Ludumina
Join the beta

SRC / IN-ENGINE CAPTURE ROMAN ARENA — IN DEVELOPMENT

SCIENCE·ART·HISTORY

Ludumina Worlds

Knowledge you can walk into.

01 Roman Arena Cross a Roman arena.

AI-guided immersive learning worlds for children aged 8–14.

Join the beta

Five worlds·Ships September 2026

An AI tutor that
teaches as you go

02 / Real-time guidance layer

Lumi is our proprietary real-time guidance layer for immersive 3D learning spaces. It watches what a learner actually does inside the world, turns that behaviour into a structured learning-state snapshot, and answers the snapshot — in context, inside the experience.

Signal Snapshot Guidance

Guidance pipeline

  1. 01Approach
  2. 02Select
  3. 03Retry
  4. 04Stall

Scroll to advance

Showing moment 03 — motion reduced

01

In-world capture

World
Living Cell — in development
Mission
Assemble D-glucose
Observed
Head pose · hands · world objects

Genuine footage from the build. The learner grabs a molecule and snaps it into place themselves — every one of those actions is a signal.

02

Behavioural signals

  • Movement0.78
  • Object proximity0.62
  • Object selection0.05
  • Wrong attempts0.02
  • Hesitation0.12
  • Task sequence0.25
  • Mission progress0.10

One behaviour each, scaled against its normal range for this mission — not a score.

Learning-state snapshot

step
1 / 4
holding
—
nearby
Compare bench · GLUT1 pocket
wrong_attempts
0
idle_for
0.4 s
guidance_given
0
03

Guidance in context

Lumi In-world tutor

  • There’s a reference model on the bench you just passed. Look at it before you pick anything up — you want to know the shape you’re aiming for.

    • movement
    • object proximity
    • mission progress

  • You’re holding a hydroxyl group. Hold it up against the reference — does the oxygen sit where yours does?

    • object selection
    • object proximity

  • Three tries into the same pocket. It isn’t the force. Turn the group and watch which face lights up — the fit is about orientation.

    • wrong attempts
    • object selection
    • task sequence

  • Still thinking? You already know ATP carries three phosphates. You’re adding the last one — so it goes on the end of the chain, not the middle.

    • hesitation
    • guidance already given

By design A nudge before an answer. Guidance already given is part of the state, so Lumi does not repeat itself — it escalates.

Disclosure Recorded gameplay on a loop, not a live feed. Living Cell and the guidance layer are in development; the signal, snapshot and guidance panels illustrate how the mechanism works on a real mission.

Not a chatbot
bolted onto a game

The difference is not the model. It is what the model can see. Same learner, same moment — here is the state each one is answering from.

Assistant bolted on

Sees only what is typed or said.

step
—
holding
—
nearby
—
wrong_attempts
—
idle_for
—

It can explain the chemistry. It cannot know you are holding the wrong group, in the wrong orientation, for the third time.

Lumi, grounded in game state

Reads authoritative state from the world itself.

step
1 / 4
holding
—
nearby
Compare bench · GLUT1 pocket
wrong_attempts
0
idle_for
0.4 s

So it responds to what the learner is doing — not only to what they thought to ask.

The environment becomes the lesson, the practice surface, the feedback system, and the assessment layer — all at once.

Ludumina Labs — design principle
  • The mechanicGrab a molecule. Snap it into place yourself.
  • Why it sticksYou build it, so you remember it.

Ludumina
Worlds

  • Science
  • Art
  • History

Five playable in engine  /  one planned

Ludumina Worlds ships September 2026

In-engine capture from Living Cell: a player stands on the lipid bilayer inside the cell with a glowing ball-and-stick glucose molecule at hand height. Candidate molecules float above, and the mission card reads: Cellular Energy, Molecule Hunt — grab the glucose you identified and dock it in the outward-facing GLUT carrier.
01 Science Playable in engine

Living Cell

“Step inside a living cell.”

