● NVIDIA Inception Program Member

From SketchUp model to
production CAD drawings
in seconds.

A2CD (Architecture-to-CAD) is an AI-driven pipeline that turns a 3D building model into a complete set of permit-grade construction drawings — architectural and MEP — automatically, with engineering accuracy.

~4s
Full pipeline run
27
CAD / drafting engines
ISO 128
Drafting standard
Arch + MEP
Drawing disciplines

The A2CD core

A2CD is not a wrapper around a chatbot. It's a production system — already built, already running on real projects — with a proprietary reasoning core at its center. We show what it does and proves; the core logic stays sealed by design.

AI vision understanding 27 drafting engines Verified-truth gates BOQ & quantities OpenUSD semantics ISO standards engine 🔒 proprietary A2CD CORE domain reasoning engine
Already built & running

Not a concept. A2CD already produces full construction sets from our own live projects, end-to-end, today.

Proprietary reasoning core

The logic that turns raw geometry into correct, code-compliant drawings is ours — encoded from years of real studio drafting, kept in-house.

Verified-truth by design

Every output is backed by traceable evidence and gated against fake-pass. It never ships what it can't prove.

Standards-native, not generic

ISO 128 / 129 / 19650 built into the engine and benchmarked to professional reference sets — not a plain CAD export.

🔒
We openly share what A2CD delivers, the stack it runs on, and the standards it meets. The core reasoning that makes it work — the genuinely hard part — stays private by design, to protect it from being copied.

What A2CD does

Architectural design lives in 3D. Construction needs 2D drawings. A2CD closes that gap — automatically generating the dimensioned, standards-compliant drawing set that today takes a drafting team days to produce by hand.

📐

Auto drawing generation

Plans, elevations, sections, roof plans and details extracted directly from the model geometry, with automatic dimensioning per ISO 128/129.

🏗️

Architectural + MEP sheets

Produces a coordinated multi-sheet construction set, benchmarked against professional reference drawing standards.

🧠

AI feature recognition

Vision-language models classify rooms, openings, circulation, shading devices and structural tiers — so the output reads like an architect drew it.

📊

Quantity take-off (BOQ)

Automatic bill of quantities and cost estimation grounded in local construction rates and material standards.

🔍

Clash & completeness checks

Cross-discipline clash detection and a drawing-completeness gate ensure nothing ships half-finished.

📄

Print-ready PDF sets

Composes title-blocked, plotted PDF drawing sets ready for permit submission and site issue.

How it works

A single command takes a raw SketchUp file to a verified drawing set — no manual drafting in between.

1

Ingest

Parse the SketchUp (.skp) model and verified dimensions.

2

Understand

AI classifies spaces, openings & structure via NVIDIA NIM vision models.

3

Generate

27 engines draft plans, elevations, sections & dimensions in parallel.

4

Validate

Clash, completeness & standards gates — no fake-pass.

5

Deliver

Title-blocked, plotted PDF construction set + BOQ.

Inside the engine

Not a black box. A traceable, standards-aligned pipeline orchestrating real AI tools — every drawing backed by evidence.

1 · End-to-end pipeline with audit trail

One command: SketchUp model → verified construction set. Every stage logs evidence, mapped to ISO 19650 information states.

Ingest.skp + verified dims UnderstandAI vision classify Generate27 drafting engines Validateclash · completeness DeliverPDF set + BOQ AUDIT TRAIL hashed evidence logged at every step — no fake-pass, fully reproducible ISO 19650 STATE WIPSHAREDPUBLISHEDARCHIVED

2 · AI Manager — multi-LLM orchestration

A routing AI manager dispatches each task to the best model, fuses the results, then a verified-truth gate blocks anything unproven.

Model + viewsgeometry · renders AI Managerrouter + scheduler Gemini VisionVertex AI · reasoning NVIDIA NIMvision LLMs · on-GPU Reasoning LLMarch standards logic Fused modelsemantic + geometry TruthGuard gateverified-truth only

3 · OpenUSD & Omniverse geometry stack

An OpenUSD backbone carries lossless 3D semantics from authoring to drafting, accelerated on NVIDIA Omniverse / RTX.

SketchUp .skpauthoring USD Authoringpxr · semantic attrs OpenUSD Stagesingle source of truth Omniverse / RTXPhysX · NanoVDB StageProcessor3D-info → drafting LOSSLESS 3D SEMANTICS — rooms, openings, structure & MEP carried end-to-end, never flattened to dumb lines

Built on an enterprise AI stack

A2CD is engineered on NVIDIA accelerated computing and an OpenUSD geometry backbone, with architectural reasoning powered by Google Cloud Vertex AI & Gemini, orchestrated by a multi-LLM engine.

Google Cloud Vertex AI Gemini (Vision + Reasoning) NVIDIA NIM Vision LLMs USD / Omniverse NVIDIA AI Enterprise Multi-LLM orchestration PhysX NanoVDB ISO 128 / 19650 Python pipeline
  • Vision-language understanding — facade, room and opening recognition from model & rendered views.
  • OpenUSD interchange — lossless 3D semantics bridge between authoring and drafting.
  • Verified-truth gates — every output backed by traceable evidence, never an unchecked claim.
  • Seconds, not days — full drawing set produced ~15× faster than the original baseline.

Who it's for

Anyone who designs in 3D but is forced to deliver in 2D.

🏛️

Architecture firms

Cut drafting hours and free designers from repetitive documentation work.

🔧

Engineering & MEP teams

Coordinated, clash-checked discipline sheets generated from one source model.

🏘️

Developers & builders

Faster permit-ready documentation and instant quantity take-offs for budgeting.

About MDMA Studio

MDMA Studio (Minimal Design Machine Architecten) is a Bangkok-based architecture and design practice. A2CD is our in-house AI product, born from real production drafting workflows on our own projects — built to give every studio the documentation speed we needed ourselves.

● NVIDIA Inception Program — verified member

MDMA Studio is an accepted member of the NVIDIA Inception Program, which has reviewed our company and product. A2CD is built on NVIDIA's accelerated-computing and AI software stack.

Request a demo or pilot

A2CD is in active development and pre-launch. We're onboarding early pilot partners — reach out to see it run on your own models.

✉️ apichit.k@mdma-studio.com
📞 +66 98 947 4464
🌐 mdma-studio.com