I build AI-native products and operating systems for fintech, regulated workflows, and long-running agents.

My work usually starts with a messy operational problem and ends with something testable: a working product, a deterministic receipt, or an observable handoff. AI accelerates the research and implementation; evidence, controls, and human ownership stay visible.

Website · Arcana Research · X

• Event Manager OS — An end-to-end conference operations system that moves one persisted record from proposal through public schedule, with cross-role handoffs and audit history. Live demo.
• Scam Discovery Platform — A public-data pipeline for finding emerging consumer-fraud patterns before they become established categories.
• FileReady — A synthetic mortgage-processing prototype for evidence-linked document review, borrower follow-up, and processor-controlled underwriting handoff.

• Evidence Workbench — A deterministic CLI that turns claims and supporting material into auditable decision receipts.
• rrule-kit — Timezone-correct recurring scheduling primitives for long-running agents and automations.
• agent-reminders — Lightweight reminder and follow-up infrastructure for AI agents.
• Researcher — A zero-dependency CLI and Raycast extension that turns URLs, files, and YouTube videos into clean Markdown dossiers.

• Working systems over slideware.
• Human approval at consequential boundaries.
• Source-linked evidence instead of confident black boxes.
• Tests at the observable boundary: persisted records, API responses, receipts, and rendered interfaces.
• Honest labels for observed, synthetic, modeled, and unavailable data.

I run Arcana Research, where I study how large organizations adopt, operate, and pay for AI. I also build local agent systems and small products that make persistent AI workflows easier to inspect and control.

Some work stays private on purpose: customer data, outreach workflows, Arcana's research platform, and the core of my personal command center. Public repositories use public or synthetic data and state their boundaries explicitly.