Projects

Here are some of the cool things I've worked on not featured elsewhere:

RelCube - Relational Embedding Transformer for Rubik's Cube Solving (In Progress) - A Transformer-based deep learning solver targeting state-of-the-art performance with orders of magnitude fewer parameters. Uses structured piece representation (corners + edges) to exploit the cube's physical geometry rather than learning geometry from scratch. Dual-head architecture: Value Head predicts distance-to-solve, Policy Head predicts optimal next move. Trained with Approximate Value Iteration and an Expanding Curriculum to bootstrap optimal paths. Integrated with A* search for full solutions. Team of 3 for class final project. [Coming Soon]
MIRA - Minecraft Iterative Reasoning Agent (In Progress) - As the sole developer, I am building an iterative AI agent designed to generate, debug, and verify complex logic circuits (Minecraft Redstone). Unlike standard generative models, MIRA uses a Reinforcement Learning from Execution Feedback (RLEF) loop to rigorously test designs against executable "contracts." The system features a custom headless simulation environment (decoupled from rendering for high-throughput training), a "Corruptor" engine that generates synthetic training data by injecting mechanical faults into valid circuits, and a "Bedrocking" system to freeze verified logic modules—mimicking formal verification checkpoints used in chip design. [View Technical Design Document]

COMAP 2023 - Modeling Ecosystem Stability Under Climate Change - Submitted to the rigorous COMAP Mathematical Contest in Modeling, my teammate and I developed models to assess ecosystem resilience to precipitation changes. My specific focus was implementing Cellular Automata methods—simulating discrete grid-based interactions similar to logic gates—to analyze spatial dynamics and predict system stability over time. Our results were validated through comparative analyses with continuous differential equation models. [Link to project]
Dynamics Simulation of Rigid Bodies on Oiled Surfaces - I developed a physics engine from scratch to model the non-linear dynamics of bowling balls. This involved constructing a system of differential equations to capture friction transitions, rotational inertia, and surface interaction, implementing the solver in Python to visualize complex trajectories. This project addresses a gap in tooling by applying rigorous physics modeling to practical sport dynamics. [Link to project]

Optimal Slime Chunk Finder (Spatial Optimization Tool) - An interactive tool that solves a spatial optimization problem for Minecraft world seeds. It uses a custom API to process large 2D grid datasets via stream processing, identifying local maxima for mob spawning rates in real-time. Designed to handle high-throughput calculations to find the statistically optimal coordinate for farm placement. [Link to app]
View All Web Utilities - A complete collection of web tools and utilities I've built. [Link to apps directory]