Somshubra Majumdar
Senior Deep Learning Engineer at NVIDIA
San Jose, California, United States
Summary
🤩
Rockstar🎓
Top SchoolSomshubra Majumdar is a Senior Deep Learning Engineer with 12 years of experience, currently building scalable ASR and generative AI systems at NVIDIA from San Jose. He blends research and production: reproducing and improving papers like Neural Style Transfer in Keras while contributing to high-profile open-source projects such as NVIDIA NeMo and keras-contrib (DenseNet, NASNet, GroupNormalization). His work spans model design, decoder improvements (beam search for RNN-T), data pipeline optimizations, and integration efforts (Hugging Face datasets, DALI) that reduce training costs at scale. Earlier research and entrepreneurial experience includes algorithmic contributions in parallel and meta-sorting (AdaSort) and founding a software solutions startup, reflecting a pragmatic focus on both novel algorithms and robust implementation.
13 years of coding experience
3 years of employment as a software developer
Masters Computer Science, Masters Computer Science at University of Illinois ChicagoBachelor of Science (B.Sc.) Computer Engineering, Bachelor of Science (B.Sc.) Computer Engineering at Bhavans CollegeJamnabai Narsee Monjee SchoolBachelor of Science (BSc) Computer Engineering, Bachelor of Science (BSc) Computer Engineering at Dwarkadas J. Sanghvi College of Engineering
English, Hindi, Japanese
Github Skills (30)
pytorch10
convolutional-neural-networks10
preprocessing10
python10
load-data10
preprocess10
machine-learning10
voice-recognition10
automatic-speech-recognition10
super-resolution10
data-preprocessing10
keras10
data-loading10
densenet10
dataprep10
Programming languages (7)
Github contributions (5)
titu1994/Neural-Style-Transfer
May 2016 - Apr 2022
Keras Implementation of Neural Style Transfer from the paper "A Neural Algorithm of Artistic Style" (http://arxiv.org/abs/1508.06576) in Keras 2.0+
Role in this project:
Back-end Developer
Back-end DeveloperContributions:35 releases, 297 commits, 13 PRs in 6 years
Contributions summary:Somshubra primarily worked on improving the core functionality of the neural style transfer script, with the primary focus on adding the ability to initialize the output image with either the content image or noise. Further improvements included enhanced logging to the Windows Helper executable, and adding a MRF Network model. The changes show modifications to parameters that influence the visual output of the style transfer process, which also includes the addition of pooling types to the script.
kerasneural-style-transfer
titu1994/neural-image-assessment
Jan 2018 - Jul 2019
Implementation of NIMA: Neural Image Assessment in Keras
Role in this project:ML Engineer
ML EngineerContributions:5 releases, 37 commits, 6 PRs in 1 year 6 months
Contributions summary:Somshubra primarily contributed to the implementation and refinement of a neural image assessment model using Keras. They focused on optimizing the data loading pipeline, including batch size reduction and data augmentation techniques. The user also made adjustments to the loss function and weight loading, as well as adding support for different model architectures like NASNet and Inception ResNet v2.
kerastensorflowava-datasetneural-image-assessment
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.