The Vault

Unpublished
Projects

A collection of functional software, hardware hacks, and tools that are actively used but haven't been polished for an official release.

Cheat Sheet + Claude Skill
2026Developer

Cheat Sheet + Claude Skill

This project provides a specialized Claude skill that takes in context like formula sheets, homework, and test solutions, and outputs a structured JSON file containing tips, warnings, and vector diagrams. The companion web editor allows you to manually format, edit, and export these super buffed up formula sheets without touching the raw JSON code.

Claude APIReactJSON Editor
Temu Lamp Controller
2026Developer

Temu Lamp Controller

I developed custom lighting software to interface with and control cheap off the shelf RGB lamps from Temu. By reverse engineering their protocols and bypassing their default apps, this software provides a unified interface to control colors, brightness, and animated patterns without relying on clunky third party apps.

IoTWeb BluetoothUI/UX
Pi LED Controller
2025Developer

Pi LED Controller

This software runs on a Raspberry Pi equipped with a touchscreen, serving as a dedicated control panel for LEDs wired directly to the Pi GPIO pins. It allows for robust control over lighting zones, custom animations, and brightness settings right from the hardware source.

Raspberry PiPythonLED Matrices
Optics Topic Visualizers
2025-2026Developer

Optics Topic Visualizers

Similar to the Thin Lens Guide, this is a collection of mathematical visualization tools built for different optics topics. They feature the same type of interactive visual elements, but are ongoing projects that haven't been fully finalized or published into standalone tools yet.

ReactSVGOptics Math
AI Model Benchmarking
2024-2026Researcher / Developer

AI Model Benchmarking

Whenever a new AI model is released, I use this ongoing benchmark test to evaluate its coding capabilities. The challenge is to build a fully interactive thin lens simulator using basic optics math. The process typically takes about four prompts per version: initial generation, adding desired features, final edits to ensure everything works, and stylization. It serves as a continuous test of AI coding capabilities rather than a finished project.

Prompt EngineeringReactAI Evaluation