First projected-reality prototype
Mapped a live interface onto moving physical objects.
CLAUDE CODE FOR PHYSICAL WORK
A lamp that projects AI guidance onto robotics and hardware work.
The system sees the bench, understands the task, and points to the exact part. No headset. No screen switching. Both hands stay on the work.
BUILT BY A FORMER APPLE VISION PRO INFRASTRUCTURE ENGINEER×FORMER HARDWARE TRADES WORKER
RSVP FOR THE AUG 20 DEMO
San FranciscoHands-on · 2026
01 / FOUNDER INSIGHT
BUILT FROM BOTH SIDES OF THE PROBLEM
The display belongs above the bench, not on your face.
Two years building Vision Pro infrastructure showed the limits of a display people remove. Years in the hardware trades showed the same gap from the other side: instructions live away from the work.
Apple Vision Pro infrastructureHardware tradesBuilding full-time in SF
FOUNDER INTRO · 01:00
Vision Pro infrastructure
Saw headset friction from inside spatial computing.
Used to build the system itself
The prototype helps build the next hardware iteration.
02 / PROBLEM
THE BROKEN WORKFLOW
AI stops at the edge of the screen.
Builders still stop, find a video or spec, re-orient to the part, do one step, then look away again. Screen-bound agents cannot point at a physical connector, fastener, or solder pad.
03 / PRODUCT
A CLOSED LOOP FOR HANDS-ON WORK
Put the interface on the thing itself.
One lamp combines cameras, a vision-language model, and a projector. It sees the task, guides the action, and checks the result.
Observe
Cameras track the bench, every part, every tool, and your hands.
Reason
A vision-language model understands the task and its current state.
Project
The next action lands in light on the exact part that needs it.
Verify
The system checks the result before moving to the next step.
04 / EVERY PUBLISHED VIDEO
THE WORKING SYSTEM, NOT A RENDER
See the loop from every angle.
These are the full set of videos published across the original site: physical registration, object awareness, perception, room understanding, and ambient interface studies.
WORKING PROTOTYPE
01 / VIDEO ARCHIVE
Instructions stay registered to the work.
A live instruction layer follows the page as it moves. The same registration loop puts assembly guidance directly onto parts and tools.
WORKING PROTOTYPE
02 / VIDEO ARCHIVE
The workbench becomes the interface.
The system sees the object, measures it, and projects the result beside it. The operator never has to translate a screen back to the bench.
WORKING PROTOTYPE
03 / VIDEO ARCHIVE
Measurements land on the object.
The camera reads dimensions from the physical scene. The projector puts the answer where it is useful: on the table, next to the thing.
WORKING PROTOTYPE
04 / VIDEO ARCHIVE
Projected guidance locks to real surfaces.
This earlier measurement pass shows the core loop in its simplest form: see the bench, understand the geometry, and answer in light.
THE SYSTEM
05 / VIDEO ARCHIVE
One rig watches the whole workspace.
An overhead camera keeps the working surface in view while the projector turns it into a shared, hands-free display.
PERCEPTION PIPELINE
06 / VIDEO ARCHIVE
It tracks the person doing the work.
Pose, motion, and presence are read from the live scene so guidance can react to the operator, not just a fixed instruction script.
WORKING PROTOTYPE
07 / VIDEO ARCHIVE
Place an object. The system responds beside it.
The system recognizes what entered the workspace and projects context next to the physical object instead of opening another window.
05 / INITIAL WEDGE
START WHERE VARIABILITY HURTS
High-mix robotics workbenches first.
Robotics teams
Small hardware teams assembling, debugging, and repairing changing systems at a shared bench.
Guide variable work
Show the next action, catch mistakes, and verify completion when the task changes too often to pre-author.
06 / UNDERSTAND THE SPACE
BEFORE THE SYSTEM CAN POINT, IT HAS TO SEE
The physical scene becomes context for the agent.
PERCEPTION PIPELINE
08 / VIDEO ARCHIVE
The system first builds a spatial view of the room.
The camera maps the physical workspace before any instruction layer reaches a surface.
PERCEPTION PIPELINE
09 / VIDEO ARCHIVE
It reads the geometry.
Walls, floors, furniture, and usable work surfaces are reconstructed from the live scene.
PERCEPTION PIPELINE
10 / VIDEO ARCHIVE
It labels every surface.
The system separates table, chair, wall, floor, and door so projected UI lands on the right physical plane.
PERCEPTION PIPELINE
11 / VIDEO ARCHIVE
The same model works across unfamiliar rooms.
The goal is adaptive guidance: understand the live environment instead of asking an engineer to pre-author every possible setup.
07 / BEYOND THE BENCH
THE INTERFACE CAN LIVE IN THE ROOM
Only the useful light appears. Everything else stays physical.
AMBIENT UI STUDY
12 / VIDEO ARCHIVE
A blank wall can become useful without becoming a screen.
Projected content occupies only the light it needs. The rest of the room stays the room.
AMBIENT UI STUDY
13 / VIDEO ARCHIVE
Quiet by default.
It can sit in a room like a framed print, then become an interface only when the task calls for one.
AMBIENT UI STUDY
14 / VIDEO ARCHIVE
Reach into the interface.
Controls appear on the physical surface and respond to a real hand. Nothing is strapped to the person using it.
AMBIENT UI STUDY
15 / VIDEO ARCHIVE
Any surface can carry the next useful action.
The workbench is the first wedge. The same projected interface can extend to cooking, repair, and shared physical spaces.
ORIGINAL SITE REEL
16 / VIDEO ARCHIVE
The interface leaves the screen.
The system puts guidance in the same physical space as the hands, tools, and objects it is helping with.
08 / NEXT PUBLIC CAPTURES
FOOTAGE NOT YET PUBLISHED
Three task demos next.
The archive above is real published media. These task-specific clips are clearly marked until the August 20 capture replaces them.
Assembly guidance
Projected steps on the exact fastener and part as the build moves.
Soldering walkthrough
The pad, component, and orientation highlighted while both hands stay on the work.
Cooking with intelligence
Ingredients, timing, and the next action projected onto the counter.
09 / PROOF + MOMENTUM
PRE-LAUNCH, WORKING PRODUCT
The prototype is used, not just shown.
No revenue or external-user claim yet. The strongest proof today is a working system and daily founder use.
10 / PUBLIC DEMOSAN FRANCISCO · AUGUST 20, 2026
COME USE THE PROTOTYPE
Build hardware and cook with intelligence.
We rented a house in San Francisco for a public, hands-on demo. Bring a task, try the prototype, and meet the robotics teams and builders shaping the first pilots.
WHEN Thursday, August 20, 2026WHERE San Francisco · location sent to guestsFOR Robotics teams, hardware builders, angels