Open any project for the architecture and the decisions behind it.
A personal assistant with local memory and retrieval, scoped tools, and real-time voice. Claude handles conversation by default; selected prompt context goes to the hosted provider. Local Ollama handles bounded utility work.
47 services in family production with single sign-on, encrypted secrets, immutable offsite backups, and full observability. I stress-tested it with an LLM-driven penetration test that surfaced a real CVSS 9.3 vulnerability.
An offline, single-player medieval settlement simulation in Godot and C#. Playable from source; commercial release acceptance is still open. The source isn't public yet.
I own the infrastructure and deployment for three products, build the internal tools our marketing team runs on, and built the MCP server that puts our CRM inside Claude.
Four public tools, with source, a release to try, and one decision behind each. These releases stand on their own.
A terminal assistant for authorized security testing, with local inference by default.
Design choice: Manual mode is the default: the model proposes commands, and the operator chooses when to run them.
A local coding agent, with its Python CLI in one standard-library file and a separate llama.cpp model server.
Design choice: Keeping the CLI free of Python package dependencies makes its implementation easier to inspect.
A human-guided music studio powered by ACE-Step 1.5, with local generation and an opt-in remote GPU lane.
Design choice: Immutable takes preserve the history of a song while its composition plan stays editable.
An image and video studio with local and remote GPU lanes.
Design choice: Each generator declares its capabilities and limits so the controls match the selected model.
In development: Wizzy, the language in The Wizard's Ink, remains private and pre-release. Public project status
A deployed web app for a tabletop campaign: player journals, a GM hub with maps and trackers, real-time shared state.
A fast, dependency-light marketing site delivered for a real advisory-firm client, deployed on Netlify.
An end-to-end ML prediction pipeline and a validated survey study analyzed with EFA, CFA, and SEM.
Engineering, teaching, research, and enterprise IT.
I build software end to end, across apps, infrastructure, AI systems, and data. I pick up new tools fast, use AI to move quickly, and like the whole range of the work rather than one narrow slice of it.
In the last year that meant a personal AI assistant with local memory and hosted inference, a 47-service self-hosted platform I hardened with an LLM-driven pentest, Smoke Rising, a settlement simulation in Godot and C#, and ownership of the infrastructure and internal tooling behind an AI startup's site and CRM, including a production MCP server that connects Claude to that CRM. Different domains, one instinct: learn the tool, build the thing, make it survive real use.
The other half of me is a teacher. I taught 90+ students solo in an intro information-systems course, holding a 93% class exam average with no faculty oversight, and before that spent two years training and leading people, turning newcomers into confident, independent leaders. The work I care about most is not just building something that works, but making sure someone else can own it after I hand it off.
Email is the fastest way to reach me. Résumé and profiles are below.