Magi Vision

A computer-vision platform for combat-sports gyms

Magi Vision pairs computer vision with purpose-built gym hardware, giving coaches and athletes a way to see what's actually happening on the mat — in real time, without a camera crew or a manual tagging pipeline. As Co-Founder & Design Chief of Seiryoku Zenyo Labs, I own product and experience design end to end: the hardware's physical presence in the gym, the software layer coaches and athletes actually use, and the story we tell investors and early partners.

Every pose model in production assumes bodies stay separate. Grappling is the sport where they don't. Systems that can call a sprint split or a jump shot still can't tell you what just happened on the mat — Magi Vision is built for exactly that gap, from the camera on the wall to the event log.

Design leadership, 0 → 1

• Set product and experience design direction across the full team
• Design across both hardware and software surfaces — the physical unit, the mounting and gym footprint, and the software coaches and athletes use
• Build and present the deck that goes in front of investors
• Own the product surface — the capture software an operator runs and how the system reads from the mat side
• Design the operator HUD hands-on, iterating against the live build rather than static mocks, and run the capture software myself on capture days
• Partner with Shaker Zaman (CEO, Operations & Engineering), who builds the capture side and reviews changes before they ship

Grappling breaks computer vision

From the first grip, two athletes occupy one silhouette. Limbs vanish behind torsos, keypoints jump between people, and the tracker rebinds identities in the exact half-second the throw resolves. On the ground it inverts: almost no motion, almost all the meaning.

• Occlusion — the decisive frames are the most hidden ones
• Identity swap — thrower and thrown trade skeletons mid-technique
• Ground phase — minimal movement, maximum consequence
• Ground truth — labels require someone who knows the sport

Cyberpunk meets ritual

The visual language draws on cyberpunk and religious motifs — high-contrast, ceremonial, built to feel like a tool an athlete respects rather than a consumer gadget. The mark is a pentacle-derived medallion paired with a liquid-metal magi wordmark; brand moments are set in Unbounded, and everything the system reports is set in IBM Plex Mono so the interface reads like instrumentation. Color is near-black with two signals: cyan and red.

The company takes its name from the judo principle of seiryoku zenyo — maximum efficiency, minimal effort — and the product follows it: nothing spent on work that does not move the outcome.

THE SITE

The public site is where the design direction gets its clearest statement: a chrome-and-liquid-metal magi lockup on near-black, the thesis line right under it — "Grappling Software providing machine verification of human energy" — and a scroll that walks from company to problem to team to demo to the dojos the footage comes from.

magivision.io

PALETTE

Screens & hardware

Magi Vision is the whole chain, from the camera on the wall to the event log. An operator starts a session and watches state commit as it happens — grip, kuzushi, the throw. The dashboard aggregates what came off the mat and hands back the highlights. The demo on magivision.io is the build, running.

The clip below is me in a white gi and green belt throwing harai goshi, captured and tracked by the software — the exact case the system is built for: two bodies in one silhouette, resolved at the half-second the throw commits.

Live, and capturing on real mats

The public site and interactive demo are live at magivision.io. Footage comes from three partner dojos that let us point cameras at their rounds: Kano Martial Arts in Manhattan, Krystek Martial Arts in Queens, and The Dojo in Brooklyn.