This displays the research that I have undertaken and participated in or am currently working on. Providing the paper along with the conference it is submitted to or published in. Research sparks an interest in me and helps me follow my passion. I may not be able to showcase all the information for the ongoing projects, however, for more information about my work or to discuss a collaboration, please feel free to get in touch through the contact page.
AI Agents for Sepsis EHR Analysis
Python
May 2025
- Designed and implemented Deep SVDD models for anomaly detection on MNIST and CIFAR-10, using geometric latent-space constraints for one-class learning.
- Enhanced model performance with z-score normalization and feature patching (FPSVDD), improving AUC scores in complex, high-variance image settings.
- Built an interactive web tool to visualize model performance and simulate the impact of architectural and pre/post-processing changes.
Solace: A Mental Health & Self-Discovery App
Python, Html, CSS, Javascript
April 2025
- Designed and implemented Deep SVDD models for anomaly detection on MNIST and CIFAR-10, using geometric latent-space constraints for one-class learning.
- Enhanced model performance with z-score normalization and feature patching (FPSVDD), improving AUC scores in complex, high-variance image settings.
- Built an interactive web tool to visualize model performance and simulate the impact of architectural and pre/post-processing changes.
Deep One-Class Classification
Python, Html, CSS, Javascript
April 2025
- Designed and implemented Deep SVDD models for anomaly detection on MNIST and CIFAR-10, using geometric latent-space constraints for one-class learning.
- Enhanced model performance with z-score normalization and feature patching (FPSVDD), improving AUC scores in complex, high-variance image settings.
- Built an interactive web tool to visualize model performance and simulate the impact of architectural and pre/post-processing changes.
Color Calibration
Python, Html, CSS, Javascript
March 2025
- Built a full-stack machine learning pipeline to calibrate and classify colors from image data using PyTorch, with automated preprocessing, model training, and evaluation via Jupyter Notebook.
- Designed and deployed an interactive website to visualize model predictions and color extraction results, allowing users to test the tool live without running code.
- Modularized the project into reusable scripts for data analysis, color feature extraction, preprocessing, and object detection, enabling both notebook-based and command-line execution.
BB84 Quantum Cryptography Simulation
Python, Html, CSS, Javascript
March 2025
- BSimulated Quantum Key Distribution (QKD) with BB84: Developed a simulation of the BB84 protocol, allowing secure key exchange by modeling quantum bit transmission, measurement, and key reconciliation between Alice and Bob, while detecting eavesdropping attempts by Eve.
- Interactive Visualization: Created an interactive website to visualize the protocol's operation, including quantum bit transmission, measurement, and detection of potential eavesdropping, providing a hands-on understanding of quantum cryptography.
- Security Evaluation & Future Enhancements: Evaluated the protocol's security properties and efficiency, with plans for future improvements like error correction techniques and the integration of quantum circuit diagrams to visualize qubit transmission.
Tom and Jerry: The Cheese Napping
Python, ROS
December 2024
- Our project involves two robots playing a simplified game of tag.
- The robot with the claw is assigned the role of "Jerry," while the other robot takes on the role of "Tom."
- The Jerry bot must navigate through obstacles, find and pick up a colored block, and then make its way through more obstacles to reach the end wall, which is marked by a fiducial.
- The Tom robot's goal is to chase and “tag” Jerry by colliding with it or getting very close.
Fiducial based navigation
Python, ROS
November 2024
- Built a fiducial-based navigation system where a mobile robot detects visual markers using ROS and OpenCV, navigates to each one, and returns to its starting point in a real-world environment.
- Implemented real-time scanning and motion control logic that allows the robot to rotate in place, recognize fiducials using camera input, and accurately move toward detected targets.
- Applied coordinate transformations using the tf2 package to understand spatial relationships between the robot and fiducials, ensuring smooth and precise navigation.
- Developed a complete ROS package with clean Python code, a launch file, and documentation, and demonstrated the project with a video walkthrough explaining the system’s architecture and behavior.
Maze Solver
Python, ROS
November 2024
- Built a wall-following robot using ROS and LIDAR by writing a node that maintains a fixed distance from the wall through continuous distance sensing and motion control.
- Processed live /scan LIDAR data to determine wall proximity and orientation, accounting for noisy or invalid values and adjusting robot trajectory accordingly using calculated angular and linear velocities.
- Tested the system in various Gazebo simulation scenarios—tight corners, angled starts, and offset distances—to evaluate how well the robot could correct its path and follow the wall reliably.
- Designed a modular ROS package with clean Python code, debug-friendly logs, and launch files, and documented everything with a clear README and demonstration video.
Line Follower
Python, ROS
November 2024
- Developed a line-following robot using ROS and OpenCV, where the robot processes live camera feed to detect and follow a clearly marked path on the ground in both simulated and real-world environments.
- Implemented image processing techniques like color filtering, contour detection, and centroid tracking to extract line features from RGB camera data and guide robot movement accurately.
- Integrated a PID control loop to adjust the robot’s heading based on the line’s position in the image, maintaining smooth and responsive path-following behavior across turns and slight deviations.
- Created a complete ROS package with well-structured Python code, launch files, and documentation, and demonstrated the system through recorded runs in Gazebo and real-world scenarios.
Wall Follower
Python, ROS
November 2024
- Built a wall-following robot using ROS and LIDAR by writing a node that maintains a fixed distance from the wall through continuous distance sensing and motion control.
- Processed live /scan LIDAR data to determine wall proximity and orientation, accounting for noisy or invalid values and adjusting robot trajectory accordingly using calculated angular and linear velocities.
- Tested the system in various Gazebo simulation scenarios—tight corners, angled starts, and offset distances—to evaluate how well the robot could correct its path and follow the wall reliably.
