Projects

I build things from scratch, usually as a way to learn. Each of the projects below started as a question: how do transformers actually work, what does production deployment look like, can a portfolio fit in a terminal? Productivity apps, LLMs, systems work. Each one taught me something I wouldn't have learned by reading about it.

Machine Learning

ShakGPT — Custom LLM

A decoder-only transformer I built and trained from scratch to learn how large language models work end to end — 345M parameters, written in PyTorch. Not a wrapper around someone else's model: the architecture, training loop, and tokenizer are all hand-built.

PyTorch
Transformers
Next.js 15
Productivity Application

Pomodoro Focus

A production-grade, immersive focus timer designed to help users maintain flow state. Features a customizable Pomodoro timer, integrated "Media Dock" for ambient soundscapes, and secure authentication via Clerk.

Next.js 15
React 19
Tailwind CSS v4
Clerk Auth
Neon / Drizzle
Systems / DevOps

Portfolio TUI

An SSH-accessible, terminal-based portfolio built in Go using Bubble Tea & Lip Gloss. Features five TrueColor themes, a procedurally animated circuit board, IP-based rate limiting, and a hardened Docker deployment secured with non-root user, read-only filesystem and dropped Linux capabilities.

Go / Bubble Tea
Wish (SSH)
Docker
GitHub Actions
Interactive RAG Agent

AI Personal Assistant

A retrieval-augmented chatbot I built to answer questions about my professional background. Grounded in my resume, portfolio, and certifications using vector search and an LLM.

Next.js 15
Python / Groq
ChromaDB
RAG