I build AI agent systems, code intelligence, and computer-use tools.

Portfolio · Ranger · LinkedIn · X

Ranger — an AI forward-deployed engineer prototype connecting customer conversations to scoped, approved code changes. It combines voice, repository context, plan approval, execution in Linux VMs, and verification. I built its TypeScript/Fastify backend, durable worker, SQLite event history, and Trace integration. Source is private.

Trace — local code intelligence for TypeScript and JavaScript repositories, exposed to agents through MCP. It combines lexical, vector, symbol, and structural retrieval, returns exact source ranges and explicit evidence gaps, and checks whether citations are still current. Ranger uses it for repository context. Source is private; see my portfolio for project details.

Barestate — my work toward self-improving machines, including computer-use agents on Apple Silicon. Atom explores compact GUI grounding with a fine-tuned, quantized TinyClick/Florence-2 derivative, alongside structured accessibility targets. This is experimental work; grounding benchmarks and end-to-end task completion are different measures.

Synara: agent-thinking UI — a draft contribution in my fork of Synara. The change makes provider-exposed thinking and summaries visible and expandable across ten agent integrations, with streaming updates, interruption recovery, and separate child-agent identities.

The PR records 301 server regression tests, 191 transcript tests, and a Chromium UI test passing. Live authenticated provider sessions remain unverified; the contribution has not been merged upstream.

I founded SurfCash, a stablecoin payments product spanning multiple markets, and previously worked in developer relations at SOON. That experience informs how I build: start with a concrete customer problem, implement the workflow, and make the result inspectable.

I'm interested in applied AI and forward-deployed engineering work across agents, developer tools, and computer use.