Prototypes,
not slide decks.
Solution Architect who answers technical questions with working software. 30+ AI projects across presales, infrastructure, and agent tooling: built, shipped, and demoed live.
How I work
Prototype first
A running container image answers the question a slide only raises. Most ideas here went from conversation to running code in hours, not sprints.
Live generation
Atlas, my positioning agent, builds a role-tailored package in front of you, in minutes. Watching it run is the demo.
Evidence over marketing
Every claim on this page links to a repo, a screenshot, or a running system. Nothing here is a mockup.
“Jack Dorsey says Block employees now bring prototypes, not slide decks, to meetings.”
My reply in that thread, a few hours later, was not a comment. It was a working container image.
Selected work
05 case studiesThe headline pair: Inception builds the coding agent; the Skyline proof of concept applies it. It works.
Inception
An open-source multi-agent framework that automates the back half of software development. A finished requirements set goes in; built, tested, running code lands in GitHub. It runs as an orchestration layer on Claude Code and stays model-agnostic by intent. Its coding harness passed its first real test: it scaffolded a production-grade feature (8 new files, 4 edits to shared code) in about 12 minutes, and the result runs live.
View the repo
Trilio inside Canonical Skyline
Canonical’s Skyline dashboard is the upstream successor to OpenStack’s Horizon, and it has no plugin API. This proof of concept forks the console and adds Trilio backup workflows that drive real product APIs against a live OpenStack cluster. The read path is proven end to end: the new Trilio Backup panel lists real workloads from the live cluster. The module was built with the Inception coding harness; the way it was built is part of the deliverable.
Del & Robin
Two OpenClaw agents run around the clock on a Hetzner server: Del, my travel agent, and Robin, my AI event coordinator. Each has its own Telegram bot, email account, workspace, memory, and isolated cron jobs. Del watches hotels and trips; Robin subscribes to AI events and newsletters and triages her inbox into consolidated Telegram updates. del-infra is their deployment and operations layer: Ansible and Terraform in two layers, daily encrypted backups, and a published pitfalls guide, generic enough that anyone can deploy their own agents from the same playbooks. They cost a couple of dollars a day, they earn it daily, and they preview what comes next: always-on coding assistants.
View the repo
Salesforce Agent
A compact MCP server (about 150 lines of Node) that connects Claude to live Salesforce. Four tools cover query, create, update, and describe; a packaged skill adds propose-then-confirm guardrails, and there is no delete tool by design. Shipped end to end in two days, proven against the production org, and now part of daily presales work: opportunities, tasks, and custom deployment records updated from chat instead of the Lightning UI.
Atlas
A positioning agent over a knowledge base of project overview files. It generates a role-tailored resume, narrative, and portfolio on demand. This website is one of its artifacts; the content you are reading came out of that knowledge base.
The day job, with AI
Customer qualification agents and solution briefs run inside live engagements, so they stay blurred. Names and proprietary detail are removed; ask about them in a live demo.