Skip to content
3D.MBA

Platform

Six modules, one pipeline, no seams to fall through

The rendering is a commodity now — three.js and a competent engineer gets you most of the pixels. What does not commoditise is the rule graph, the CAD ingest and the telemetry binding. All three live here.

18
Ingest formats
94%
Median size reduction
10⁹
Addressable permutations
< 400 ms
Telemetry latency

Module 01

Configurator Studio

Detail

Rules, variants and pricing wired to a 60 fps canvas.

A visual rules engine sitting on top of a real-time WebGL scene. Define options, constraints, dependencies and price logic once; the canvas, the SKU, the BOM and the quote all stay in lockstep.

  • Visual rule graph with compatibility, requirement and exclusion constraints
  • Deterministic SKU + BOM emission on every valid configuration
  • Price, lead-time and margin modifiers evaluated client-side in under 2 ms
  • Variant permutation explorer — find the 40 SKUs that never validate
  • Server-side render fallback for email, PDF and print catalogues
  • Shareable configuration URLs with a signed, replayable state hash
Rule evaluation
< 2 ms for 10k-node graphs
Scene budget
180k tris / 60 draw calls
Cold start
1.4 s to first interactive frame
Variant ceiling
10^9 permutations, lazily resolved

Module 02

Spatial Commerce

Detail

3D product pages, AR placement and a cart that understands variants.

The commerce half: 3D on the product detail page, WebXR and Quick Look for in-room placement, and a variant-aware cart that survives the handoff to Shopify, Magento, SAP Commerce or your own checkout.

  • Drop-in 3D product viewer with lazy WebGL boot after first interaction
  • AR: USDZ Quick Look, Android Scene Viewer and WebXR from one glTF
  • Variant-aware cart payloads that round-trip through the order system
  • Automated hero renders and 360 spins per variant, cached at the edge
  • Try-in-your-room with plane detection, occlusion and real-world scale
  • Analytics: which option combinations get built, abandoned and bought
Viewer payload
480 kB median, Draco + KTX2
LCP impact
0 ms — canvas boots post-interaction
AR conversion
USDZ generated in 900 ms, cached
Cart round-trip
Versioned JSON, schema-validated

Module 03

Reality Capture

Detail

Photogrammetry, terrestrial LiDAR and drone survey into usable geometry.

The as-built side. We fly, scan and photograph the thing that already exists, then register, clean, classify and decimate it into something a browser can actually open.

  • Drone photogrammetry and LiDAR missions with GCP-verified accuracy
  • Terrestrial scanning for plant, heritage and interior capture
  • Automatic registration, denoising and moving-object removal
  • Semantic classification: ground, structure, vegetation, equipment, pipework
  • Octree-streamed point cloud publishing — billions of points in a browser tab
  • 3D Gaussian splat reconstruction for photoreal showroom backdrops
Survey accuracy
±18 mm with ground control
Capture rate
12 ha/day drone, 40 scans/day terrestrial
Streaming ceiling
3.2 B points, LOD-paged
Turnaround
Registered cloud in 48 h

Module 04

Digital Twin

Detail

Live telemetry bound to the geometry it came from.

Take the factory model — CAD layout, scan, or both — and bind every machine to its live signal. OEE, temperature, cycle time and alarms rendered in the place they physically occur.

  • Node-bound telemetry from OPC-UA, MQTT, Modbus gateways and REST
  • Live OEE, cycle time, alarm and utilisation overlays per asset
  • Time-scrub a rolling 30-day window and replay any incident
  • What-if layout planning with a discrete-event throughput model
  • Maintenance view: work orders and part history anchored to the machine
  • Robot and AMR path visualisation with reach and collision envelopes
Telemetry latency
< 400 ms end to end
Assets per scene
2,000 instanced nodes at 60 fps
History window
30 days scrubbable, 2 years archived
Protocols
OPC-UA, MQTT, Modbus TCP, REST, Kafka

Module 05

AI Studio

Detail

Generative materials, automatic retopology and 3D semantic search.

The tedious parts of a 3D pipeline, automated: PBR material synthesis from a photograph, mesh decimation that preserves silhouettes, UV unwrapping, and natural-language search across a model library.

