Apr 2025 – May 2025 · 1 mo

Jun 2023 – Apr 2025 · 1 yr 10 mos

• Learning and Perception for humanoid robot Digit, agility's flagship robot, targeted towards warehouse automation.
• I work on deep learning based perception pipelines to help digit detect and localise important elements in its surroundings (including dataset curation, model training and evaluation), as well as develop innovative algorithms that run real-time on the robot to process these neural network evaluation outputs into meaningful action priors

Jan 2023 – Jun 2023 · 5 mos

• Graduate level Class ESE650 - Learning for Robotics, Spring 2023

Sep 2022 – Dec 2022 · 3 mos

• Computation Motion Planning course, Robotics Specialisation by University of Pennsylvania, on Coursera - Fall 2022

Aug 2022 – Dec 2022 · 4 mos

• CIS 519 Applied Machine Learning Fall 2022

Feb 2022 – May 2022 · 3 mos · Philadelphia, Pennsylvania, United States

• In the mentorship of Dr Oleh Rybkin
• Investigating model based Reinforcement Learning for efficient and planned exploration of unknown simulation environments.
• Analysing latent disagreement based RL strategies along with Unsupervised Representation Learning for high dimensional pixel input data to enable zero-shot or few-shot transfer at test time.

Jan 2022 – Jun 2022 · 5 mos · Philadelphia, Pennsylvania, United States

• CIS519 Applied Machine Learning, Spring 2022

May 2022 – Aug 2022 · 3 mos · United States

• Autonomous Driver Assist System (ADAS) Team, Autonomy R&D Division
• Worked on Hardware aware Neural Architecture search for deep vision backbones, optimisation and ML Systems.
• 1) Starting from Efficientnet-b0, which is a well optimised but hardware agnostic backbone neural network architecture for perception tasks, conducted Neural Architecture Search (NAS) to find neural network architectures that are highly optimised to run on Qualcomm's hardware and compiler, while achieving better accuracy than baseline.
• 2) Surveyed state-of-the-art research in efficient deep neural network architectures and designed a novel search space with proper efficacy analysis in place. Our results enveloped the fastest models in Qualcomm's Auto Model Zoo today.
• 3) Implemented the end-to-end deep learning pipeline in Pytorch with Hardware in loop using Google's Vertex NAS Platform and increased data-loading pipeline speed by 5x under cloud based distributed training regime.

Oct 2020 – Sep 2021 · 11 mos · Bengaluru, Karnataka, India

• Built data flow pipeline features for the large scale distributed real time operations and analytics platform, vRealize Network Insights (vRNI) that supports next generation datacenters with enterprise grade reliability.

Jul 2018 – Oct 2020 · 2 yrs 3 mos · Karnataka, India

• Built the architecture, design, and solution implementation for device management software on Junos (Juniper OS) specific to Junos Node Slicing Project.
• Junos Node Slicing enables service providers and large enterprises to create a network infrastructure that consolidates multiple routing functions into a single physical device. It helps leverage the benefits of virtualization without compromising on performance, by enabling the convergence of multiple services on a single physical infrastructure.

May 2017 – Jul 2017 · 2 mos · Dallas, Texas

• Internship project in Natural Language Processing

Aug 2016 – Apr 2017 · 8 mos · IIT Kanpur

• International autonomous Ground Vehicle Competition
• Built and tested numerous computer vision algorithms to detect chalk lanes on grassy surface and avoid obstacles using sensors like LIDAR and Camera.
• Created a system for online mapping of surroundings (lanes and obstacles) using continuous LIDAR and camera data stream.