Sep 2017 – Feb 2025 · 7 yrs 5 mos · Bengaluru, Karnataka, India · On-site

• At Slang Labs, we’re building next generation voice interfaces for Indian languages that, we hope, will enable easier access to the internet to millions of Indians. Our first product is Slang, a Voice-to-Action mobile SDK, that enables app developers to easily add a natural voice interface to control their apps.

Jan 2014 – Jun 2017 · 3 yrs 5 mos · Menlo Park, California, United States · On-site

• I built performance tools for Android at Facebook. I continued to work on Little Eye, while also bootstrapping new tools, such as disk I/O and memory-allocation profilers from scratch. In addition, I back-ported Linux kernel tools like perf and kprobes from the mainline kernel to a couple of AOSP kernels.

Jan 2013 – Jan 2014 · 1 yr · Bengaluru, Karnataka, India · On-site

• At Little Eye Labs, we built Little Eye - a comprehensive performance analysis tools for Android applications. Little Eye provided insights to application developers about causes for excessive battery drain and suboptimal consumption of resources such as memory, CPU and network.
• Little Eye Labs was acquired by Facebook in January 2014.

Oct 2009 – Feb 2013 · 3 yrs 4 mos · Cupertino, California, United States · On-site

• At Apple, I worked on CoreImage - a high performance image processing framework for Mac OS X and iOS. My primary focus was on improving the performance of CoreImage on Mac OS, and bringing up an OpenCL based software renderer for CoreImage on iOS.

Jun 2008 – Sep 2008 · 3 mos · Marlborough, Massachusetts, United States · On-site

• I interned with AMD's GPU shader compiler team. For my internship, I implemented a linear-scan register allocator targeting an upcoming AMD GPU, with support for multiple vector channels, and global as well as local allocation.

Aug 2007 – Oct 2009 · 2 yrs 2 mos · Columbus, Ohio, United States

• My graduate studies focused on compiler optimizations for GPGPU programs. For my thesis, I developed a technique for semi-automatic unrolling of loops in GPGPU programs. This technique maximized runtime performance by maintaining an optimal trade-off between Instruction-Level-Parallelism and GPU occupancy.
• Master's thesis: Optimal Loop Unrolling for GPGPU programs (http://rave.ohiolink.edu/etdc/view?acc_num=osu1253131903)
• Source code: https://github.com/gmurthy/ptx-analyze

Jul 2004 – Aug 2007 · 3 yrs 1 mo · Bengaluru, Karnataka, India · On-site

• I worked on building dynamic program analysis (memory errors, performance, code coverage) tools for HP-UX and AIX platforms