  • PBR material synthesis from one photograph — albedo, roughness, normal, AO
  • Silhouette-preserving decimation with per-part triangle budgets
  • Automatic UV unwrap and texture atlasing across an assembly
  • Geometry-derived part naming and classification for unnamed CAD trees
  • Natural-language and shape-similarity search over a model library
  • Text-to-variant: describe a colourway, get a validated material set
Decimation
92% triangle reduction, 0.4% silhouette error
Material synthesis
2k PBR set in 8 s
Search corpus
500k parts, sub-200 ms queries
Human review
Every step diffable and revertable

Module 06

Asset Pipeline

Detail

Eighteen formats in. Optimised, budgeted glTF out.

The unglamorous core. Tessellation, conversion, compression, LOD generation, validation and versioning — run as a build system with a cache, not as a person with a copy of Blender.

  • CAD tessellation with configurable chord and angular tolerance
  • Content-addressed incremental builds — rebuild one part, not the assembly
  • Automatic LOD chain generation with per-level budgets
  • Draco geometry and KTX2 / Basis texture compression
  • Validation gates: manifold checks, budget ceilings, naming conventions
  • Versioned asset registry with rollback and diff between releases
Ingest formats
18 in, 11 out
Compression
94% median size reduction from source CAD
Incremental build
4,000-part assembly in 90 s warm
Manifest
Every output traceable to an input hash

The pipeline

Run it yourself

Pick a source asset and a target surface, then step through. The reductions, the skipped stages and the budget gate are the ones our build system actually applies.

Source asset
Target surface
  1. 1

    Capture & ingest

    13 s

    Whatever the geometry already lives in — CAD release, drone flight, scanner, or a modeller's FBX — lands in a content-addressed store.

    STEP AP242 assembly hashed and stored, 4,120 parts indexed

    Size
    412.0 MB
  2. 2

    Reconstruct & tessellate

    B-rep becomes triangles; photo sets become meshes or splats; raw scans get registered, denoised and classified.

  3. 3

    Optimise to budget

    Decimation, LOD chains, UV atlasing, Draco and KTX2 — until the manifest fits the budget you set.

Current artifact
412.0 MB
from 412.0 MB at source
Build time so far
13 s
warm cache, incremental
Triangle budget
180k
Mobile configurator

Engagements

Four ways to start

Most clients begin with a Feasibility Sprint. It is fixed-price, and it sometimes concludes that the project should not go ahead.

12 weeks

Feasibility Sprint

Fixed price

We take one real product — your worst CAD assembly, not the easy one — through the whole pipeline and put it on a phone. You get the frame budget, the honest triangle count and a go / no-go.

  • One product through the full pipeline
  • Frame budget on a mid-range Android
  • Rule model for the top 20 options
  • Go / no-go with the reasoning
28–14 weeks

Configurator Launch

From $85,000

A production configurator on your storefront: rules, pricing, BOM emission, AR, analytics and the CMS handover so your team owns it afterwards.

  • Production configurator
  • Rules + pricing engine
  • Cart and ERP integration
  • Team handover and source
34–10 weeks

Capture Programme

Per site

Drone and terrestrial capture of a site or plant, registered to survey accuracy, published as a streamable cloud and a clean mesh for downstream use.

  • Flight and scan plan
  • Registered, classified point cloud
  • Streamable web publication
  • Mesh + IFC derivatives
412–20 weeks

Twin Rollout

From $150,000

A live digital twin of one line or plant, wired to your existing OPC-UA / MQTT infrastructure, with the what-if layout model your ops team will actually use.

  • Plant scene from CAD + scan
  • Telemetry bindings
  • OEE and alarm overlays
  • Throughput what-if model
Pricing and ROI calculator

Formats

26 formats, in and out

CAD & engineering

Precise B-rep and parametric formats coming out of mechanical design and PLM.

4 formats

Mesh & exchange

Polygon interchange formats — what modellers, DCC tools and game engines pass around.

3 formats

Scan & point cloud

Reality capture output: LiDAR, photogrammetry and structured-light scanners.

4 formats

BIM & geospatial

Built-environment formats carrying semantics, not just geometry.

3 formats

Additive & manufacturing

What actually goes to a printer, a CNC post-processor or a quality station.

4 formats

Web & real-time

The delivery end — what ships to a browser, a headset or a phone.

5 formats

Neural & volumetric

Radiance-field and splat representations. Photoreal, awkward to edit, improving fast.

3 formats

Open the format explorer

Bring us the product you were told could not be configured.

Tell us the option count, the CAD you already have and the frame budget you need to hit. An engineer reads it and replies with a straight answer — including when the answer is that good photography would serve you better.

Book a demoTry the configurator

We reply to every enquiry within one business day, in your timezone — with a frame budget, not a brochure.