Fremont, California, United States
Hand Tracking on AR/VR technology https://www.oculus.com/blog/introducing-hand-tracking-on-oculus-quest-bringing-your-real-hands-into-vr/
I worked on infrastructure to process incoming AdSense accounts as well as training and productionizing risk models to reduce bad spend for advertisers. Ownership of Approvals System (Dups) •Primary owner of AdSpam Risk’s approvals system, a complex system of multiple components that checks that incoming accounts are not duplicates of existing accounts •Created canary instance that runs the newly checked in packages and diffs the results to an identical run in production, ensuring that the production pipeline would not get affected by any binary update that could bring down the pipeline. Since launch, there has been no outages caused by broken packages •Created monitoring system to track application flow, signal ingestion, and decision counts •Maintain stability of pipeline by issuing migrations to new systems and fixes during any production outage •Designed pipeline to integrate ML models in parallel to each other and the existing legacy system AdMob Risk Propagation •Developed basic smearing heuristic to propagate risk scores across account similarity graph Signup Risk Model •Feedforward neural network Keras model predicting risk of incoming accounts to reduce bad spend Throttling Risk Models •Created the infrastructure for risk models; connected models to real time traffic annotators to throttle ad serving •Launched notifications to help publish risk model assessment notes providing clarity to publishers on the enforcement
I developed a tool to help engineers transform SQL queries into Spark queries to pull in large amounts of data efficiently. SQL Support for Complex Spark Queries: •Provides the same expressiveness found in Hadoop/Hive for more the powerful Spark •Presents UI for user to enter in SQL query, parsed that query and created UDAF functions for Spark to execute •Automates the choosing of the optimal parquet to use for a query given a date range and columns of interest
-- iDevice Camera Streaming App: -Utilized C libraries with Swift with bridging headers - Drove hardware-level compression to convert camera buffers into elementary stream - Processed and delivered data via FASPSTREAM to receiver for near-instant real time streaming -- Performance/Speed Test App: -Tool compare mobile FASP transfers to other transfer methods such as HTTP, - Dynamically creates files of user selected size and erases the file after use - Allows user to specify a FASP transfer rate - Ran basic data analysis to show trends on different bandwidths/providers - Created a simple server in Python to receive HTTP POST requests -- Photo Picker: - Created a CocoaPod to assist the transition of all pre-iOS 8 apps to iOS 8 - Fixed up consistency issues and allows user to fetch asset information easily
- Labs: Assisted in answering student questions and checked off completed labs - Discussions: Led discussions, presented steps and solutions to problems, answered student questions, and lectured on core concepts - Office Hours: Helped run office hours for introductory courses