I'm a Member of Technical Staff at ATG, a YC-backed applied AI lab founded by the people who built Paperspace. I work remotely from San Francisco.
Outside of work I build open-source tools for agent-driven development and compete in hackathons: eight awards, four of them first place. Lately I’ve been judging them too.
Hackathons
The three I'm proudest of, then the rest.
1st place · January 2026
Bake-Off. A marketplace where you post a task and AI agents compete to do the work and get paid. Built in twelve hours at YC headquarters; 495 people applied for 190 spots.
First prize came with $25,000 in cash and a guaranteed YC interview.
1st place · February 2026
Memory Palace. An education tool that turns any study topic into an interactive memory palace, generated with Odyssey's world-model API. Built in four hours.
First prize: $25,000 cash plus $50,000 in Odyssey and AWS credits.
1st place · September 2025
Ross. An AI system that finds clients for personal-injury law firms: it reads police reports, picks out the viable cases, and handles the outreach. 44 teams competed.
My first first-place finish — twelve days before I joined ATG.
Conductor Control. Turns a room full of phones into one conductor-led instrument.
Watch & Learn. Turns a screen recording into step-by-step skills an AI agent can follow.
Fall Activities. A selfie in, autumn travel photos out — no prompts to write.
Kaizen. Reads GitHub, Linear, and Datadog to show engineering leaders where the team actually spends its effort.
Powwow. A personal chief-of-staff for conferences: coordination, intros, follow-ups.
Judged AI projects from 83 competing teams alongside folks from AWS, Datadog, Section 32, Sierra Ventures, and xAI.
Judged the San Francisco edition of a six-city on-device-AI hackathon, hosted by AI Tinkerers SF.
Open source
Memory and workflow system for Claude Code: repo context, rules, and patterns that survive between sessions instead of dying with each one.
Docs, standards, and review defaults for AI-coded repos, so a project agents have worked on for months is still understandable by a human on day one.
A context router: watches the conversation and loads the docs and skills an agent needs before it starts working, and nothing more.
Local-first issue tracker that lives inside the git repo, with a CLI and a small browser UI.
Some of my open-source work lives in other orgs' repos: I built official demo applications for CopilotKit, and LangChain featured one of them, Scene Creator Copilot.