United States
My role lies at the intersection of embedded machine learning and code generation, spanning multiple levels of abstraction, from algorithmic design and model compression, to memory optimizations and assembly code. I am fluent in PyTorch, Python, and C, with experience in designing and building tools from scratch and in root-causing obscure HW/SW bugs. Key Accomplishments: * Designed, implemented, and maintained a suite of automatic code generation tools (downstream from compiler proper), which connects heterogeneous algorithmic components and allows customers to better utilize Ambarella’s hardware. * Implemented LLM, CV, and cryptographic algo. that were previously deemed infeasible on the CNN coprocessor. * Mentored 2 new hires and worked closely with them on complex projects, focusing on shortening our development cycle.
Augmented the existing SDK for CVflow vector processor to handle more CNN/RNN architectures.
Developed an algorithm to reconstruct 3D scenes with semantic segmentation from RGB-D video.
Streamlined internal workflows in a data-driven approach. Automated data ETL and report generation processes.