- Designed a modular ROS package with clean Python code, debug-friendly logs, and launch files, and documented everything with a clear README and demonstration video.
Basic Mover
Python, ROS
October 2024
- Programmed a ROS-based autonomous robot to drive precise paths in simulation using velocity and odometry data.
- Implemented movement routines to drive forward, rotate, return, and trace geometric paths like squares and circles.
- Subscribed to real-time odometry data to calculate position, distance, and orientation for accurate motion control.
- Demonstrated control logic, algorithmic decision-making, and ROS pub/sub architecture through live simulation.
Double Follow
Python, ROS
October 2024
- Modified a ROS TF2 simulation to support multiple follower robots, forming a real-time “duckling” chain behind a user-controlled leader.
- Used coordinate frame transformations to calculate each follower's position relative to its leader, enabling smooth, chained motion.
- Demonstrated multi-robot coordination using TF2, custom Python nodes, and ROS pub/sub messaging in a Gazebo-based simulation
Covey.Town Library
React, Typescript, HTML, CSS, Javascript
February 2023
- Covey.Town provides a virtual meeting space where different groups of people can have simultaneous video calls, allowing participants to drift between different conversations, just like in real life.
- However, covey.town does not have a library and we wanted to be able to add an interactable library area. Our aim was to solve the current problem that there are no spaces for education in covey.town for knowledge acquisition.
- With this we were able to: (i) read books through the web app, (ii) download books to a personal device, (iii) leave book ratings for other users to view, (iv) engage in discussion threads with other users about specific books, (iv) access audiobooks and listen to them in a records area and (v) rate and discuss the audiobooks.
Pet Dataset Segmentation
Python
February 2023
- Developed a Multi-Model Approach: Utilized 5 distinct deep learning models (CNN, ensemble CNN, ResNet, ViT, and a combination of pre-trained models) to accurately classify and segment 37 different breeds of cats and dogs from the Oxford-IIIT Pet Dataset, achieving up to 95% accuracy.
- Innovative Segmentation & Bounding Box Detection: Created a baseline segmentation model, then extended it with advanced models like ResNet and ViT, and incorporated bounding box detection to improve pet identification, with practical applications for animal shelters and clinics.
- Optimized with Data Augmentation: Leveraged state-of-the-art data augmentation techniques to enhance model performance, using a large, diverse dataset to train models efficiently without the need for massive computational resources typically required for such complex tasks.
Valiant Analyst - An NLP Mechanism
Python, HTML, PureBasic
January 2023
- We worked on creating a dashboard and a pipeline that uses natural language processing to make question answering analytics easier.
- We created a platform that takes in audio input of the question, then gets its context and provides a preliminary, supportive answer (using question answering analysis), and named entity recognition (NER). Along with this, we get the sentiment analysis for the professor. It can be depicted as shown below.
- It is very interesting because it is highly applicable and actually usable in classrooms to assist with learning and teacher support.
Machine Learning in Finance
Python
April 2023
- We worked on different models and machine learning algorithms to create predictive analyses.
- Some projects that we worked on included:
- Bankruptcy prediction
- Lending club originations and credit score information content
- Twitter sentiment analysis and stock returns
- Pre-paid card marketing
- Crowdfunding campaign success prediction
- Default rate prediction using historic credit card data
Website @Generate
React, HTML, CSS, JavaScript, Fly.io
January 2023
- Managing a team of 10 engineers and designers to develop the website for Generate, the product development club to showcase the work we do and a hub for all the information related to the club
- Leading weekly meetings with the team and the clients, scoping project, creating a project charter, setting deadlines and planning project execution while motivating the team and holding technical workshop
Jurni @Generate
Python, Figma
August 2022
- Managing a team of 12 engineers and designers to develop a natural language processing algorithm, wireframe prototyping, and branding for Jurni, focusing on mental wellness and insight driven journaling
- Leading weekly meetings with the team and the clients, scoping project, creating a project charter, setting deadlines and planning project execution while motivating the team and holding technical workshops
- Jurni fosters mental wellness for employees through an insight driven journaling platform that provides direct access to verified mental health professionals
Wordle Visualization
HTML, CSS, JavaScript, D3
January 2022
- Collaborated with a team of 5 to create a website for the popular game of Wordle
- Explored and creatively visualized the data to analyze player and word trends, frequencies, and performances
- Wrote a research paper outlining the process and findings of creating this visualization and presented the findings at Women's Research and Engagement Network (WREN) research conference
Chess Bot
Python
July 2021
- Developed a chess bot using Monte Carlo Tree Search, heuristic playouts and state evaluation functions based on depth and runtime
- Utilized game plays against other bots on lichess.org to collect data on the bot’s performance
Animator
Java
June 2020
- Developed a user interface to add, remove and edit shapes, key frames, and scrub them
- Built the project using Java Swing and MVC model
N Bullets
Java
April 2020
- "Nbullets" is a simple Java game that involves shooting bullets to destroy enemy ships. The game world consists of two main entities: bullets and ships.
- Players control a spaceship that can shoot bullets, while enemy ships move across the screen horizontally.
- The objective of the game is to shoot down as many enemy ships as possible while avoiding collision with them.
FIRST Robotics Competition (FRC)
3Ds Max, C++, Java
2018-2019
- I dedicated myself to learning various animation software, coding, construction, and marketing techniques.
- I actively sought opportunities to serve society and create a meaningful impact.
- Throughout my journey, I embraced challenges, learned from them, and honed skills in teamwork, cooperation, and collaboration.
- I explored different aspects of marketing, including seeking sponsorships, gaining exposure, and networking with influential individuals.