May 2018 – Aug 2018 · 3 mos · Greater Minneapolis-St. Paul Area

• Worked on optimizing Big Data Analytics Suite (BDAS) for Cray HPC systems. Re-designed big data distributed algorithms (namely K-Means, SVM and PCA) for GPUs nodes boosting the average performance by a factor of 20-70 times while maintaining scaling efficiency >90%.
• Developed distributed training framework for Pytorch for Cray super-computers. The goal was to make fast and easy to use add-on for distributed deep-learning enabling users to train model across multiple CPU or CPU-GPU nodes. The library is developed in C++ for maximum performance and achieving scaling efficiency >90%. Another goal was to address the issues associated with massive parallel training of model (especially when batch_size*num_nodes=training_data).
• Addressed the performance issues associated with FFTW, an open-source FFT library for HPC systems, by analyzing the code at a low level (cache-misses, CPU architecture, and memory hierarchy), and gained 5% improvement in performance after prefetching some data effectively.

Aug 2017 – May 2018 · 9 mos · Bryan/College Station, Texas Area

• Artificial Intelligence for Online Tutors
• Research thesis on understanding and recognizing facial expression using deep neural nets in conjugation with conventional image processing methods, utilizing the work in understanding user (learner’s) sentiments for developing intelligent online tutors.
• Generated database of subjects with different facial expression by feeding the web collected videos to a pre-trained Inception Resnet model cascaded with an unsupervised KNN model, enables us to build a large database for complicated facial expression.
• Developed a more robust deep neural model by forcing the network to learn important features and fine-tuning hyperparameter results in increasing understanding accuracy by 15%.
• Skills: Python, TensorFlow & Qt
• Self-Driving car steering modeling
• Built an automatic car steering model by training a 6-layer CNN model with Udacity self-driving car database, achieved a testing accuracy of 92.7\% and finally tested on an emulator for real-time working.
• Skills: Python, TensorFlow & OpenCV
• ‘Books for Aggies’ Library website design
• Designed University Library website for sharing book among students on Ruby on Rails framework
• Followed Agile Development Process, performed client meeting at each iteration and tracked each user story using Pivotal Tracker.
• Performed Test Driven Development and Behavior Driven Development and achieved 92% code coverage.
• Skills: Ruby, Ruby on Rails, HTML, CSS
• Assembly code Indirect Branch Predictor
• Implemented Indirect Branch Predictor by registering history of the geometric length of the incoming instruction, resulting in only 0.093 mispredictions per 1000 instructions (MPKI) with SPEC CPU 2006 benchmarks.
• Skills: C++
• Read Write Partition based cache replacement policy
• Implemented Read Write Partition (RWP) based cache replacement policy in C++ which achieved 4.8% overall IPC geometric mean speedup over baseline LRU when run on CPU spec benchmarks.
• Skills: C++

Aug 2015 – Jul 2017 · 1 yr 11 mos · Chandipur, India

• Real-Time Object Tracking:
• Reduction of object detection time in a 640x480 frame to 8ms by introducing two-stage coarse and fine parallel pipeline in tracking, making real-time tracking 50% faster and 20% more accurate.
• Passive Lift-off Time Detection System (PLTDS):
• Developed real-time event sequence detection with multiple cameras by discovering and implementing improved feature extraction algorithms, making PLTDS able to visually identify ongoing launching events in 10ms time-frame.

May 2014 – Jul 2014 · 2 mos · New Delhi Area, India

• Project on fault tolerance on 400KV Sub-Stations
• Simulated the effect of various faults and their effect on sub-station.
• Designed more efficient algorithm in pre-detecting fault and model the algorithm in MATLAB