How ZURU improved the accuracy of floor plan generation by 109% using Amazon Bedrock and Amazon SageMaker

ZURU Tech improved its floor plan generation accuracy by 109% using Amazon Bedrock and Amazon SageMaker, developing an innovative text-to-floor plan generator for their Dreamcatcher platform.

• Used generative AI to allow non-technical users to design buildings via natural language prompts
• Explored two primary approaches to improve model accuracy:
  • Prompt engineering with Anthropic's Claude 3.5 Sonnet
  • Full parameter fine-tuning of Llama 3.1 8B model
• Developed a sophisticated evaluation framework to measure model performance
• Achieved significant improvements in instruction adherence and mathematical correctness
• Leveraged AWS services like Amazon Bedrock, SageMaker, and OpenSearch for development

The project demonstrates how advanced generative AI techniques can dramatically improve complex design generation tasks.