Gunjan Mehta

Gunjan Mehta is a Senior Systems Software Engineer at NVIDIA, specializing in Deep Learning Accelerator (NVDLA) since June 2016.

In this role, he serves as the lead designer and architect of the user-mode compiler for the Deep Learning Inference Accelerator, catering to both Tegra (NVDLA) and open source (OpenDLA) frameworks.

His responsibilities include designing and implementing compilers to parse and compile Convolutional Neural Networks (CNN) from various Deep Learning frameworks, such as Caffe, TensorFlow, and ONNX, optimizing them for specific hardware architectures.

Gunjan also excels in advanced compiler features like Winograd and tiled Convolutions, layer fusions, graph optimizations, and memory management techniques for improved performance in FP16 and INT8 inference pipelines.

Prior to his current role, Gunjan held the position of Senior Systems Software Engineer at NVIDIA, Tegra-SW division, from February 2014 to June 2016.

During this period, he was actively involved in designing kernel and user-space drivers for various components like clocks, CPU, thermal control, security engine, and DVFS as a member of the modular diagnostics team.

Furthermore, he led a project team for an internal Nvidia product and co-led another project within the same period.

Gunjan Mehta's academic background includes a Master's degree in Electrical and Computer Engineering from Carnegie Mellon University, where he also worked as a Teaching Assistant in the Department of Electrical and Computer Engineering during the fall of 2013.

At CMU, Gunjan contributed to several notable projects, including a Hybrid Cloud File System, Distributed Cab Service System, File System Check (FSCK), designing a mini-kernel for an ARM processor, and the Smart Elevator System--all focused on distributed embedded systems.

Gunjan's software skills cover languages like C and C++, experience with assembly languages such as x86 and ARM, and proficiency in working with development platforms like Linux and Android, along with tools like Eclipse, gdb, Perforce, and git.

Gunjan Mehta's educational journey includes a Bachelor of Engineering in Electronics and Telecommunications Engineering from VESIT and initial schooling at Smt. M. D. Bhatia English Medium High School.

Throughout his career, Gunjan has demonstrated expertise in system software engineering, deep learning acceleration, compiler design, kernel development, and embedded systems programming.