Research
Alpha Omega Labs
- Architect and principal engineer of Cortex, a self-hostable control plane that lets an organisation run AI coding agents from ordinary chat channels under enforced policy with a full audit trail — roughly 182,600 lines of first-party Python and TypeScript across 53 backend service modules and a 17-area React operations console.
- Designed its two-plane authorisation model — 15 control surfaces × 4 account groups × per-runtime capability policies — so a lower-privilege human can safely drive a higher-capability agent without privilege escalation, backed by an append-only audit log and TTL-scoped cross-organisation delegation.
- Architect of omegaXiv, a nine-service platform on which AI agents run a scientific investigation end to end: plan it, execute code in isolated sandboxes, self-check against quality contracts, and publish a reviewable paper.
- Built spend governance directly into the run lifecycle — payment pre-authorisation before dispatch, settled by capture-or-cancel through a single idempotent finaliser — so an autonomous run cannot silently burn budget.
- Built Gauntlet, a four-track agent-harness benchmark on UK AISI's Inspect AI that measures harness delta rather than absolute model score, running safety probes inside a network-blocked sandbox against a fabricated decoy.