cv

Research Engineer

CNH Industrial, Boston MA

• Proposed and led a team of 3 to prototype/implement a multimodal (camera + LiDAR) semantic segmentation model for terrain traversability estimation; iterated architectures and trained PointTransformerv3-based model on in-house simulation and real dataset to achieve 0.815 mIOU
• Productionized and deployed a real-time, multi-threaded inference engine service for an image-based driveable area segmentation model using TensorRT, C++; performed end-to-end testing and documentation
• Developed a C++ service for communication between ISOBUS/CAN and publisher/subscriber middleware on a real-time system; tested and supported integration efforts on and off-field

Software Engineering Intern, Deep LiDAR

Argo AI, Pittsburgh PA

• Proposed a probabilistic 3D object detection pipeline for estimating uncertainties in bounding box detections; performed in -depth literature review; presented to the team
• Developed loss functions and model architecture upgrades to optimize and learn parameters for direct modeling of bounding box parameters as distributions using Tensorflow
• Created frame-wise BEV and pointcloud visualizations for bounding box parameter uncertainties as ellipsoids, cuboids, and arcs using Open3D and Python3

Computer Vision Engineer

Comofi MedTech, Bangalore, India

• Proposed and implemented an ensemble method for 3D segmentation algorithm in Python utilizing connected components; improved performance by 20%
• Implemented region-growing algorithm for segmentation of organs on CT scan (3D) data with 85% accuracy in Python
• Built a pointcloud preprocessing pipeline with Intel Realsense, PCL and ROS in C++ for filtering and downsampling incoming pointcloud data

Software Engineer

MagikEye Inc, Bangalore, India

• Deployed web service on Amazon EC2 with Docker containers as SLURM nodes for customer support and testing
• Conducted performance optimization on RaspberryPi Zero using QPULib by enabling non-blocking GPU load and stores for repeated mathematical operations in convolutions
• Wrote production-level, low-latency Python and C++ ROS packages for depth sensor; extended existing SDK to support ROS

Product Engineer

QtPi Robotics, Bangalore, India

• Led a team of 4 for development of an autonomous scaled-down simulation of a digital supply chain utilizing ESP32, RFID, IR sensors, Firebase, and an interactive dashboard for Robert Bosch
• Conceptualized design elements for website's UI/UX on AdobeXD to highlight key product USPs with, increasing customer engagement and session duration

Masters in Robotic Systems Development

Carnegie Mellon University, Pittsburgh, PA

• GPA 4.17/4.33
• Electives - Computer vision, Deep Learning, Geometric Vision, Multimodal ML, 3D Learning

Bachelors in Technology, Computer Science and Engineering

REVA University, Bangalore, India

• GPA 8.87/10

Autonomous Reaming for Hip Replacement Surgery

• Led the development of Perception and Sensing subsystem- implemented pose tracking and error detection for dynamic compensation of robot arm using Atracsys SpryTrack in C++, integrated with Planning and Controls via ROS
• Developed a GUI using Python and Open3D for manipulating pointclouds, displaying system metrics; integrated with all subsystems
• Managed systems engineering and project management efforts between all stakeholders with regular communication, updates and presentations

Deformable SuperPoint

• Proposed use of deformable convolutions for improved interest point detection for learning-based detectors
• Implemented deformable convolution layer; integrated with SuperPoint's VGG-style encoder
• Improved interest point detector repeatabilty score on MSCOCO by 7.4% within 1/2 number of iterations; showed boosted performance in homography estimation and SfM

Researcher

R-PAD Lab, CMU, Pittsburgh PA

• Performing literature review on keypoints-based representations for object detection in 2D and 3D, adding uncertainty estimates to predictions

Researcher

Spatial Experience Lab, School of Design, CMU

• Integrated Optitrack Motive Camera with TouchDesigner, Derivative