Projects

Needle In A Haystack

Testing a LLMs ability to recall information from a long context.

I created a benchmark to test how well LLMs can recall information from a long context window. It was tested on gpt-4o & claude-2.1

Terra Mano

Handcrafted bronze maps of American terrain.

In need of a physical project, I managed 3d printing, casting in bronze, and hand finishing maps of US mountains.

ChunkViz

Visualizing different chunking strategies.

LLMs do better with shorter context windows. I built a tool to visualize how different chunking strategies to help you pick the one that is best for your use case.

SnakeBench

Benchmarking LLM through multi-headed snake games.

50 LLMs battle it out on Snake.

I Vibe More Than You

If you don't have 85 terminals open, are you even vibing?

Meme site that allows you to open up as many vibe terminals you want.

Agent Traffic Control

A radar view of agents completing work in parallel.

A UX exploration combining radar charts and agentic work.

TBPN Limited

Physical Technology Trading Cards

Phsyical trading cards based off of popular TBPN moments

agnts.sh

Plain-text links for agents.

A link service built for agents. Every URL serves three response shapes from the same endpoint - an HTML stub for humans, raw markdown for agents, and full JSON - based on the request's Accept header. Skip the scraping, just hand your agent a link.