
Work I've Built
14+ years of shipped products across mobile apps, web platforms, and cloud infrastructure serving hundreds of thousands of users across 26+ countries.
- 78Projects Shipped
- 400K+MagLoft Readers
- 26+Countries Served
- 35M+API Requests / mo
Selected work across AI agents, mobile, web & infrastructure
Six projects that show range, from a desktop AI agent suite and enterprise mobile apps to the GCP backend behind 35M+ requests a month.


MagLoft API
2020–PresentMagLoft API Functions
2020–PresentComplete archive of 78 shipped projects
More projects24

AI Assistant
Personal
A macOS app that puts a fleet of AI agents behind a tray-resident robot face, Rust and Tauri 2 behind, React 19 in front, native NSPanels through Cocoa. Every agent drives a persistent Claude Code CLI session that calls back into an in-process MCP server. A brain mode works out what is worth doing today and gets on with it unasked, Voice Mode turns a spoken request into an action, and agents generate ideas, draft content, find leads, work the web and code on a review branch, under approval.
- Brain mode plans the day when nothing is being asked of it, works out what is worth doing, writes the plan and works it task by task, pausing for approval wherever a step reaches outside the machine.
- Voice Mode carries the whole context, the Knowledge Base, the open work and every agent, and acts through the agents' own tool surface, so a spoken request opens apps, drives the browser or starts an agent.
- The Idea Generator runs a Generate to Validate to Blueprint funnel with live web research, the Content Agent drafts and publishes, the Lead Generator finds and qualifies leads with outreach sent only on approval.
- The Web Agent drives a dedicated Chrome over the Chrome DevTools Protocol, scans a form into typed fields, fills it and reads every value back before the human gate.
- The Developer Agent runs a Claude Code session inside a git workspace, plans, builds, tests and commits to a review branch, shown in a native NSPanel with an edit-by-edit replay.
- Each agent runs on a persistent Claude Code CLI session over stream-json and calls back into an in-process MCP server on axum, so every tool stays local and the session stays warm between turns.



















































