Skip to content

DesignGenAI

DesignGenAI

Generates high-fidelity, structurally valid designs from semantic type embeddings using ANE-accelerated neural networks.

✨ Features

DesignGenAI bridges the gap between abstract design intent and concrete, functional UI code. Our system leverages advanced neural architectures optimized for rapid, high-quality design synthesis.

Semantic Data Conversion

Automatically transforms 50-100 existing design JSONs into structured, labeled training datasets for robust learning.

ANE-Accelerated Modeling

Utilizes specialized ANE-accelerated neural networks to map design type embeddings directly to component JSON vectors.

High-Fidelity Generation

Generates distinct, valid HTML pages, achieving over 80% structural validity from model output.

🚀 Quick Start

Getting started with DesignGenAI is straightforward. Clone the repository and install the necessary dependencies.

Installation Command: 
pip install designgenai[all]
Get Started Now View API Docs

📚 Documentation

Explore the core components of the DesignGenAI pipeline:

  • Dataset Builder: Learn how dataset_builder.py prepares your initial design corpus.
  • Model Architecture: Dive into model.py to understand the 2-layer embedding mapping.
  • Training Pipeline: Review train.py and the integration with ane_trainer.
  • Generation Flow: See how generate.py synthesizes and converts designs into HTML.