Dec 2021 – Mar 2023 · 1 yr 3 mos

Sep 2019 – Dec 2021 · 2 yrs 3 mos

• Developed machine learning models to track faces in real-time enabling face sticker experiences in video calls using Alexa devices. The feature increased customer engagement by 18%.
• Developed a face detection model to detect location of faces in a video stream. The model had much higher accuracy on
• different skin-tones, gender, age groups etc. compared to SOTA real time detectors.
• Developed a dense 3D Face Alignment model to track facial pose and facial features such as eye, lip movements etc.
• Proposed an adversarial technique to preserve privacy in visual images. The model learns a segmentation mask which reveals
• parts of the image that are necessary for a downstream task. Publication accepted at FG 2021.
• Studied the bias in performance of SOTA face detection models for different age groups, gender skin-tone etc. Proposed a
• novel loss function to de-bias features for detection models and improve accuracy. Publication accepted at AIES 2022.

Jan 2019 – May 2019 · 4 mos · Greater Boston Area

• Worked with the wake word detection team in the Alexa division.
• Researched, proposed and developed on-device models to estimate the rate of wake-up attempts falsely rejected by the wake-word detection model.

May 2018 – Jan 2019 · 8 mos · Palo Alto, California

• Proposed, implemented and trained end-to-end deep learning models that directly consume point cloud data.
• The methodology produced state-of-the-art results for the tasks of 3D objects classification, semantic segmentation and instance segmentation on an internal dataset.
• Proposed an approximation for efficient 3D pooling in the model reducing the inference time of models by 50%.
• Implemented different Convolution, Pooling operations in OpenCL for portable deployment on different platforms such as AMD, Nvidia, Intel of the above models at inference time.

Sep 2017 – Sep 2018 · 1 yr · Bryan/College Station, Texas Area

• Graph Data Analytics.
• Present work includes generating low-dimension embedding for social networks.
• Adapting Convolution Neural Networks on non-euclidean data such as graphs.

Sep 2016 – Jul 2017 · 10 mos · Singapore

• I worked under the supervision of Dr. Gan Woon Seng at the DSP Lab. Implemented a Multi-channel feedforward active noise control algorithm for real-time in-situ deployment using Verilog HDL on Xilinx’s Virtex-7 platform.

Jul 2015 – Aug 2016 · 1 yr 1 mo · Bengaluru Area, India

• Software developer in the Application Common team developing and maintaining the framework and services used by over 10 teams to develop computed tomography (CT Scanner) image post processing applications. Acquired development experience in C# and C++ as well as proficiency in Object Oriented Modelling and Agile Software development with Scrum framework.

Jul 2011 – Aug 2014 · 3 yrs 1 mo · Delhi Area, India

• Embedded Systems :- Responsible for setting up the entire hardware stack for the AUV. Interfacing different sensors such as Inertial Measurement Unit and Pressure Sensor with a Single Board Computer (SBC).
• Control System:- Developed a robust control algorithm to maintain the orientation and depth of the AUV.
• Led the team and represented the university at AUVSI Robosub 2014, San Diego and NIOT SAVE 2014, Chennai.
• Was selected for R&D grant from ONGC for AUV project.