AWS Artificial Intelligence.Working on AI solutions for cloud computing.

Video machine learning inference platform. A distributed synchronous system that categorizes all video and audio (both Live and VOD) uploaded to Meta, while detecting offensive/harmful content. Greatly increased media discoverability and ad matching across all Meta platforms. Greatly reduced harmful content uploaded to all Meta platforms. Worked from end to end (from video decoding, to machine learning predictor to data storage and callbacks).Backend storage temperature management system. Automatically rotates data between hot and cold storage depending on access rate and data age. Struck a balance between storage cost and access speed. The amount of data dealt was in the order of petabytes per year.Realtime Automatic Speech Recognition model. Collected data, tuned and trained speech model based on the wav2letter paper by Facebook AI Research. Able to transcribe live English videos with a 3 second delay. Collected data and trained language model to work with the speech model. Asynchronous video and audio inference platform. For bulk/offline processing that runs on Apache Spark. Designed to be easily configurable and runnable. Does not affect production/synchronous bandwidth during peak hours.Integrated framework that automatically canaried new machine learning models onto existing infrastructure. Helped client teams to safely onboard and deploy new machine learning models. Mitigated potential model regressions.Develop, train and deploy machine learning models that process video and audio using PyTorch and Caffe2.

In the eCommerce Risk team that assessed financial risk for Expedia online payment platform.Devised a service that used pattern recognition to detect fraudulent transactions.Increased accuracy of the existing automated fraud detection system.Decreased amount of transactions that went to human auditors.

Core team member. Initialted the Polymer Film with Integrated Servo project.