I’m Matt, a computer science student at CU Boulder. I build AI tools for spatial creation and wearable context. My latest project is Matrix: a persistent world you can create and edit with AI, then enter from a browser, VR, or AR.

Portfolio · Résumé · LinkedIn · Email · Personal site

Describe what you want to build, work with Operator, then step into the result. Matrix connects a Three.js/WebXR world to Codex, procedural creation, and Blender asset workflows. The same saved world can be explored on desktop, in VR, or through AR.

• Create and edit: use voice or text, review changes, and refine objects with context from the current scene.
• Place in AR: select a point on a measured support surface, move a whole layout relative to the room, then save and reopen it.
• Keep a world running: a PC-hosted AI Citizens prototype explores persistent routines, needs, memory, and bounded interactions.

Released: v1.1.0, September 29, 2026. It adds active-turn steering, precise placement points, and room-relative AR layout controls. v1.2 work is in progress.

Watch the 2½-minute demo · Source and setup · Release evidence and limits

Stack: TypeScript, Three.js, WebXR, Python, Codex, Blender, and automated regression tests. Local services need setup; the release notes distinguish desktop/package checks from Quest wearer testing.

I contributed to Marinara Engine and Marinara Agents, a local-first AI roleplay and game platform.

My work connects the story we write during roleplay to a working MiniMax H3 animation pipeline. Roleplay narration becomes a storyboard; character sheets and personas ground who is in the scene; World Maps artwork grounds where it happens; generated keyframes establish each shot; and ComfyUI sends the grounded frame and motion plan into MiniMax H3 for animation.

The H3 flow is implemented and has produced a successful multi-shot clip with synchronized motion and audio. The packaged Storyboard prompts reuse that tested workflow. Animation pipeline · MiniMax H3 prompts

Roleplayers already create scripts every day through dialogue and narration. The next step is to let those stories become illustrations and animated scenes without losing the characters and locations that make the world recognizable.

• Character continuity: character-sheet artwork and persona data provide visual and written references for each character.
• Location continuity: World Maps artwork gives scenes a persistent visual reference for recurring places, including fandom settings.
• Story to animation: narration becomes a scene plan, keyframes, motion instructions, and then an animated clip.

The broader direction is an interactive anime / multiverse experience: combine characters, worlds, roleplay and game modes, make choices, and continue talking with characters who can respond in the context of recent adventures. That larger experience is the direction of the project; the working H3 animation pipeline is one piece already in place.

I’m also interested in how far local video models can go on consumer/pro hardware. Reaching Seedance-like action quality locally is an experiment and goal, not a claim of demonstrated parity.

Manfred is my wearable AI project. An Omi wearable captures daily audio for transcription and lifelogging; those transcripts can feed the personal LLM wiki I use to organize conversations, project notes, and recurring context.

I’m also connecting E09 smart glasses to ChatGPT Live for voice conversations and image uploads. The goal is to use the normal ChatGPT account experience—rather than a blank, stateless API session—so a wearable interface can benefit from context already available there while letting the user speak naturally and share what they are seeing.

Voice through the glasses works. Photo capture and delivery have been tested and are still being made more reliable. This is a prototype, not a finished always-on assistant.

Demerzel is the layer behind my other projects: daily AI briefs, LLM wiki ingestion from daily conversations, and management of the machines running local AI workloads.

The wiki keeps useful context from disappearing into old chats. Daily briefs surface what changed and what needs attention. Fleet tools track what is running, start and stop supported services, and keep workloads from competing for the same GPU.

I use the machines as a small managed compute pool so each project does not have to reinvent hardware control.

Separate autonomous coding/research workers are still being refined. Personal wiki content, recordings, credentials, and machine-access details remain private.

School of the Ancients: conversational learning with AI historical mentors across web and VR prototypes.

AI Radar: automated AI news/research collection and daily briefing.

NASA SUITS 2026 prototype: contributed to an experimental AI/VR assistant, rover-autonomy concepts, and augmented-reality interfaces for a student project.

Open to software and AI development opportunities alongside finishing my degree.