Dec 2021 – Dec 2023 · 2 yrs · Redmond, Washington, United States · On-site

• Azure Stack: Azure services on-premises
• Developed and implemented methods for registering edge devices with Azure, enabling billing, OS activation, and other services. Collaborated with the Azure team to improve existing tools, including VS Code extensions, GitHub actions, and PowerShell SDK for Azure Stack compatibility and consistent user experience.

Jul 2019 – Dec 2021 · 2 yrs 5 mos · Redmond, Washington, United States · On-site

• Azure Stack

Jun 2018 – Aug 2018 · 2 mos · Greater Seattle Area · On-site

• Automated Time Series Outlier Detection and Forecasting
• Developed ARIMA (Autoregressive Integrated Moving Average) and ETS (Error, Trend and Seasonality) models for time series outliers detection and correction and forecasting.
• Automated the hyper-parameters estimation of ARIMA and ETS models for any time series data.

Sep 2017 – May 2019 · 1 yr 8 mos · West Lafayette, Indiana · On-site

• Failure Analysis in High Performance Computing Systems
• Analyzed extensive system usage and failure data from two centrally administered computing clusters at two Tier 1 US research universities, comprising a total of 3.0M and 2.2M jobs from the two clusters, respectively.
• Conducted analysis of different types of data such as node failures and recovery, job failures, and job resource usages to bring out important insights into how the clusters behave and implications for how they can be managed more effectively.
• Key findings include the need for event-driven application-level checkpointing, the varying effects of resource pressure on different types of jobs, the use of user historical resource usage to predict resource usage of jobs currently in queue, and the observation of job failures due to resource contention at levels much less than available capacity.
• Approximate Video Processing
• Developed histogram based Scene Change Detector and edge data based Image Complexity Categorizer.

Mar 2017 – Aug 2017 · 5 mos · On-site

• Energy Management System (EMS) in Smart-Buildings
• Developed a Recurrent Neural Network model to predict optimal start time for HVAC systems and another algorithm to derive optimal air flow rate and set-point temperature for HVAC systems based on occupancy, temperature, humidity, and people preferences, as part of my work on Energy Management System (EMS) in Smart-Buildings.
• Proposed a novel approach to create zones dynamically and implemented a caching system based on ehcache, resulting in improved overall performance of the EMS.
• Contributed to the successful pilot testing of the EMS solution at Samsung Office in Poland, which led to a reduction of more than 20% in the building's overall energy consumption.

Jul 2015 – Mar 2017 · 1 yr 8 mos · On-site

• RGB-D Video Segmentation
• Developed an RGB-D Video Segmentation system that uses foreground-background information from color and depth to assign foreground probabilities for each pixel. The color data is modelled using these probabilities, and energy minimization is performed in a graph cut framework using the resulting models.
• Introduced novel approaches to reduce model computation time, including temporal reuse of the color model, and to smooth the transition of boundaries across frames using localized motion-based compensation. Implemented a skeleton-based post-processing step to remove non-human parts from the foreground and a fast guided filter to improve the overall look of the segmentation.
• Achieved very high quality segmentation (F score 0.975) in almost real-time (38ms per frame) without using GPU.

May 2014 – Jul 2014 · 2 mos · Bangalore · On-site

• Enhanced Face Recognition System
• Developed and integrated pre-processing techniques like Mirror Image Superposition, Histogram Equalization, and Gaussian Filtering to enhance the accuracy of the Face Recognition (FR) System based on Independent Component Analysis.
• Proposed a novel classification scheme using the mode as a criterion, improving recognition rate by around 14% from 82% to 96%.