I’m working on ways to make complex ideas easier to hold, develop, and use with language models.

I'm interested in helping people manage complexity, identify what matters, make it addressable, and keep working with it over time without needing a degree or a subscription.

I’m developing PGX, a lightweight natural-language format for making meaning addressable in LLM conversations and documents. PGX sits between ordinary language and graph practice: it lets people and LLMs name point at ideas, connect them, cite them, and reuse them while staying close to the way writing, thinking, and conversations actually work.

I’m also building Parmesan, a local, SQLite-backed tool for durable structured work with LLMs. The goal is simple: open a conversation, bring in your material, and develop a knowledge structure that can survive the chat.

• Developing PGX and collecting rigorous, reproducible case studies.
• Building Parmesan as a practical home for structured LLM work.
• Exploring mechanistic interpretability, applied AI, language-model interaction design, and software that gives regular people more agency.

The work is early and experimental. I’m interested in examples, objections, collaborators, and case studies that test what holds up.

Start with Parmesan.