Feb 2021 – Mar 2023 · 2 yrs 1 mo · Sunnyvale, California, United States

• Cloudnet Control Plane: Designed and developed a Google¬ Infrastructure service to program various cloud networking products like load balancers L7 ILB, L7 xLB, Health Checked services, VMs via GSLB and GFE.
• Optimized the scalability & reliability of microservice to handle Millions of requests, a 10x increase from previous implementation. This was achieved by utilizing chunk mechanism, message queues, implementing monitoring dashboards and effectively managing the concurrency of user requests. [GCE, Multi-threading]
• Implemented Adaptive Throttling via batching to reduce the number of transactions by 20x, thus effectively controlling the load on downstream systems as well as preventing load-induced outages. [C++]
• Effectively implemented sharding and replication mechanism along with leader election algorithm for microservice, resulting in a 30% improvement in load distribution and a 4-minute reduction in end-to-end latency. Recipient of Network System Health award for the same. [Python, Google Cloud SDK]
• Ensured the consistency of system by implementing consistency checker tool to compare the data between upstream and downstream persistent storages. This helped Google to maintain programming correctness.
• Centralized Endpoint ID Generation: Designed and developed a unified way to generate endpoint IDs in a centralized way, eliminating the issue of customized endpoint IDs for different regions & endpoint types.
• Developed an effective way to send those Endpoint IDs by consolidating the required parameters into a 64-bit number, reducing the memory usage. [Spanner DB (NoSQL)]
• Reviewed and Authored 20+ design documents. These discussions ensured 100% reliable Production rollouts and maintained system’s availability.

Jun 2020 – Aug 2020 · 2 mos · Seattle, Washington, United States

• Built a micro service that predicted the top contributors for anomalies given the data sources and corresponding anomaly hypothesis as input.
• I was responsible to put together the system end to end including API definitions, hosting machine learning models in docker, and rendering the output as APIs as well as visualizations

Jan 2020 – May 2020 · 4 mos · Stony Brook

• Graduate Teaching Assistant for Analysis of Algorithms course. It involved guiding the students for algorithm concepts during their assignments and course work.

Apr 2019 – May 2019 · 1 mo · Mountain View, California

Jul 2014 – Aug 2019 · 5 yrs 1 mo · Bangalore

• I have been working with Samsung R&D, Bangalore since 5 Years. The projects I have been working on include:
• Bixby 2.0 (Voice Assistant Engine)
• Developed an AI engine to execute an event in smartphone corresponding to voice command utterance. It involved implementing various Deep Learning algorithms like Convolution Neural Network for Domain Classification, Recurrent Neural Network for Intent Classification and Bi-directional Long Short-Term Memory Network for Tagging identification. Models were trained by distributed training in various cores of GPU. Also, Worked on Theano library in implementation side.
• Employed Transfer Learning with Word2Vec. Proposed Boosting solution for combining weighted sum of models at different epochs using the AdaBoost algorithm which increased accuracy by ~4%. Also, proposed intelligent data distribution techniques to improve utterance execution accuracy based on screen context.
• Skills Learnt: Natural Language Processing & Deep Learning
• Web Classification Engine in Browser: Developed an integrated Web Classification engine which included implementing an optimized feature construction using Word2Vec and classification using SVM classifier with multi-feature set model. This product was well appreciated at Samsung Best Paper Awards by NLP experts.
• Optimization of Samsung Web Browser: Improved WebGL performance by optimally handling WebGL libraries, Touch by implementing gesture configuration modules and Memory by handling multiple usages of same memory buffers in various processes.

May 2013 – Jul 2013 · 2 mos · Bengaluru, Karnataka, India

• Developed Optimized Backward Planning software on MATLAB and C sharp language to aid the employees and automate the process of selling coal stockpiles optimized in various aspects (calorific value, tonnage, etc.) using LPSolve.

May 2012 – Jul 2012 · 2 mos · Greater Kolkata Area

• In this project, I worked on offline recognition of handwritten numerals of three Indian scripts. Here, I proposed a novel approach to combine multiple MLP classifiers with the varying number of hidden nodes based on Adaboost technique. Finally, Improved the accuracy of test data by ~7% as compared to prior-art and published this work too.
• Skills Learnt: Computer Vision and Machine Learning