AI & Automation

Open source

Agent Workbench brings several AI runtimes into one local-first workspace. Instructions route to the right agent, session context stays visible, PTY controls keep execution inspectable, and the operator remains in the loop when judgment matters.

Open source

The Polymarket bot handles the pieces that should be systematic: market ingestion, normalization, edge scoring, liquidity checks, timing signals, and sizing. Execution stays gated, because the point is not to automate risk faster; it is to make the signal auditable before it gets authority.

Active infrastructure

Metis combines local-model lanes, persistent memory, governed task lifecycles, and multi-machine orchestration into one operating system. The framework is source-available; the private repo layers identity, projects, and integrations on top.

In progress

The gallery is part of my broader toolkit: AI-assisted research and data prep feed into hand-crafted D3 visualizations. The system handles the ingestion; the editorial judgment stays human.

The portfolio is part of my AI-assisted operating loop. Browser comments turn into implementation tasks, structured project data powers multiple views, and agent-supported workflows keep the site current without turning every update into manual website maintenance.