Mar 2019 – Jul 2019 · 4 mos · Bangalore

Oct 2018 – Dec 2018 · 2 mos

• GPU Performance Prediction using Multi Layer Perceptron(MLP) : Developed a MLP based classifier which maps CUDA/DirectX applications into simpler lower level work- loads to estimate the performance of applications using lower-level workloads.
• Reduced over-fitting in the MLP through data augmentation using Auxiliary Classifier Generative Adversarial Networks (ACGAN).
• Benefited hardware engineers to estimate performance of applications early on thereby facilitating shorter chip development cycle.
• All the deep learning programs were written in python using tensorflow library, while the scripts were in perl.

Jul 2015 – Sep 2018 · 3 yrs 2 mos

• Compute Trace Library : Part of a team which developed a library which enables devel- opment of parallel compute workloads for the GPU.
• It was an internal representation of the CUDA driver. It took a CUDA program as input and produced
• the information transferred from CPU to GPU.
• This helped HW engineers to write CUDA programs and run on software simulations in early stages
• of chip development.
• The library was developed in CPP with a python interface as well.
• Shader Substitution : Developed a tool which substitutes a shader in a parallel GPU workload. Shaders are programs that run in parallel on multiple GPU cores. This tool was written in CPP and perl for both kind of interfaces.
• Trace Disassembler and Assembler : Developed assembler and disassembler for workload binaries of the GPU. These tools get used by the HW engineers to view contents and edit workloads in development of next generation of GPU. These components were written in CPP.

May 2013 – Jul 2013 · 2 mos · Pune

• While at IISER Pune, India , I worked on implementation of an attack on A5/1 cipher which is the primary encryption technique in GSM mobile phones.I implemented the project in CPP. This enabled me to use object oriented concepts of CPP and increased my interest in CPP overall.

Dec 2012 – Jan 2013 · 1 mo · Noida Area, India

• I worked on a approximate nearest neighbor search using kd tree in Open-SURF for feature matching of images. Moreover, applied Intel SSE 4.2 instructions to optimize the code. This project was integrated into a cross platform Augmented Reality app.