Research Scientist | Understanding Intelligence
AI is analyzing your overall score…
Identifying your key strengths…
Evaluating your skill match against the job requirements…
Assessing your cultural and operational fit
I’m interested in building machines with animal-like intelligence. To this end, I aim to understand computational principles that could enable agents such as robots to continually learn, adapt, develop, and improve throughout their lives. I have worked extensively on online reinforcement learning, imitation learning, and sim-to-real approaches for robotic manipulation.
University of Alberta
Doctor of Philosophy - PhD, Statistical Machine Learning
January 1, 2020 – January 1, 2025
University of Alberta
Master’s Degree, Computing Science
January 1, 2015 – January 1, 2017
National Institute of Technology, Tiruchirappalli
Bachelor of Technology (B.Tech.), Instrumentation and Control Engineering
January 1, 2011 – January 1, 2015
Zyphra
Member of Technical Staff
February 1, 2026 – Present
San Francisco Bay Area · On-site
Sanctuary AI
Research Intern: Reinforcement/Imitation Learning
May 1, 2025 – September 1, 2025
Vancouver, British Columbia, Canada · On-site
University of Freiburg
Visiting Researcher | DAAD Scholar
March 1, 2023 – June 1, 2023
Freiburg, Baden-Württemberg, Germany · On-site
University of Alberta
Doctoral Student
September 1, 2020 – Present
Edmonton, Alberta, Canada
Kindred.ai
Machine Learning Researcher
June 1, 2019 – August 1, 2020
Greater Toronto Area, Canada
Kindred.ai
Artificial Intelligence (AI) Engineer
September 1, 2017 – May 1, 2019
Greater Toronto Area, Canada
University of Alberta
Graduate Research Assistant
May 1, 2016 – August 1, 2017
University of Alberta
Masters Candidate
September 1, 2015 – August 1, 2017
University of Alberta
Graduate Teaching Assistant
September 1, 2015 – April 1, 2016
IIIT Hyderabad
Research Intern
May 1, 2014 – July 1, 2014
Hyderabad Area, India
Festember
Head of Treasury
April 1, 2014 – May 1, 2015
India
Pragyan Guest Lectures
Manager
October 1, 2013 – April 1, 2014
India
Spider (R&D Club)
Researcher
May 1, 2013 – March 1, 2015
Trichy
Connect NIT
Web Developer
July 1, 2012 – April 1, 2013
Trichy
NITTFEST'13
Coordinator
July 1, 2012 – April 1, 2013
Trichy
Pragyan Crossfire
Coordinator
July 1, 2012 – May 1, 2013
Trichy
NIT Trichy
Deputy Manager, Festember Marketing & Media Relations
April 1, 2012 – April 1, 2014
Trichy, Tamil Nadu
Learning from Demonstration: Teaching a Myoelectric Prosthesis using an intact Limb
May 1, 2016 – Present
Prosthetic arms should restore and extend the capabilities of someone with an amputation. They should move naturally and be able to perform elegant, coordinated movements like a biological arm. However control of modern day prostheses is non-intuitive and taxing; it does not give rise to graceful, synergistic movements. Using actor-critic reinforcement learning, we present a method that allows someone with an amputation to use their non-amputated arm to teach their prosthetic arm how to move through a wide range of coordinated motions and grasp patterns. We evaluate our method during the myo-electric control of a multi-joint robot arm and demonstrate that the robot arm is able to learn a control policy for a particular sequence of gestures and movements involving 3 Degrees of Freedom (DoFs).
Deep Reinforcement Learning for Hex
March 1, 2016 – May 1, 2016
DeepMind’s recent spectacular success in using deep convolutional neural nets and machine learning to build superhuman level agents — e.g. for Atari games via deep Q-learning and for the game of Go via Reinforcement Learning — raises many questions, including to what extent these methods will succeed in other domains. In this paper we consider DQL for the game of Hex: after supervised initializing, we use selfplay to train NeuroHex, an 11-layer CNN that plays Hex on the 13 × 13 board. Hex is the classic two-player alternate-turn stone placement game played on a rhombus of hexagonal cells in which the winner is whomever connects their two opposing sides. Despite the large action and state space, our system trains a Q-network capable of strong lay with no search. After two weeks of Q-learning, NeuroHex achieves win-rates of 20.4% as first layer and 2.1% as second player against a 1-second/move version of MoHex, the current ICGA olympiad Hex champion. Our data suggests further improvement might be possible with more training time.
Robot Control with Policy Gradient Actor-Critic Methods
January 1, 2016 – April 1, 2016
Reinforcement learning offers to robotics a framework and set of tools for the design of sophisticated and hard-to-engineer behaviors. Conversely, the challenges of robotic problems provide both inspiration, impact, and validation for developments in reinforcement learning. The aim of this independent study is to gain understanding of the current state-of-the-art and historical background of reinforcement learning (RL) research in policy gradient methods and actor-critic algorithms. In this article, we attempt to understand the different algorithms used to incorporate the above topics into deployed (real-world) reinforcement learning systems. We highlight both key challenges in robot reinforcement learning as well as notable successes. We discuss how contributions from Actor-Critic algorithms, Policy gradient methods, etc., handle the complexity of the domain and study the role of algorithms, representations, and prior knowledge in achieving these successes. The main objective for the experiments in this independent study is to deploy Prediction Learning methods coupled with Natural Actor Critic algorithms in real-world robots.
