I am passionate about designing computer languages and building compilers that make it easier for programmers to write high-performance software. I have a decade of experience working on a wide variety of production-grade and research compilers. I have also published at multiple top programming language and compiler conferences.
Currently, I am a senior software engineer at Google, where I work on the XLA TPU compiler and help drive bring-ups of the compiler for new generations of TPUs. Before that, I completed my PhD in computer science at MIT under the supervision of Prof. Saman Amarasinghe. I was a primary contributor to the TACO sparse tensor algebra compiler, and I developed new techniques to automatically generate efficient code for computing on sparse tensors that are stored in disparate data layouts.
A compiler that generates fast code to perform sparse tensor algebra computations. I extended TACO so that it can emit code to efficiently compute with sparse tensors that are stored in a wide range of specialized data structures. I also helped maintain many other parts of the compiler, as well as developed a web interface that lets users try out TACO without having to install it on their own machines.
A domain-specific programming language for computing on sparse systems using linear algebra. I implemented a new front end for the Simit compiler that was significantly more user-friendly and robust than the original one. This new front end was later also adapted for use in the compiler for GraphIt, a programming language for graph computations.