Mar 2019 – Mar 2022 · 3 yrs · India

• Project: DSP backend support for NN models (C++)
• Worked as a core team member in the Development of DSP backend support as computing unit for running neural network models for camera applications . My work involves enabling neural network caffemodels to run on DSP computeunit using C++ which results in faster execution of models used in camera systems in Samsung and received Spot Award for enablement of the computing unit.
• Worked on Quantization schemes and Quantizing float caffe models to int8 NN models. Creating Model Debugging code representing quantization ranges and weight values and graph representing it.
• Went to QIPL for supporting DSP backend singlehandedly

Sep 2018 – Feb 2019 · 5 mos · India

• Project: Samsung Neural SDK(C++)
• Worked as a core team member in the Development of Samsung Neural SDK from scratch using C++. My work involves designing the interface apis for the end users and internally linking to the internal .so libraries. It also involves encryption and decryption of the model data and creating shell script for verification purpose.

Nov 2017 – Aug 2018 · 9 mos · India

• Project: Image Classification using CNN(Python)
• In depth study of various Image classification papers and many other papers related to network architectures, different convolutions, activations, full receptive field etc. Worked on storing and retrieving tfrecord files and preprocessing.
• Completed a research work to improve accuracy of Mobilenet without increasing FLOPs or inference time. Currently made various experiments on activation functions and made different architectures using dilation, Depthwise separable convolutions and Group convolutions and achieved better performance with less FLOPs for CALTECH256 and CIFAR10 dataset using Python Tensorflow framework.
• Completed a research work to decrease FLOPS in Mobilenet without decreasing accuracy for CIFAR10 and CALTECH256.
• Running bazel benchmark for taking inference time and FLOPS on android device and as well as Linux CPU machine.
• Made experiments on shuffle layer as well .All experiments are done using Tensorflow and CUDA version 7.5 is used for training the models
• Currently working on distillation process with Teacher-student model .

Jul 2016 – Nov 2017 · 1 yr 4 mos · India

• Currently involved in Mission Critical Push to Talk project based on 3GPP for KT (South Korea) operator.
• Worked as a team member in understanding the 3GPP standards for MCPTT and operator requirements.
• Involved in development of various services related to MCPTT including Emergency calls ,Recording feature ,Upgrade and downgrade calls (audio-video, halfduplex-fullduplex),Session refresh scenario based on SIP, Floor Control based on MBCP .
• Optimization of state machines in 3GPP doc for emergency call.
• Ensuring full support for developed module for any potential bug during commercialization.
• Worked in nodejs for supporting emergency calls from server side.

Jul 2015 – Jun 2016 · 11 mos · India

• Involved in commercialization of IMS Stack version 4.0 for various operators worldwide
• including ATT, Verizon, TMO, SFR, SWC, EE, KDDI, Vodafone, and CMCC.
• Involved in implementation of operator requirements and issue fixing while working as a part of Enabler layer of IMS Stack 4.5
• Supported local testing using Sipp testing scripts.

May 2014 – Jun 2014 · 1 mo · Bengaluru Area, India

• Project1: Completed a project on human genomics using spring mvc, HTML, JavaScript and hibernate both at frontend and backend. A web designing project which has different modules which fetches data and displays results based on user entries, shows chemical structures and the compounds matching a particular structure and many other module
• Project 2: Completed a research work on SAM tools and BAM tools for compression techniques, Studied different compression techniques and sequence alignment algorithms to sequence genomic structure and also comparison of various compression techniques available in the market for storing effectively