AdaPhish

AdaPhish is an adaptive phishing detection and submission platform designed to enable secure sharing of phishing emails and provide a robust solution to evolving phishing scams. It is now being tested within the United Nations University, and the paper has been accepted to the 4th International Conference on AI in Cybersecurity (ICAIC). An update will be posted soon with the link to the paper once it is officially available on the IEEE Xplore.

Raphael

Raphael is a chess engine I made in C++, and it is accompanied with a GUI built with SFML. It is capable of easily beating most players and even some publicly available chess engines. It uses the recursive negamax function to rapidly explore outcomes of different moves, and it leverages several optimizations incuding transposition tables, iterative deepening, SEE pruning, and pondering. It also has a NNUE mode to evaluate using a neural network, though I am currently in the process of training it using data from my own evaluation function rather than Stockfish's.

Kineval

Kineval is a Python-based interactive learning tool made for EECS 367 at the University of Michigan and it includes a wide variety of features including URDF loading, forward kinematics, inverse kinematics, collision detection, and RRT motion planning. It's written in a manner to allow students to implement these features themselves while only focusing on the key details (i.e., not the rendering, keyboard control, etc.). It is planned to be deployed in Fall 2026.

Folio

Folio is a startup which connects students and freelancers with startups. I contributed to many aspects of the platform, including its robust AI-powered search engines, its notification architecture, and user migrations from the old platform. The platform is powered by a Django backend and a Next.js frontend.

ConverseCart Search Engine

I worked with the startup ConverseCart to build an AI-powered search engine by leveraging word embeddings, bag-of-words models, and cross encoders. It uses a 3-step approach, where it first retrieves a bunch of candidate products using a vector search on the query and product embeddings, then uses a cross encoder to rerank the products by relevancy, and a personalizer to boost certain product relevancies. It also caches results to boost speed.

RTD Toolkit

RTD is a technique developed at the University of Michigan for making provably safe motion plans. It accounts for uncertainties in the measurements as well as the robot movement using polynomial zonotopes to avoid collisions. I worked with ROAHM lab, the team that developed this technique, to built a Python toolkit for easily implementing RTD on any grounded robot.

Plaite

Plaite is a project my team and I developed in 24 hours during MHacks 2024, and it won us the Best Use of Intel AI award. Plaite is a nutrition tracking app which uses YOLOv8 for real time food detection to simplify the tracking process for users. Plaite also leverages LLMs and a food nutrition API to offer tailored advice on improving health and meeting dietary goals.

Mandelbrot Fractal Explorer

This is a Mandelbrot Fractal Explorer I built in C++ & SFML. It uses thread pools to calculate the pixel colors concurrently, and is able to achieve zoom levels of 10^-16 before floating point precision comes into effect. It also implements a few other optimizations such as cardoid and periodicity checking to skip known regions that are in the Mandelbrot set.

JSA Website

I overhauled the website for the Japan Student Association (JSA) @ The University of Michigan, mainly by moving the website to GitHub Pages for free hosting, and by rewriting some of the hard-coded page elements to pull dynamically from a JSON file so that non-CS majors could still update the website with ease. This solved the problem of the website not being updated for months due to people not knowing how, and allowed us to advertise our events to a wider community.