Apr 2021 – Apr 2022 · 1 yr · San Francisco Bay Area

• Designed and developed Self-Services that enabled data scientists to quickly deploy deep learning models into model inferencing service for both cloud and on-premise environments.
• Built REST APIs and python based Command Line Interface through which users can interact with Self-Services.

Aug 2018 – Mar 2021 · 2 yrs 7 mos · San Francisco Bay Area

• Developed a software platform for data scientists to train and test deep learning models using Amazon SageMaker.
• Designed and implemented a feature which allows users to evaluate their deep learning models.
• Implemented a docker container for pre-processing of the medical images and creating corresponding TensorFlow records.
• Developed training and testing capabilities for deep learning models for on-premise environment using Kubernetes.
• Added support for mask and ROI annotations which enabled training for segmentation deep learning models.
• Enhanced UI to improve user workflow on the platform using TypeScript and Angular 7 framework.
• Technologies: Python, AWS, Amazon Sagemaker, TensorFlow, Kubernetes, Helm, Docker, Angular 7, TypeScript

Jun 2017 – Mar 2018 · 9 mos · La Jolla, California

• 1) CSE 150: Introduction to Artificial Intelligence (Jan 2018 - Mar 2018)
• I was a Teaching Assistant for the course "Introduction to Artificial Intelligence: Search and Reasoning (CSE 150)" taught by Prof. Sicun Gao.
• ====================================================================
• 2) CSE 258: Web Mining and Recommender System (Sep 2017 - Dec 2017)
• I was a Teaching Assistant for the course "Web Mining and Recommender System (CSE 258)" taught by Prof. Julian McAuley. My responsibilities were as follows:
• a) Holding 3 office hours every week.
• b) Grading assignments and homework.
• c) Answering questions on Piazza.
• ====================================================================
• 3) CSE 150: Introduction to Artificial Intelligence (Jun 2017 - Aug 2017)
• I was a Teaching Assistant for the course "Introduction to Artificial Intelligence: Probabilistic Reasoning and Decision Making (CSE 150)" taught by Prof. Lawrence Saul. My responsibilities were as follows:
• a) Holding a discussion session every week.
• b) Holding 2 office hours every week.
• c) Correction of assignments and exams.

Mar 2016 – Jul 2016 · 4 mos · Bengaluru Area, India

• Background:
• Brain tissue can be classified into 3 classes called gray matter (GM), white matter (WM) and cerebrospinal fluid (CSF). Due to finite imaging resolution, the intensity level at a voxel may be due to a mixture of tissue classes. Such voxels are called 'partial volume (PV)' voxels and the phenomenon is known as 'partial volume effect (PVE)'.
• Contributions:
• Worked on PV estimation in 3D T1 weighted MR images of brain for better tissue quantification.
• After investigating state-of-the-art techniques, refined the criteria for identification of voxels with PVE.
• The results matched the ground truth with an accuracy of 92-98%.
• Technologies: C/C++, Insight Segmentation and Registration Toolkit

Jul 2014 – Feb 2016 · 1 yr 7 mos · Bengaluru Area, India

• Brain tissue classification:
• Worked on 3 class (gray matter, white matter and cerebrospinal fluid) classification of 3D T1 weighted MR images of brain.
• The task was accomplished using the concepts of Markov Random Fields, k-means and Expectation-Maximization algorithm.
• Brain MR image segmentation:
• Segmented several brain structures like Hippocampus, Amygdala, Caudate and Brain stem in 3D T1 weighted MR images of brain.
• The task involved several steps like skull stripping, rigid, affine and non-rigid registration.
• DICE similarity coefficients were computed to compare the results with state-of-the-art techniques.
• Technologies: C/C++, Insight Segmentation and Registration Toolkit

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

• Project Title: Adaptive Streaming and Scalable Transcoding
• Worked on adaptive streaming and scalable transcoding for Internet Protocol camera.
• Developed 'Scalable Transcoder' module which produced three outputs of different resolutions using Pyramidal-Decomposition. For resizing the image, I implemented bilinear, bicubic, lanczos, bspline and Gaussian blur algorithm using C++.
• Also developed 'Bandwidth Adaptation Logic' module which checked the available bandwidth at regular intervals. Based on that, it decided whether to stream low or high resolution video. HTML 5 and JavaScript were used for developing this module.

May 2012 – Jul 2012 · 2 mos · Ansan, Gyeonggi-do, Korea

• Project Title: Edge Directed Image Interpolation
• Developed an edge directed, high performance and low complexity algorithm for obtaining high resolution image from low resolution image.
• Explored fundamentals of image processing such as nearest neighbor, bilinear, bicubic interpolation, Canny, Sobel and Prewitt edge detection. MATLAB was used extensively for programming purposes.
• The proposed algorithm produced results which were comparable to state-of-the-art techniques such as New Edge-Directed Interpolation and Soft-Decision Adaptive Interpolation with 10-15% improvement in (run-time) performance.