I translate complex data
into actionable intelligence

This Portfolio Site

Frontend Development, UI/UX Design

A highly responsive Single-Page Application (SPA) designed with a utility-first CSS architecture. Features dynamic modal states, Base64 asset embedding, and performance-optimized layout rendering for seamless cross-device compatibility.

Metal Genre Classifier

Machine Learning, Transfer Learning, Neural Network Training, Web Scraping, Database Management, Front-end Integration

Interactive client for a logo classification service. Handles user image uploads, front-end resizing/pre-processing (to 244x244), and displays predicted genres and confidence scores from the model endpoint.

Slay the Spire Content Mod

Java Development, Game Design

A comprehensive gameplay expansion developed in Java. Engineered complex relic mechanics involving exponential scaling functions and probabilistic event triggers, requiring deep integration with the base game loop via the ModTheSpire API.

Global Population Projector

Statistical Modeling, Numerical Optimization, Data Visualization, Cloud Deployment

A data-driven forecasting tool that models demographic trends using a generalized logistic function. Implements a custom Minimax optimization loop to minimize maximum relative error, ensuring high-fidelity fits across non-linear historical data points.

Video Sorter

Multimedia Processing, Containerization (Docker), Streaming/VOD Integration, Cloud Deployment

A real-time audio-visual processing application that deconstructs a video into individual frames, analyzes the Root Mean Square (RMS) volume of each frame's corresponding audio segment, sorts the frames into a crescendo of intensity, and re-renders the final output.

Reinforcement Learning, Systems Engineering, Lua Modding, Neural Networks

A deep-learning agent capable of autonomously playing the complex roguelike deckbuilder Balatro to completion. Engineered a custom Lua mod to extract raw game state via Inter-Process Communication (IPC). The Python backend vectorizes the board state and evaluates hundreds of potential action-state pairs per turn using a custom PyTorch neural network, employing temporal penalty discounting to solve the credit assignment problem.