ArchiCraft

AI-Driven Conceptual Architecture & Interior Design

Overview

ArchiCraft is a web-based platform that leverages multimodal generative AI to accelerate the early stages of architectural planning and interior design. The application serves two primary functions: transforming existing floor plan images into detailed design concepts, and generating new architectural layouts based on raw constraints (plot size, room requirements). It is designed to bridge the gap between abstract requirements and visual schematics, providing instant feedback for spatial exploration.

Live Demo

URL: https://archicraft-250712192556.us-west1.run.app

Judge Access / Unlock Codes

To prevent API quota abuse during the public viewing period, the "Visualize Interiors" (High-Res Rendering) feature is gated behind an access code.

Judges: Please use the access codes provided in the official hackathon submission details/notes to unlock full functionality.

Key Features

• Floor Plan Analysis: Uploads 2D floor plans to identifying distinct rooms, approximate dimensions, and spatial flow.
• Interior Concept Generation: Automatically suggests furniture layouts, color palettes, and material finishes based on user-selected styles (e.g., Modern, Scandinavian, Industrial).
• Generative Layouts: Creates structural floor plan schematics (SVG) from text-based inputs, such as plot width/depth and required room counts.
• Photorealistic Visualization: Transforms vector schematics into high-fidelity top-down architectural renders.
• Technical Specifications: Outputs structured JSON data for room schedules and supports dynamic unit switching (Meters/Feet).
• Professional Export: Built-in CSS print styling for exporting "Design Reports" to PDF.

Technology Stack

• Frontend: React 19, TypeScript, Tailwind CSS
• AI SDK: Google GenAI SDK (@google/genai)
• Visualization: React-DOM for SVG generation, HTML5 Canvas for rasterization
• Hosting: Google Cloud Run

AI & Google Gemini Usage

This project was built using Google AI Studio to orchestrate multimodal workflows, utilizing Gemini 3 Pro exclusively across all features.

• Gemini 3 Pro Image (Nano Banana Pro 3): Used for the "Visualize Interiors" feature. The application rasterizes the generated SVG schematic and prompts the Gemini 3 vision model to "texture" the floor plan, adding realistic lighting, flooring, and furniture while respecting the structural boundaries defined by the code.

• Gemini 3 Pro Preview: Utilized for the core architectural reasoning engine. Gemini 3 Pro was selected for the analyzeFloorPlan and generateArchitecturalConcept functions due to its advanced reasoning capabilities for complex spatial planning and superior adherence to technical JSON schemas, which is critical for rendering accurate coordinate-based drawings on the frontend.

Project Scope & Limitations

ArchiCraft is a conceptual design tool intended for ideation and exploration.

• Conceptual Only: Generated layouts do not account for local zoning laws, load-bearing structural engineering, or MEP (Mechanical, Electrical, Plumbing) constraints.
• Accuracy: Dimensions are estimates based on standard door widths and pixel-to-meter ratios.
• Professional Review: All outputs should be reviewed by a licensed architect or structural engineer before being used for construction.

Getting Started

To run this project locally:

1. Clone the repository.
2. Install dependencies:

   npm install

3. Set up environment variables:
   • Create a .env file.
   • Add your Google Gemini API key: API_KEY=your_key_here.
   • Note: For full functionality, the API key must be associated with a Google Cloud project with billing enabled to access the Gemini 3 Pro Image model.
4. Run the development server:

   npm start

Hackathon Context

This project was built for Google DeepMind - Vibe Code with Gemini 3 Pro in AI Studio. It demonstrates the practical application of Multimodal LLMs in the AEC (Architecture, Engineering, and Construction) industry, showcasing how generative models can move beyond text chat to produce structured, visual, and spatial data.