I build local-first AI agents, voice interfaces, and developer tools that turn model capability into safe, testable products. I bring 15+ years of software experience to the work—from architecture and delivery to the details of implementation, verification, and release.
Portfolio · Featured AI project · Open-source contributions
A local-first Gmail cleanup workbench with optional OpenAI classification, human review, and reversible actions.
- Go API, React workbench, Gmail OAuth, SQLite, and the OpenAI Responses API.
- Metadata-only Gmail reads and bounded AI classification batches.
- Preview-and-confirm flows for Trash and standards-based one-click unsubscribe actions.
- Demo data, guided onboarding, responsive UI, automated tests, and documented trust boundaries.
|
A local Windows supervisor that routes Telegram text and voice through OpenAI or tracked Codex tasks. Focus: agent routing, SQLite memory, background jobs, recovery, and local execution boundaries. |
A voice-first Windows desktop agent built with Electron and OpenAI Realtime. Focus: live voice, tool approvals, resumable task planning, and browser interaction. |
|
A Rust and Tauri desktop runner for persistent Codex workflows. Focus: event-stream parsing, SQLite thread history, cancellation, retry, and typed lifecycle state. |
A 20-season college-football simulation with 70 fictional programs and deep browser verification. Focus: product depth, simulation systems, TypeScript, cross-browser tests, and long-loop reliability. |
My working loop is Frame → Build → Verify → Ship:
- Turn ambiguity into constraints, risks, and a testable first slice.
- Keep architecture and interfaces understandable enough for another person to own.
- Validate with tests, screenshots, telemetry, and real usage.
- Release software that can be explained, operated, and improved after launch.
The projects above are working labs for agent reliability, human approval flows, persistent memory, AI-assisted development, browser performance, and evidence-led quality.
My upstream work includes merged framework and documentation improvements across:
- Laravel framework #53109
- Laravel framework #53097
- PHP documentation #3135
- Laravel documentation #8123
I am currently returning to regular open-source participation in the AI agent, MCP, TypeScript, Rust, and developer-tooling ecosystems.
- Reliable local and desktop agent systems.
- Human-in-the-loop controls for tools that can change external state.
- Evaluation, observability, memory, and recovery for long-running agent work.
- Practical uses of OpenAI, Codex, TypeScript, Go, Rust, React, Electron, and Tauri.
I am always interested in thoughtful collaboration around applied AI, agent infrastructure, developer experience, and ambitious interactive systems.