Python developer building production AI on the Claude API: retrieval systems, LLM agents, and an MCP server any model can call. I ship polished developer tools, and I write about the craft. AR/VR specialist at Meta by day.
By day I work in AR/VR at Meta, where I help customers discover Ray-Ban Meta smart glasses and the Meta Quest headset family; spanning live product demos, consultative selling, and hands-on technical support across the full spatial computing lineup.
Outside of that, I build under the RNVizion banner, lately centered on AI. I've shipped a retrieval-augmented assistant over my own writing and an agentic publishing pipeline, both on the Claude API, plus rnv-color-mcp, a server published to the official Model Context Protocol registry that any LLM can call, and AIII, an open standard for AI agent identity with a reference implementation. They sit on top of the RNVizion toolkit: five professional desktop apps for text transformation, color management, and icon generation, each with full CLI support, multi-theme UIs, cross-platform CI, and comprehensive test coverage.
I also write: an ongoing series on building with intention in the AI era, part of a wider renaissance-man practice that runs from software into design and craft. My background spans game programming, IT support (Google IT Support certified), technical content development, and customer-facing work in regulated industries. I'm targeting remote AI Engineer and Solutions Engineer roles where building, communicating, and shipping production systems all matter.
Artificial Intelligence Identification Initiative: an open proposal and Apache-2.0 reference implementation for AI agent identity. An L1–L3 identity-and-authorization layer for MCP servers, built on one principle: resolve or refuse, never guess. Holder-of-key verification, declarative per-tool authorization, 46 tests with property-based eval gates, and a runnable FastMCP demo.
A retrieval-augmented chatbot over my published writing. It embeds every blog post into a vector index, pulls the passages relevant to a question, and answers only from them; grounded in the source, guardrailed against off-topic prompts, and honest when the corpus doesn't cover something. Built on the Claude API and deployed live on Hugging Face.
My desktop color engine, lifted out and exposed once as a remote MCP server; an AI assistant calls it today, my own apps call the same backend tomorrow. Nine tools cover mixing (including Kubelka-Munk paint physics), conversion, harmony, Delta-E difference, and WCAG contrast, plus a persistent palette memory; it resolves hex, CSS, and brand color names, and refuses an unknown color rather than guessing. Published to the official MCP registry and running live on Hugging Face.
11 transformation modes across 9+ file formats with regex builder, folder watching, compare view, and a full CLI. Backed by 786 tests at ~76% coverage running on Linux and Windows CI.
View on GitHub →
Professional palette management for designers and developers: 16+ export formats, 7 color mixing algorithms, and built-in WCAG accessibility tools.
View on GitHub →
Screen color extraction with a magnifier, WCAG contrast checker, and color harmony generation. Built for designers and developers who need precision color workflows.
View on GitHub →
Multi-resolution ICO creation from PNG, SVG, and ICO sources. Cross-platform exports for Android, iOS, and favicons with batch processing, folder watching, and a full CLI.
View on GitHub →
A desktop paint-mixing application simulating real-world color mixing using Kubelka-Munk theory. Bridges physical color theory with digital workflows.
View on GitHub →
An ongoing series on what AI can be when it's built to deepen real work, not replace the thinking behind it. Part of a wider renaissance-man practice; the same instinct that ships the tools writes about the craft.
On constraint as a creative force: why the gap between what you have and what you need is an invitation, not a wall.
Read it →The human decision that turns software from working into loved; the soul that speed alone can't generate.
Read it →Strategic economy of effort: doing less on purpose, so the energy lands where it actually moves the work.
Read it →The RAG demo, running live and embedded right here. Ask it about my work; it answers only from published sources, and tells you when something isn't covered. It's on free hosting, so it naps when idle: it starts waking the moment this page loads, and should be ready by the time you scroll down to it.
Open to remote AI Engineer, Solutions Engineer, and Developer Advocate roles. Always happy to talk production AI, developer tooling, or the craft of building things that feel right.
