Projects

June 2026

Best Agentic Architecture (US & Europe) · Ford Agathon

A seven-agent LLM system that consolidates decades of noisy, free-text automotive warranty claims into a library of consensus "canonical repairs" — clean, confidence-scored ground-truth labels for downstream machine learning. A chain of single-responsibility agents shares one context and self-corrects through an LLM critic loop, combining statistical consensus with regional cost grounding, while all numeric outputs stay deterministic and auditable. Built with Rider Harrison; won Best Agentic Architecture (US & Europe) at Ford's Agathon hackathon, competing against 250+ teams.

October 2024 – December 2024

A physical chess bot driven by a local minimax algorithm. Magnetic reed sensors detect pieces, stepper motors and a linear actuator move magnetic pieces beneath the board, and an STM32 orchestrates everything.

September 2023

A full-stack event recommender for UMich built during MHacks using an Azure-hosted GPT-4 API. It offered tailored event recommendations from users' interests and updated JSON event data online in real time.

February 2023 – April 2023

A real-time algorithm that tracks eye movement and projects the gaze onto the screen being looked at. Documented with related work, motivations, and conclusions in the style of a miniature publication.

August 2022 – December 2022

An A* search algorithm in C plus CNN-based image classification (PyTorch) enabling a differential-drive robot to navigate and map a maze.

October 2022 – November 2022

A C++ simulation of the Among Us game built on the Traveling Salesman problem and a set of routing heuristics.