Multiple Sclerosis Lesion Segmentation in MRI Images
October 1, 2015 – December 1, 2015
Magnetic Resonance Imaging (MRI) is commonly used to detect Multiple Sclerosis Lesions in the Central Nervous System. The lesions are anatomically variable, thus, accurate segmentation of MS lesions is challenging. Since manual segmentation requires domain specific knowledge and is expensive both in terms of time and cost, we aim to automatically segment lesions. Our focus is on comparing classification techniques using a wide range of features in order to find the most successful method. We also attempt to combine some techniques to improve detection success. Our results indicate that Markov Random Fields with Random Forest provides the best result but there are other options useful in segmenting lesions. Interestingly, Random Forest is better in terms of computational complexity than other algorithms, especially when compared to SVM. This makes it the most practical in terms of both accuracy and speed.
Model Predictive Control for Micro Aerial Vehicle (MAV) Systems
May 1, 2014 – Present
I'm responsible for the development of an algorithm for trajectory tracking for quad rotors. The modelling and simulation of the trajectory tracking algorithm for the Quad rotor was successfully implemented. The algorithm is being currently being tested on the Parrot Ar Drone using ROS and Python coupled with CVXOPT for solving the quadratic constraints.
A Control Strategy for an Autonomous Robotic Vacuum Cleaner for Solar Panels
December 1, 2013 – January 1, 2014
I was primarily responsible for the Embedded Coding and design aspects of the robot. Abstract: Accumulation of dust on the surface of solar panels reduces the amount of radiation reaching it. This leads to loss in generated electric power and formation of hotspots which would permanently damage the solar panel. This project aims at developing an autonomous vacuum cleaning method which can be used on a regular basis to maximize the lifetime and efficiency of a solar panel. This system is implemented using two subsystems namely a Robotic Vacuum Cleaner and a Docking Station. The Robotic Vacuum Cleaner uses a two stage cleaning process to remove the dust from the solar panel. It is designed to work on inclined and slippery surfaces. A control strategy is formulated to navigate the robot in the required path using an appropriate feedback mechanism. The battery voltage of the robot is determined periodically and if it goes below a threshold, it returns to the docking station and charges itself automatically using power drawn from the solar panels. The operation of the robotic vacuum cleaner has been verified and relevant results are presented. The DC Charging circuit in the docking station is simulated in Proteus environment and is implemented in hardware. An economical, robust Robotic Vacuum Cleaner which can clean arrays of Solar panels (with or without inclination) interlinked by rails and recharge itself automatically at a docking station is designed and implemented.
Shadow Arm
July 1, 2013 – August 1, 2013
A Kinect based robotic arm that seamlessly replicates the motion of the human arm (Up to the wrist) using TI Stellaris LM4F210, Servo-Motors & OpenCV.
Efficient Fault Tolerant Routing Algorithm for New Hybrid Topology for regular NoC
May 1, 2013 – June 1, 2013
Implemented Network on Chip (NoC) concepts using Verilog HDL & developed an efficient fault tolerant algorithm for making the system robust & efficient. It was under the guidance of Prof. Dr.G. Lakshmi Narayanan at ECE Dept, NIT Trichy.
Autonomous Path Finder
February 1, 2013 – Present
Performs Colour & Shape Recognition in Matlab from live image feed & finds the shortest path required for autonomous navigation. It was controlled from the Laptop using UART communication.
In-Plant Training at TVS Sundaram Fasteners Ltd
December 1, 2012 – Present
During the training period, various practices of the organization of TVS Sundaram Fasteners Ltd was learnt. Also an insight into how sprockets, hubs & shafts are manufactured and the various tests it undergoes to be a commercially viable product was obtained.
Cricket, Hangman & Tic Tac Toe
January 1, 2011 – Present
The games were developed in C++ using graphic libraries. It uses unique user interfaces with different difficulty modes, load/save options etc.
Solar Tracker
September 1, 2010 – Present
It is a solar panel designed to orient itself toward the sun all day to make the system efficient & compact. Our method improves the efficiency of the system by 30%. It was submitted for CBSE Regional Science Exhibition
edX Honor Code Certificate for Autonomous Navigation for Flying Robots
edX
June 24, 2026 – Present
Cultural Fit Analysis
The candidate's extensive involvement in academic research, including a PhD, and roles at AI-focused companies like Kindred.ai and Sanctuary AI, indicates a strong alignment with a research-driven, innovative, and technically challenging environment. The diversity of personal projects, ranging from robotics to game development and image processing, shows a broad intellectual curiosity and a proactive approach to learning and application. The candidate's experience in various organizational roles for university festivals also suggests an ability to work in teams and manage complex initiatives, contributing to a well-rounded profile. However, the lack of explicit collaboration or team-oriented project descriptions in the technical projects makes it difficult to fully assess team collaboration style beyond organizational roles.
Soft Skills & Operational Fit
The candidate's project descriptions indicate a strong drive for research and problem-solving, often in collaborative academic settings. Roles like 'Head of Treasury' and 'Manager' for cultural/technical festivals suggest organizational and leadership capabilities. The detailed descriptions of research projects imply strong analytical and communication skills in a technical context. The focus on advanced AI/ML and robotics research suggests a fit for roles requiring innovation and deep technical contributions.