Built in public, 3 stages so far

Human AI Interface

Hands, face and voice for AEC agents. Point at a wall, say "make it 300 thick", nod to accept. All from a normal laptop webcam.

Watch the stagesCode on GitHub

Stages

One stage a week, each with its clip and the LinkedIn post that went out with it. Open a stage to watch it in full.

  1. 03Week 3

    Orbit a building with one hand

    Pinch and drag and the model turns. A fist works as a clutch, two hands zoom, pan and turn, and a flick lets the view coast to a stop.

    Watch the clip

  2. 02Week 2

    Hands, face and a pinch you can see

    Everything runs in the browser from a normal laptop webcam: face mesh, both hands, a live pinch meter and a cursor that follows the pointing hand.

    Watch the clip

  3. 01Week 1

    Teaching my Rhino model to read my hands

    Before writing any code, a week of research: 20 AEC platforms, 17 vision libraries and 16 datasets, and one rule for the project. Build the hands, not the modeler.

    See the research

Build the hands, not the modeler

AI agents can already drive design tools, but talking to them still means typing into a chat box. HAI rethinks how you direct the agent, from a normal laptop webcam.

Download for Windows Read the code

Open source under Apache-2.0.

  1. 01

    Hands, face and voice

    Both hands as equal pointers, a nod for yes, a shake for no, and speech that stays on the laptop. Nothing fires until you engage, so talking with your hands doesn't move the model.

  2. 02

    Private by design

    Camera video is processed inside the browser on your machine. It is never recorded or uploaded; only gesture events reach the app.

  3. 03

    Your own agent, real geometry

    HAI drives your own signed-in Claude as a small design team, and keeps the precise modelling in tools like Rhino.