cv

Machine Learning Engineer

Apple

Integrated Google Cartographer for 3D SLAM, developed object detection algorithms, and fused Lidar streams for autonomous drone projects.

• Integrated Google Cartographer to perform 3D SLAM for an autonomous drone with 2D-Lidars and IMU as sensor modalities using Nvidia Xavier.
• Developed a DBSCAN-based object detection algorithm for Multi-Object Tracking on 3D Lidar stream.
• Set up a 3D Lidar stream using REST API and Amazon S3.
• Fused data from multiple Lidar streams to a single stream using ICP and feature extraction.

Machine Learning Engineer

Rizse

Designed and integrated real-time obstacle avoidance for UAVs and refactored algorithms for improved efficiency.

• Designed and integrated real-time obstacle avoidance to path planner of a UAV utilizing octomaps.
• Refactored the CPTSP algorithm to increase code readability and reduced operation time by 30%.
• Tuned and trained a machine learning model based on ResNet for aviation inspection.

Robotics Software Intern

Asensus Surgical

Designed XML builders and improved the performance of robotics systems.

• Designed an XML Builder using Factory and Singleton Design Patterns.
• Coded extensively in C++ and upgraded tinyxml2.0 to parse the XML document in an RTOS like TwinCat-3.1.
• Stabilized the performance of the robotics arm by finely tuning the PID constants for the control system.

Graduate Research Assistant

EcoPRT Lab, North Carolina State University

Integrated object detection algorithms in autonomous driving environments and collaborated on research projects.

• Integrated PointPillars for object detection with ROS-based Autonomous Driving Environment.
• Implemented HDL-Graph-SLAM for an L-4 autonomous vehicle utilizing Velodyne LIDAR.
• Developed a project on people’s reaction to autonomous vehicles in collaboration with the psych department, designing a lighting system using Raspberry-Pi controlled LEDs and C++.

Product Developer

Toppr Technologies Pvt. Ltd.

Reduced operational costs and managed a team for various projects.

• Reduced operational cost by 23% by removing duplicate questions utilizing a word-to-vector algorithm implemented with Python and Keras.
• Managed a team of 5 individuals and mentored 15 interns on projects like 'Duplicate question detector.'

Master of Science

North Carolina State University

Electrical Engineering

• Neural Networks
• Computer Vision
• Pattern Recognition
• Mechatronics
• Digital Imaging Systems
• Random Processes
• Design and Analysis of Algorithms
• Probabilistic and Graphical Models

Bachelor of Technology

Maulana Azad National Institute of Technology (NIT-B)

Electrical Engineering

PointPillars++ for Object Detection (Thesis)

Designed a novel object detection algorithm improving state-of-the-art performance on the KITTI detection benchmark.

• Improved the state of the art for 3-D object detection on KITTI detection benchmark from 59.20 mAP to 66.83 mAP.
• Improved performance for all classes in all difficulty strata.

Leaf Wilting Classifier

Trained a CNN model for classifying wilting levels in leaves using transfer and ensemble learning techniques.

• Ranked 3rd in a competition of 50 teams based on model performance.
• Mitigated issues caused by an unbalanced dataset using XGBoost and ensemble learning.

Facial Recognition Models: A Comparative Analysis

Compared classical machine learning techniques with deep learning models for facial recognition tasks.

• Concluded that classical approaches can perform better on smaller datasets when transfer learning is not involved.
• Based findings on ROC curve analysis.

Shortest Route Finder

Developed an application to find the shortest route between cities in North Carolina using C++, OpenCV, and OpenGL.

• Improved search algorithm performance through low-level optimizations.
• Utilized Open3d and QT for visualization.

PointPillars++: an encoder for 3-D Object Detection and Classification from Point Clouds

North Carolina State University