$ ls -la ./projects/

This project explores test-time adaptation of hybrid State Space Models (SSMs) by injecting goal-conditioned perturbations directly into the recurrent latent states during inference. The goal is to steer model behavior toward desired outcomes without any weight updates and create a lightweight form of test-time fine-tuning that leverages the natural recurrence of SSMs.

A minimal, high-performance voice AI assistant built on Raspberry Pi Zero 2W + PiSugar Whisplay HAT. Button press → listen (STT) → custom prompt pipeline → model response → speak (TTS), with real-time display/RGB feedback.

A terminal-themed Next.js portfolio and blog showcasing projects, thoughts, and technical explorations in AI research, data analysis, and systems development.

Juliet is a modular AI framework where independent isomorphic agents develop persistent, personalized intelligence via swappable adapters that inject structured, tagged context into a standardized prompt pipeline enabling plug-and-play, divergent learning without rigid orchestration.

A minimal Fortran interface to llama.cpp using ISO_C_BINDING. If the Fortran Renaissance is real, it deserves its own 🦙 bindings.

A fun, terminal-based Rock, Paper, Scissors game written in modern C++23. Perfect for kids and adults alike! Built by a dad for his son to enjoy.

A classic Hangman game implemented in C++23 with a programming-themed word dictionary. Perfect for learning and entertainment!

A fun and simple C++ number guessing game that's perfect for kids and adults alike! This project includes three different versions of the classic guessing game, each with unique features.

A classic Snake game implementation using Pygame with modern features including a persistent leaderboard system and smooth gameplay.

TinyMamba is a minimal, educational implementation of a hybrid language model architecture combining state-space modeling (SSM) and transformer attention. It is designed for clarity, extensibility, and experimentation with memory-based reasoning and context-dependent computation.

A tiny command-line tool that searches **Grokipedia**, grabs the top article, and **saves the full text to a `.txt` file**.

A comprehensive financial data visualization platform that provides insights into U.S. economic indicators, commodity prices, inflation trends, and housing market metrics. Built with Next.js and powered by Federal Reserve Economic Data (FRED) API integration.

A comprehensive government data aggregation platform that simplifies access to official information from multiple U.S. government agencies and institutions. Deployed at: **[gov-feeds.discorover.com](https://gov-feeds.discorover.com)**

A lightweight Python pipeline to ingest, deduplicate, scrape, and summarize speeches, testimony, and comments published by Federal Reserve Board members. The resulting structured and summarized content is intended for use as model context, monitoring, analysis, or downstream NLP workflows—built to stay lean on cost by leveraging Grok-3-mini via the XAI API.

Fetches and displays the latest official **United States public debt data** from TreasuryDirect’s “Debt to the Penny” RSS feed. It provides a clear day-over-day summary, including the change in debt held by the public, intragovernmental holdings, and total public debt outstanding.

A Python-based pipeline to collect, parse, and structure U.S. Department of Defense (DoD) contract award announcements for use as model context or training data.

Watcher is a lightweight Python package designed for developers who prefer hot-reloading files from the terminal.