AI engineer with 8+ years building web software in TypeScript. These days I work on agentic systems — agents that verify their own work instead of just generating.
Sidekick — a self-verifying agent. A worker drafts, a second model judges the draft against explicit success criteria, and the worker revises until it passes or hits the cap. Includes a 10-case deterministic eval harness, tool retries with timeouts, and prompt-injection defense.
Baton — shared memory for AI coding agents. An MCP server + CLI that saves a checkpoint in one agent and hands it to the next, so context survives the handoff. Local SQLite, 13 tests, CI.
Agentic systems · MCP · evaluation · guardrails · TypeScript