Walk into the cell, pick a molecule up with your hands and run cellular respiration a step at a time — while Lumi watches what you are doing and guides you from inside the world.

  • Grab
  • Build
  • Be guided
Join the beta
In-engine capture: an orbital station against black space — a ring truss, a pressurised module and a robotic arm lit hard from one side, with an instruction panel floating alongside it.
02 Science

Orbital Station

“Repair a station in orbit.”

Truss, pressurised modules and a robotic arm in open space — with the repair called out step by step in-world, on a panel that floats beside the job.

Playable in engine
In-engine capture: the Athena Parthenos standing in gold inside the cella, Nike perched on her outstretched hand, the hall lit warm by braziers.
03 Art

The Treasures

“Meet the treasures up close.”

The cella by brazier light — the Athena Parthenos in gold, Nike standing in her hand.

Playable in engine
The project's model of the intact Parthenon, lit from above: the full tiled roof, the painted frieze running under the cornice, and the colonnade below it.
04 History

The Parthenon

“Walk the Parthenon while it stands.”

Ground level, roof still on, and Iktinos the architect on site to be asked about it.

Playable in engine
In-engine capture: a Roman amphitheatre — banked terracotta seating above a wooden arena floor, with the cross of the hypogeum showing through the boards.
05 History

Roman Arena

“Cross a Roman arena.”

Banked seating, the wooden floor, and the hypogeum cross running underneath it.

Playable in engine
Render of the project's anatomically detailed brain model: cortical surface with the sulci picked out in magenta, cerebellum and brainstem below. A source model, not gameplay.
06 Science

Human Body

No trailer line — there is no footage to quote.

Around 200 named structures, cortex to cranial nerves. The model exists; no world has been built from it yet.

Planned No loop — nothing built

Status Five of these worlds are playable in engine today, and every hover loop on this page is a capture from live play. So is every still except two, and both say so on the plate: the Parthenon still is the project’s own model of the building rather than a frame of it being walked through — its loop is the live capture — and Human Body is not playable at all. That one is planned only; its card shows the anatomical source model, around 200 named structures, that no world has been built from yet. None of the five has shipped: the release date the company states is September 2026.

Build Log

Built in the open — shipping September 2026

Ludumina Labs is a company mid-build, not a company with a back catalogue. Nothing has shipped and there is nothing to quote — but there is a date, and it is ours: LUDUMINA WORLDS is due September 2026. What follows is the work itself, and every entry carries the state it is actually in.

In-engine capture of a Roman amphitheatre: banked terracotta seating on the left, the planked wooden arena floor with the dark cross of the hypogeum at its centre, and a learner standing out on the sand at the near edge.

01 Playable History

The Roman arena is crossed on foot — up the banked seating, across the wooden floor, over the dark cross of the hypogeum below it. It is one of four worlds now running in engine alongside Living Cell, each of them captured with a learner inside.

That figure on the steps is the whole point. This is a capture taken while somebody played, not a camera move flown over a model — which is why every entry below states plainly what state its subject is in, and why the states are ranked.

In-engine capture: a learner stands among the fluted columns of the intact Parthenon, facing an NPC whose name floats above him reading Iktinos, Architect.

02 Walkable History

Walked at ground level, intact, from between the columns rather than from a helicopter. Under the colonnade there is a man with a name over his head: Iktinos, Architect — the one who drew it.

An earlier entry in this log called the Parthenon a source model. That was true when it was written and is not true now: the reconstruction has been taken into the engine, given collision, a way in and an inhabitant. It stopped being the thing you look at and became the thing you are standing in.

In-engine capture: an orbital station lit hard white against black space — pressurised cylindrical modules, a truss ring and a robotic arm — with an instruction panel at the lower left.

03 Playable Science

Pressurised modules, a truss spine and a robotic arm held against open space. The science world sets a repair rather than a tour: read the instruction, find the fault, fix it where it is.

