CONTENTS
Research Areas
Self-Driving Car
Network Science
Recommender Systems
Generative Design
Achievements
Publications - 1 (UR)
Conferences - 1
Internships - 2
Academic Projects - 3
ABOUT
Currently, working at an A.I Stealth Startup as a data scientist. My work here involves working with Large Language Models (LLMs) and other Natural Language Processing Techniques to build recommender systems and tools for personal use.
Artificial Intelligence (A.I.), and to a larger degree, Machine Learning (M.L.), are behind some of the most important technologies used today. From autonomous vehicles like Tesla to recommender systems on Netflix and world-building tools used by Facebook's metaverse, some of the largest companies by market capitalization spend fortunes on research and development on A.I. and M.L. Since August of 2020, I have focused much of my time on developing skills in these fields. The following paragraphs highlight some of the areas I have worked towards.
I used machine learning to recognize gender through audio. In this academic project (WPI CS539), my team and I implemented various algorithms and compared them to see which algorithms have better performance metrics. I took a deep learning class right after, where I worked on a similar group project, this time, deploying deep learning algorithms to predict the value of stocks in a market (WPI CS541). Since then, I also had the opportunity to work as a data science intern at Juniper Networks Inc. Here, under my manager, Zhongyi Jin, I worked on various recommender engines including the Service Request (SR) Classification project. A significant contribution of mine was the initiation of the Software Upgrade Recommender. In this project, I researched and developed a machine learning pipeline to tackle the issue of upgrading the operating system of a router in a cluster. At the end of my internship, I transferred the tool to full-time employees.
As a Teaching Assistant for my master's thesis advisor, I was fortunate to work with a team that developed the Modular Package for Autonomous Driving (MPAD). I co-wrote a paper (under review) on the implementation of this tool in classroom settings. As a research assistant at the Order-Disorder Phenomena Lab, I have been working on connecting energy studies on far-from-equilibrium systems like Rayleigh-Bénard Convection to network theory: check section 5.2 for a basic idea. I also worked on designing an automated design framework for powertrains which heavily relies on artificial intelligence especially when it comes to design choices. Design choices in this project are based on graph grammars and tree search (using a tool called Graphsynth).
Gallery
Driving dashboard on MPAD’s web application with lane recognition display.
Example Final Autonomous RC Car with payload (screwdrivers) and sensor package.
Gephi Visualization of Room Temperature Thermal-Image of Non-Turbulent Rayleigh Bénard Convection. This can be done by converting the thermal image into a matrix of pixel intensities, using autocorrelation functions, thresholds, and multigraph components.
Gephi Visualization of a Steady State Thermal-Image of Non-Turbulent Rayleigh Bénard Convection. Notice how there are two distinct regions in the steady-state image, the segments account for the hot and cold spots respectively.
Research Output
Paper (Under Review): Developing a Modular Package for Autonomous Driving and the Experiences in Implementing it in a College-Level Design Course
Radio-Controlled (RC) scaled cars are a great way for prospective engineers to learn real-world technical skills including the basics of autonomous driving. Current self-driving RC car modules require hours of setup configurations and course-specific training before driving autonomously. Even after training, most modules do not provide the capability to avoid obstacles or traverse non-standard terrain. An ideal self-driving module must be capable of being downloaded and installed onto an RC car regardless of its dimensions. Since this project’s audience is typically mechanical engineering students with little computer science and electrical engineering experience, having controls and driving data to be easily accessible on a website would prove to be extremely beneficial. Hence, we developed the Modular Package for Autonomous Driving (MPAD) to allow Mechanical Engineering students to test their work in a complete system without requiring a coding or electrical design background. MPAD consists of two development boards, Raspberry Pi 4 8GB and Elegoo Mega 2560. The sensors (Ultrasonic, inertial measurement unit, temperature, and hall-effect) are connected to the Elegoo, which then connects to the Raspberry Pi. The sensor package uses a Raspberry Pi camera for lane recognition and ultrasonic sensors for collision avoidance. The package is specifically designed to be easily transferable from one RC car to another. MPAD was implemented in an introductory engineering design course at XYZ University, where groups of students had to design a custom scale car with a gearbox, steering linkage, and chassis that can accommodate MPAD. The objective was to check the modularity of MPAD and to verify if the documentation is sufficient to help mechanical engineering students successfully create a self-driving RC car. Eight teams of mechanical engineering students each received documentation in the form of two guides to create their own version of MPAD. While there were challenges and learnings, the integration was successful, and the students were able to fully utilize MPADs capabilities with their own RC car design and demonstrate a self-driving scale car. This article will discuss the implementation, testing details, and future work for a self-driving modular package.