- Developed machine learning models to extract tree-level metrics (e.g., height, diameter, species, tree count) from satellite imagery, integrating ground-level sensor data (LiDAR, GPS, etc.) collected across diverse forest types and terrains- Conducting on-site field visits for data collection, system testing, and validation, directly contributing to product refinement and improved customer satisfaction through hands-on integration and training efforts- Contributed to a patent-pending innovation, “Methods of Deriving Forest-Scale Insights from Narrowly Scoped Below-Canopy Datasets,” leveraging hyperspectral and multi-resolution satellite imagery for sub-canopy data integration

Developing a streamlined pipeline for a multi 3D object detection and tracking using only monocular vision for autonomous vehicles to handle different traffic scenarios on the road like traffic signals, intersections, etc.

- Created an interactive client-server network using sockets in Python to stream 3D point cloud data from a RGBD camera and visualizing it on a webpage using Node.js- Designed an end-to-end calibration pipeline product to complete Eye-to-Hand calibration and calculate the extrinsic parameters and the relative pose of a RGBD camera mounted on a robotic arm with respect to the end effector and base frames- Developed a web-based GUI using HTTP server to remotely control a robot through a ROS package on NVIDIA Jetson AGX Xavier- Quantized large deep learning models to lower precisions using ONNX and TensorRT to reduce inference time to a fourth- Designing smaller deep-learning models for faster inference by applying knowledge distillation techniques on large Vision Transformers- Designing the communication logic for the company’s robotic system using Allen Bradley PLC as the central control unit, facilitating interactions with the robot, camera, and cloud server for optimized automation

18-793 Image and Video Processing

• Developing a Reinforcement Learning based framework under Prof. Carlee Joe-Wong to incorporate altruism in an Autonomous Vehicle Fleet in a mixed-autonomy environment to mitigate various problems faced on the road• Experimenting with different network designs and other techniques to optimize and generalize the performance in mixed-autonomy transportation system and reach an equilibrium state sooner

• Worked under Professor Howie Choset• Worked on developing a novel algorithm to solve Multi-Agent Path Finding problem using Deep Learning techniques• Attempted to implement a learning-based Conflict Based Search (CBS) algorithm using Attention and Reinforcement learning• Studied development of robust algorithms for path planning for multiple articulated manipulators using Deep Learning techniques• Incorporated Reinforcement Learning concepts to improve sampling strategies of sampling-based path planning algorithms• Volunteered as a demonstration expert as well as improved the control calibrations for the Titan6 and Daisy6 hexapod robots

• Wrote Python scripts to analyse simulator outputs and automate end-to-end timetable generation for coaching trains for the Indian Railways, resulting in improved train scheduling and reduced downtime by 75%• Performed data clustering on more than 48,000 freight trains spanning over 6 major Indian Railway routes, to identify train timings and route patterns, thus optimizing resource allocation of crews and goods and increasing the net revenue by 3%• Developed and integrated a new feature component in the in-house Java based train simulator• Project funded by Western Railways

• Designed and back-tested quantitative trading algorithms in Python to maximize profits and cap losses to 10% of buying power• Interpreted and analyzed large historical financial datasets to identify trading opportunities using statistical analysis and curated comprehensive reports to assist future strategic actions

• Intern for IoT Solutions for Remote Asset Monitoring• Designed and fabricated temperature calibration circuits using Altium Designer for testing and generated the standard test set for prototype IoT devices• Created the standard emulator on Raspberry Pi for testing and calibrating sensors for the Jio Vehicle Tracker IoT device

• Developed a 3-axes autonomous quadruped robot capable of following a line, climbing steps and jumping over ropes• Designed prototypes for different orientations of the actuators in the mechanical structure• Developed the navigation algorithm for the robot on a Computer Vision based - android application built using MATLAB

• Designed a robot based on the specified dimensions for the Fastest Line Follower Challenge• Designed a micromouse robot with proper sensor and actuator placement for Maze Solving Competiton