The instruction sits out in the world with the learner instead of in a menu behind it — the same principle as the cell, applied where the consequences of getting it wrong are more obvious to an eleven-year-old.

In-engine capture: the Athena Parthenos standing in gold inside the Parthenon's cella, a winged Nike held out on her open right hand, Doric columns and a coffered ceiling around her.

04 Walkable Art

Inside the cella, by brazier-light: the statue in gold, Nike standing in her outstretched hand. The art world is about meeting the object rather than reading the label beside it.

The reason to walk into the temple is that the statue is the size of a building. A photograph of it flattens that out; standing at the foot of it does not. This is what the room looks like from where a learner is put.

In-game capture from Living Cell: a glucose molecule held over the open binding pocket of a GLUT1 transporter, with an in-world label reading GLUT1 — inward-open.

05 Playable Prototype

The docking mechanic in Living Cell is playable, and captured. A learner inside the cell picks up a glucose molecule and seats it in hexokinase's binding pocket, with Lumi giving instruction in-world while it happens.

None of it is a pre-rendered sequence. The learner walks, grabs and mis-seats the molecule under their own control, and the correction arrives in the world at the moment of the attempt — not on a results screen afterwards.

In-game capture: a PICK UP MOLECULE prompt floating in the cell interior beside a molecule comparison panel referencing D-glucose.

06 In engine AI Layer

The guidance layer is fed structured learner-state snapshots taken from authoritative game state: movement, object proximity, current selection, wrong attempts, hesitation, task sequence, mission progress.

That is the difference between a tutor that can tell a child has stalled in front of the same structure, and a chatbot waiting to be typed at. Lumi answers what the learner is doing, not only what they ask.

Render of the anatomy world's brain model in profile, its vasculature picked out in magenta across the cortical folds.

07 Source model built Art

The anatomy world's source model is built and exported: cortical regions, white-matter tracts, ventricles, cranial nerves, deep nuclei, vasculature and meninges — separately named, one mesh each.

Naming and separating every structure is unglamorous work, but it is what lets a learner pull the model apart, isolate a single tract and be told what they are holding. A fused brain mesh can only ever be looked at.

Computed render from the carbon-12 nucleus solution: three bright lobes in a triangular arrangement inside a magenta density cloud.

08 Computed Science

Physics Playground's source science is computed: a carbon-12 nucleus, plus electron-density and orbital solutions worked out properly rather than drawn as rings around a dot.

These are renders of the actual solutions, not diagrams redrawn to look scientific. If a child is going to fly through an atom, the shape they fly through should be the real distribution.

09 In source Platform

The hub that links the worlds exists in source — a shared progression profile carrying science coins, XP, levels, badges and mission history, with mission completion verified on the server.

So progress earned in the cell is still there in the arena, and a learner is one person moving through a curriculum rather than five separate save files.

Entries 01–09 · state ladder

  1. Playable
  2. Walkable
  3. In engine
  4. Computed
  5. Source model built
  6. In source

The road ahead

Stages, and one date

Five worlds running in engine, a guidance layer being written around them, and one date we are willing to put in writing. This is the path forward rather than a history, and every stage on it is marked with the state it is actually in.

  • Built — exists and runs today
  • In progress — partly written
  • Next — started after this
  • Planned — intent only

One date, and only one: September 2026. No quarters, no funding, no headcount. Nothing else below is a promise about when.

Stage 01 / 07

Five worlds are already in engine

Built

Living Cell runs: a learner stands inside a cell, picks up a glucose molecule and docks it into hexokinase’s sugar pocket, with Lumi guiding the attempt as it happens. Four more run alongside it, and every frame on this page was captured inside one of those builds rather than mocked up for a website.

