- Analysed the performance of the RL agents in the semiconductor field that showcased significant improvements in KPIs- Built an interactive dashboard that helps the team to compare the critical KPIs between different RL agents and visualize them across time- Performed the tool modeling analysis on an open-source dataset leading to a journal paper

- Developed an MVP of fixed-wing flight stack (R&D) in less than three months, that includes: autonomous control in offboard mode of PX4 (SITL, HITL, and real-life), robust simulation with flight dynamics in Gazebo & Unity3D, decentralized collision avoidance algorithms (RVO2), and a communication module with ROS2 & ROS1 bridge - Built an in-house Ground Control Station software for complex mission planning, obstacle/geofence building, path planning, and status monitoring using PyQt5

- Built an autonomous robot (Pixie-four wheeled drive) from scratch in Nvidia Isaac SDKwith custom stereo visual odometry for localization, april tag based relocalization andobstacle avoidance in the span of two months- Achieved a 0.87 IoU score on a bean field dataset (proprietary) by training a SegNet andHoughCNet in tandem for crop row detection pipeline

Unsupervised domain adaptation for low-resource entity recognition:- Determined the efficient loss-centric method in unsupervised domain-adaptation of apre-trained transformer (BERT) for entity recognition; performance gain of 3.2 F1 score- Contributed an extensively pre-processed Enron email dataset and annotation setvaluable for retrieval and extraction testing purposes at ZyLAB

Synthetic Medical Records:- Built a generative autoencoder (s2s LSTM) tool for synthesizing Electronic Health Recordsresulting in 0 waiting-time of confidential data acquiring for testing query pipeline- Created generic to specific tunable results through tempered softmax in the outputs

Course: Image Processing, Bachelor Artificial Intelligence, 2019- Assisted in programming assignment creation and evaluation in MATLAB- Provided personal guidance with a facetime of 8 hrs/week for the students

Embodied Cognition:- Solely fabricated the perception guided grasping pipeline on the Moveit! stack of ROS asan atomic task, to build the behavior repertoire of the robot with a gripping arm

Institute robotics team- Qualified 13th among 34 global teams in the Intelligent Ground Vehicle Competition - IGVC 2017, Michigan USA- Implemented the crucial navigation stack: localization through sensor fusion by ExtendedKalman Filter (EKF), enabling obstacle-avoiding GPS waypoint navigation- Designed and simulated the robot in Gazebo to test SLAM and lane-detection algorithmsthus reducing manual testing times by 75%

Co-authored Journal paper published in Soft Matter, RSC - 2017- Simulated a Monte-Carlo Brownian dynamics to estimate the number of cubosomes (lipidnano-particles) adsorbed and corroborated with experimental droplet retraction times