PTENES
3D learning world · Babylon.js

A 3D world for teaching anything.

Three stations, one guide mascot, and a one-minute World Tour. Each station shows an idea in action. The first world teaches three habits for using AI in everyday life.

astra3dany banner: a 3D world for teaching anything
What it is

A lesson you can watch happen.

It’s not a video or a slide: it’s a miniature stage in the browser. The learner walks around, clicks a pedestal, watches the machine work, and reads alongside it what’s happening and what to remember.

🎬 Rhythmic World Tour

One-minute guided tour: arrive (3 s) → move closer (3 s) → demonstrate (6 s) → hold on the result (5 s) at each station, with a card synced to the action. Pause, restart, change the speed, exit.

🕹️ Free exploration

Orbit, zoom in, pan, reset. Click a pedestal to open the station, run and replay the demonstration at your own pace. Keyboard, mouse, and touch.

🔁 Change the subject, keep the recipe

All the text and stations live in a single configuration file. Teaching something else means editing that file; the camera, cards, and tour stay the same.

How it works

One station = one principle = one visible change.

Follow the field guide recipe Build a Learning World (Mark Kashef, Early AI Adopters): six parts, a fixed camera pace, and an acceptance checklist. The current world teaches three habits:

01 · The gate→ 02 · The drawers→ 03 · The review→ Recap
01

Confirm before acting

A request arrives at a closed gate and goes for review. The second one passes the check: the gate opens and the wheel (the action) turns. Ask the AI to confirm before sending, paying, or deleting.

02

Give context where AI reads

An instruction card goes into the project drawer, and the two colleagues stay connected. Keep the instructions in one place; everyone, including the AI, reads the same thing.

03

Check the source

The response format checks out, but the source says 20 + 30 + 40. The total is 90, and the date remains open without a source. A well-formatted answer isn’t necessarily a correct answer.

Prerequisites

To use it: just a browser. To edit it: Node.

The site is 100% static. There’s no backend, API key, database, or Blender involved.

🌐 Browser with WebGL2

Chrome, Edge, Firefox, or Safari 15+. A dedicated GPU isn't required, but it helps. Works on mobile.

# open the published world
https://inematds.github.io/astra3dany/app/

🟢 Node.js 20+ and npm

Just to run locally or change the topic. Developed and tested with Node 24.

node -v   # v20 or newer
npm -v

📦 Git

To clone the repository. The Vite and Babylon.js versions are pinned in the package-lock.

git clone https://github.com/inematds/astra3dany.git
User guide · step by step

From scratch to your own world.

Steps 1 and 2 are for anyone who just wants to visit. From step 3 on, they’re for anyone who wants to run it locally or teach something else.

1

Take the World Tour

Open the world and click the orange button World Tour. In one minute, the camera visits the three stations, runs each demonstration, and holds on the result while the card explains it. At the end, the recap with the three habits appears.

https://inematds.github.io/astra3dany/app/   # Pause/Resume · Restart · 0.75× · Exit (or Esc)
2

Explore at your own pace

Click a pedestal (or buttons 01/02/03). Take a closer look and run it moves the camera closer and runs the demonstration; Replay run it again; Wide view backs away. Outside the tour, the camera is free:

ActionMouseKeyboardTouch
Orbitdragarrow keysone finger
Zoom in / outscroll wheel, + and − buttons+ / −pinch
Panshift + drag, right-click—two fingers
Return to framingResetRReset

Add ?diagnostics to the URL to see FPS and resolution. The button Reduce motion replace the camera animations with hard cuts (and respect the system preference).

3

Run locally

Clone, install the locked versions, and start the development server.

git clone https://github.com/inematds/astra3dany.git
cd astra3dany/game
npm ci            # Vite 8.2.2 + Babylon.js 9.25.0, exactly as in the lockfile
npm run dev       # http://127.0.0.1:43220
4

Change the topic

Edit game/src/mundo.config.js: title, intro, audience, mascot, and the three stations (name, principle, explanation, sequence, per-second demonstration notes, takeaway, and camera poses). All site text comes from there.

// game/src/mundo.config.js — one station
{
  id: 'portao', demo: 'portao', x: -3,
  principio: 'Confirme antes de agir',
  titulo: 'Uma checagem que pode parar uma ação.',
  sequencia: ['Pedido', 'Checagem', 'Ação'],
  observacoes: [[1.5, 'O pedido se aproxima…'], [2.8, 'Checagem falhou…'], [6, 'Checagem passou…']],
  takeaway: 'Peça pra IA confirmar antes de agir: enviar, pagar, apagar.',
  camera: { wide: {…}, closeStart: {…}, closeEnd: {…} },
}

Need a new mechanic? Create a module in game/src/estacoes/ with { reset(), update(demoTime), foco, pick } and register it in DEMOS in the main.js. The three existing ones serve as templates.

5

Test and inspect

The tests read the config and check the duration, the 5 s hold, and the phrases for each phase. Then check in the browser: wide view, close-up, halfway through the demonstration, and the result, on desktop and mobile.

npm test          # 7 timeline and caption tests (node --test)
npm run build     # generates ../app/ (static, relative paths)
npm run preview   # http://127.0.0.1:43221 — inspect the build that will go live
6

Publish

Commit the folder app/ and push. GitHub Pages serves from the root of the main branch; the same build runs on Vercel, Netlify, or Here.Now by pointing to app/.

git add app && git commit -m "build" && git push
# app:  https://inematds.github.io/astra3dany/app/
# guide: https://inematds.github.io/astra3dany/guia/
Examples

What appears on screen.

Inspection screenshots (desktop 1280 px and mobile 390 px). The full report is at evidence/revisao.md in the repository.

Wide view of the stage with the three pedestals and the Any mascot
Overview: three numbered pedestals, the Any mascot, and the intro panel with the route through the stations.
Close-up of the gate station during the demonstration
Station 01 in explore mode: the blocked request in the review tray, the second request going through, and the synchronized caption.
World Tour at the verification station with the card beside it
World Tour at station 03: the scene is rendered beside the card, which shows the takeaway and the verified values.
Mobile layout with the scene on top and the card below
On mobile, the tour puts the scene on top and the card below; nothing covers the demonstration.
Known limitations

What it still doesn’t do.

So you don’t find out at the wrong time.

🤖 Code-based mascot

Blender wasn't available on the build machine, so Any is generated procedurally. A GLB made in Blender can be plugged in through the config (legs Perna1..4 get the upper level).

📦 ~5.8 MB bundle

1.2 MB compressed, because of Babylon.js. The first load on 4G takes a few seconds; there is no code-splitting yet.

📱 Tested only in headless mode

Desktop and mobile were inspected in Chromium without a GPU. No physical device, Safari, Firefox, or real touch gestures have been tested yet; the measured fps don't represent a real device.

🔇 No audio, quiz, or live AI

The demonstrations are deterministic animations. Voice, music, and questions are planned for a future version.

🌐 One language

Texts in Portuguese, in the config. There’s no i18n; translating means copying the config.

♿ Partial accessibility

Keyboard-navigable buttons and panels, tour card in aria-live, reduced-motion. The 3D scene itself has no description beyond the cards.

Roadmap

Next phases.

In the order that matters to learners.

Done
World with three stations, a tour, free exploration, recap, mobile support, tests, and evidenceComplete field guide recipe applied to "AI in everyday life".
Phase 2
Blender mascot and voiceRigged GLB asset in place of the procedural mascot; TTS narration for the captions.
Phase 3
More worlds, same engineOne config per topic (e.g., prompts, automation, security) and a world selector at the entrance.
Phase 4
Quiz and progressOne question per station and a recap of what the person got right.