  • Living Cell — molecular docking, in first person
  • An orbital station — truss, modules, robotic arm, a repair to make
  • The Parthenon — intact, walked at ground level, with an NPC in it
  • A Roman arena — banked seating, wooden floor, the hypogeum below
  • The cella — the Athena Parthenos in gold, by brazier-light
In-engine capture: a Roman amphitheatre from above, wooden arena floor and banked stone seating, with a player on the sand. Arena — crossed on foot
In-engine capture: the Athena Parthenos standing in gold inside the cella, Nike on her outstretched hand. Cella — Athena Parthenos
In-engine capture: an orbital station of modules and truss against black space, with an instruction panel. Orbit — station repair
Captures from the builds — each of these is walked, not rendered

Stage 02 / 07

Teaching Lumi to read the room

In progress

The layer that turns behaviour into understanding. Raw play is compiled into structured learning-state snapshots, and Lumi is grounded in those snapshots — authoritative game state rather than guesswork — so it responds to what a child is doing, not only to what they type. This is the part currently being written, and the part that has to hold in all five worlds.

Signals feeding the snapshot

  • Movement
  • Object proximity
  • Selection
  • Wrong attempts
  • Hesitation
  • Task sequence
  • Mission progress
In-game capture: a glucose molecule mid-reaction beside an instruction panel reading lock the terminal phosphoryl group.
Capture from the build — guidance surfacing mid-task

Stage 03 / 07

The one date we will put in writing

Next

LUDUMINA WORLDS — science, art and history — is due September 2026. It is our own date, taken from the closing frame of our own trailer, and it is the only one on this site. The rest of this rail is ordered, not scheduled: the stages run in sequence, and each one carries the state it is in rather than a quarter.

Stage 04 / 07

In front of the children it is for

Next

The people who decide whether any of this works are eight to fourteen years old. Beta testing with them is the real test — not whether the worlds run, which they do, but whether a child who has never been told what a hexokinase is comes out the other side knowing.

  • Complete mission chains, start to finish
  • Guidance coverage on every task, including the wrong turns
  • Onboarding, and failure paths that teach rather than just stop you
  • Testing with children aged 8–14
In-game capture: the cell interior with a mission tablet panel on the left reading carry the routine, and a labelled D-glucose molecule overhead.
Capture from the build — a mission running in the cell

Stage 05 / 07

Progress that follows the learner

In progress

One progression profile shared across worlds, so what a child earns inside the cell still counts when they walk into the arena. It lives in the hub rather than in any single world, and mission completion is verified server-side rather than trusted from the client.

The record as it is written in source

Stage 06 / 07

Three worlds still on the workbench

Planned

Human Body, Protein Factory and Physics Playground have not been built. What exists is the material they would be made from — roughly two hundred anatomical structures modelled and named separately, one mesh each, and real carbon electron-density and orbital solutions computed rather than drawn as rings around a dot. The material is finished; the worlds around it are not started.

Render of a detailed anatomical brain model, sulci picked out in magenta against black. Brain — source model
Scientific plot of a carbon electron-density solution, three glowing lobes on black. Carbon — electron density
The material: modelled, named and computed

Stage 07 / 07

From bedrooms to classrooms

Planned

Direct-to-parent subscription first, because it is the shortest route to real children actually using the thing. School licences, enrichment centres and STEM workshops follow from there, in that order and for that reason.

  1. Route 01 Direct to parents Subscription — the first channel
  2. Route 02 School licences After the worlds hold up with real children
  3. Route 03 Centres & workshops Enrichment centres, STEM workshops
Intended order of channels

Closed beta · Ludumina Worlds

Step inside
a living cell.

Five worlds are playable in engine today — Living Cell, The Parthenon, Orbital Station, Roman Arena and The Treasures. Ludumina Worlds ships September 2026. Add your email to the beta list and we will get in touch when family testing opens.

Beta list not yet open — this form is not connected to a mailing list.

  • Ages8–14
  • Built onRoblox
  • Worlds5 playable · 1 planned
  • ShipsSeptember 2026