Andy Beckwith
Principal engineer — AI orchestration and platform.
Twenty four years in production, with agentic AI at Zendesk. Open to full-time, contract, and advisory work — tell me the delivery problem and I'll scope it.
Plan first, name the unknowns, ship.
Demos
Multi-agent systems you can watch run.
Orchestration is hard to convey in prose, so these are built to be watched. No API keys required, every run simulated and deterministic.
Open the demosAI-Driven Wildland Fire Response
Eight specialist agents read a simulated wildfire incident and assemble a fireground picture: spread prediction, structure triage, evacuation triggers, resource tracking, and an IC decision-support briefing. Pick a scenario, advance the operational clock, watch the orchestration run.
AI-Driven Wildfire Mitigation Planning
The planning-side companion: a specialized internal intelligence platform fuses beetle kill, dead-and-down loading, structure density, topography, and historical + live weather into ranked mitigation wins for the US 285 corridor (thinning, fuel breaks, and home-ignition-zone work where they newly protect the most structures) and predicts outbreaks from live lightning.
A five-agent system, made watchable
Replay how five agents assemble a resume section from raw career notes: cluster, draft, critique, revise, emit. Framed as Monte Carlo Tree Search with the backup phase most production agent systems leave out. Pre-recorded and deterministic.