Self-taught systems work focused on local-first AI, Rust infrastructure prototypes, and practical tools with runnable demos and clear implementation boundaries.
500 constructed practice runs across 8 kinds of break: 350 recovered, 40 after a retry, 50 partial, 60 failed. The public file is the 40-row sample. The runs are constructed practice cases, and the recovery decision follows a template. This sample does not show that training on the pack makes an agent better.
Rust mesh-network prototype with Pulse discovery, JSON-over-UDP protocol messages, heartbeat recovery state, registry inspection, CI, and a one-command demo.
Visual memory prototype that renders documents into image tiles, stores hypergraph evidence, retrieves similar tiles, and exports inspectable reports.
Domain-aware operating calendar built around deterministic planning rules, plain-English task reasons, and reusable farm playbooks.
Architecture/spec artifact for a governed local-first agent runtime, including invariants, authority rules, crate boundaries, and verification gates.
Start with the runnable repos. Look for tests, CI, small source trees, quick demos, generated artifacts, and status sections that separate implemented behavior from roadmap ideas.