analysis, and tradeoffs in time and memory; also implementation of fundamental dynamic data structures, such as Linked Lists, Trees, HashMaps, Heaps, Queues, Graphs, sorting & searching algorithms.

• Augmented camera tracking confidence algorithm with ML-driven solution to improve accuracy by creating advanced decision metrics and leveraging spatial distributions in order to make real-time broadcast decisions for the NFL First-Down-Line; increased on-air decision accuracy by 45%.• Created new template matching scheme based on 2018 CVPR concept; matched templates 1.25x faster allowing for an increase in data throughput.• Developed framework for new, comprehensive texturing technique with automatic outlier removal to enhance robustness of tracking applications.• Above modifications adopted by multiple football broadcast companies (ESPN, FOX, CBS, etc.) with over a billion total annual viewers. • Played an integral role in the development of a visual odometry framework • Helped resolve key differences between real-time applications and offline testing modules.

• Over 250,000,000 people viewed an original design created for the social media app Snapchat; nominated by Odyssey as Top 25 Geofilter in America.