# Formas will open Cartesian preview for editable AI-generated CAD models > Source: > Published: 2026-09-15 16:10:50+00:00 # Formas will open Cartesian preview for editable AI-generated CAD models **Yiping Goh and Carlos Banon are betting structured solids and BIM context can move generative design beyond polished renderings.** By [RuntimeWire Staff](https://runtimewire.com/author/runtimewire-staff) ยท Published Primary source: [Formas](https://www.formas.ai/cartesian) ## Why it matters AI design tools have been strongest at producing images. Cartesian is betting that editable solids, named parts and CAD exports can move generation into the actual architecture and product-design workflow. [Yiping Goh](https://www.tatlerasia.com/people/gohyiping?ref=runtimewire) and [Carlos Banon](https://www.sutd.edu.sg/profile/carlos-banon/?ref=runtimewire) will begin preview access on September 18th for [Cartesian](https://www.formas.ai/cartesian?ref=runtimewire), a new product from [Formas](https://www.formas.ai/?ref=runtimewire) that turns prompts, sketches, photographs and plans into editable 3D models for architecture and product design. Access will be selective. Formas is taking applications through a [preview waitlist](https://www.formas.ai/cartesian/waitlist?ref=runtimewire), with priority for existing Founding Circle and Studio members, and says joining the list does not guarantee an invitation. The founders bring two different histories to the same problem. Goh previously co-founded promotions marketplace All Deals Asia, which Lippo Group acquired, before helping launch Indonesian e-commerce platform MatahariMall. She later worked in venture capital at Quest Ventures and became a McKinsey partner. Banon is an architect, associate professor at the Singapore University of Technology and Design, and co-founder of SUTD's Architectural Intelligence Research Lab. His research has covered parametric design, artificial intelligence and 3D-printed building components. That pairing explains Cartesian's focus. Goh has approached Formas as a software founder trying to replace fragmented workflows. Banon has spent years working with the geometry, fabrication constraints and professional conventions that make architecture difficult to reduce to an image-generation problem. ### Generating the model underneath the rendering The central claim behind Cartesian is that generative design should produce an editable model, rather than leave a designer with a convincing picture and the job of rebuilding it in CAD. Formas says Cartesian divides a scene into identifiable objects and preserves their relationships. In its restaurant demonstration, chairs, tables, plants, booths and structural columns remain separate parts. A user can ask to change a chair while retaining the room, or keep booths fixed while reconstructing the rest of an interior from a photograph and rough plan. Formas also says the underlying geometry includes inspectable faces, edges and solids, with named elements and explicit relationships that can feed into building information modeling workflows. Public [model studies](https://www.formas.ai/cartesian/examples?ref=runtimewire) cover interiors, furniture, residential projects, urban design, complex structures, manufacturing and 3D printing. The examples are unusually tangible for a generative 3D preview. Formas lets visitors download 3DM and STL files, including a restaurant model listed at 130 MB in 3DM and 112.9 MB in STL. The product page describes 3DM as native geometry measured in meters and STL as mesh coordinates measured in millimeters. Meshes therefore remain part of the export story. Formas' technical distinction rests on the claim that Cartesian can create structured solids, native surfaces and editable assemblies before exporting a mesh when a workflow requires one. Some larger examples are explicitly labeled as mixtures of NURBS solids and mesh parts. The BIM pitch also has a clear boundary at preview. Formas lists native 3DM and SketchUp exports, while DWG and IFC support are marked as planned. Named objects and relationships may provide a route into BIM, but that structure is different from demonstrated interoperability with production Revit or IFC workflows. ### An architect and an operator take on legacy design software Formas grew out of Goh and Banon's shared argument that architecture software forces designers to spread conceptual work, modeling, rendering and presentation across separate tools. In Formas' [April funding announcement](https://www.formas.ai/blog/formas-ai-raises-ususd3-98-million-in-oversubscribed-pre-seed-round-to-build-the-ai-native-design-workspace-for-architecture?ref=runtimewire), Goh described architecture, engineering and construction as a large, under-digitized market burdened by steep learning curves. Banon framed the product around preserving designer intent. Cartesian pushes that thesis deeper into the workflow. Formas' existing products concentrate on visual exploration, rendering and animation. Cartesian is an attempt to generate the underlying design artifact that professionals can measure, modify and move into established software. That places Formas between two entrenched categories. Rhino, SketchUp, AutoCAD and Revit give professionals detailed control but demand substantial training. AI visualization products such as [Veras](https://www.evolvelab.io/veras?ref=runtimewire) can generate and edit architectural imagery using sketches or existing models. Cartesian is trying to start earlier, generating the editable 3D structure from the designer's instructions and reference material. The promise carries a higher technical burden than image generation. Architectural geometry has to remain coherent across edits. Components need usable dimensions and relationships. Curved surfaces, assemblies and topology must survive export. Fabrication introduces tolerances and material constraints that a visually plausible model can ignore. Formas has published controlled examples across a broad range of design tasks. The September 18th preview will show how the system performs when outside users bring incomplete drawings, contradictory references and projects that were not selected for a product demonstration. ### $3.98M to build past the rendering layer Formas announced a $3.98M pre-seed round on April 29th, led by [Vertex Ventures Southeast Asia and India](https://www.vertexventures.sg/projects/formas-ai/?ref=runtimewire). UntroD Capital, Hustle Fund, Big Sky Capital and Orvel Ventures participated alongside angel investors including PatSnap co-founder Jeffrey Tiong, members of the Zopim founding team, Kantox co-founder Antonio Rami and Hammertech co-founder Bradley Tabone. Formas did not disclose a valuation. Formas said at the time that users had generated over 500,000 designs across 135 countries since the broader platform entered public beta in November 2025. Those figures are self-reported and describe generated designs, rather than paid customers, recurring revenue or professional projects completed in production. The financing gives Goh and Banon room to turn early design activity into a product that can sit inside professional workflows. Formas said the proceeds would support hiring, product development and sales, with the US and Europe identified as primary markets. The gated preview gives Formas control over the first set of projects and users. It also raises the stakes for how Formas handles professional design data. The [Formas privacy policy](https://www.formas.ai/privacy?ref=runtimewire) says anonymized generations from Free and Basic plans may be used to improve its models, and selected work from those plans may appear in public galleries. Pro, Education and Enterprise work is private by default and excluded from training without consent. Cartesian is a logical product for this founding pair. Goh knows how to package complicated workflows into software and sell into new markets. Banon brings direct experience with the geometries and fabrication methods that generative 3D systems often approximate. Their preview will test whether those backgrounds can produce models that remain useful after the first impressive image disappears from the screen.