Fixed-Income team, developing ETL and data pipeline for quant trading strategies.

Google Search Infrastructure - Discovery

Worked with professor Nandita Vijaykumar on accelerating coordinate-based multilayer perceptron (MLP) networks. This project started as a part of the undergraduate summer research program, and is funded by the Natural Sciences and Engineering Research Council (NSERC) and the computer science department at the University of Toronto.Designed a new architecture for coordinate-based neural representations that significantly improve the training and inference speed without incurring a reduction in reconstruction accuracy.The result of this research has been published on ICLR 2022.

Worked with professor Florian Shkurti on imitating a game engine using a differentiable neural network renderer to help train an autonomous driving policy.Implemented the differentiable neural network renderer, which acts as an implicit neural representation of the scene.Modified the network to improve its accuracy on large scenes with dynamic objects.

Worked in the automotive video team, primarily worked on QNX Hypervisor (with Android GVM) on the latest Qualcomm Automotive Chipsets.Designed and developed features for video drivers that increased their overall performance and added support for new video codec formats to video drivers. Developed a complete unit testing pipeline for video components.Performed more than 40 validations on the video product requirements (decoding, encoding, video pre-processing, etc.), debugged and assessed their performance across various Qualcomm chipsets.