Systems Builder | AI automation, full-stack products, infrastructure, and business operations
I turn messy real-world operations into working software. My recent work spans construction intelligence, hospitality systems, local-business operations, and governed AI infrastructure—from interfaces and integrations through deployment, monitoring, and operator handoff.
Tenant-aware signed webhook verification with canonical-host checks, bounded timestamps, timing-safe signatures, and replay refusal. TypeScript; 8 focused tests.
A resumable accept/edit/reject/defer contract that preserves originals, binds decisions to digests, and blocks unresolved export. Python; 15 focused tests.
A model-agnostic specification and offline evaluator for isolated contexts, deny-by-default capability policy, one-use approvals, and redacted audit records. Python; 24 focused tests.
A clean-room reconstruction of a restaurant field deployment: passive no-popup capture boundaries, broad red-panel qualification, and clear-aware ticket-identity deduplication informed by two Windows deployments. Python; 23 focused tests with synthetic fixtures only. No client screens, identities, payment records, credentials, or private topology.
- AI agent control plane — operating: isolated user contexts, familiar human surfaces, replaceable model workers, approval-gated tools, and audit state.
- Self-hosted family infrastructure — operating: NixOS/Linux services, private networking, email, password management, PostgreSQL, encrypted backups, monitoring, runbooks, and rollback.
- AltonBnB operations — operating: direct booking, OwnerRez/Airbnb/Vrbo coordination, event alerts, and private revenue/guest-operations tooling.
- BetterBis — private operating lab: websites, business communications, CRM, media, intake, and launch systems packaged as reusable components for local businesses.
- Instruxit drawing intelligence — built/private collaboration: evidence contracts, confidence scoring, rejected-candidate tracking, and human review.
- Contractor operating system — separate proposal: designed but not deployed.
Private source, family/customer data, and operational topology remain private. The public repositories above are original clean-room examples with synthetic fixtures, fresh histories, explicit status boundaries, and Apache-2.0 licensing.