Competitionline: ASP Architects Build Their Own AI Tools

As an architect at ASP Architects in Berlin, I have started developing our own AI tools. Our experimental group in the Berlin office initiated the project and sparked company-wide interest in artificial intelligence.

Together with Hans Aescht, we initially experimented with facade design, using analog facade studies from a competition as a basis. We developed two innovative approaches:

1. LoRa Model (Low-Rank Adaptation): We developed a specially trained model with 50 different facade patterns that optimizes computational efficiency.

2. ControlNet: We enhanced Stable Diffusion with an extension that provides better control over image generation. We trained the model based on our own 3D visualizations.

In addition to facade design, we also developed tools to increase efficiency in document organization:

• An automated protocol program that creates transcripts and fills pre-made DOCX templates with relevant information
• A test chatbot that accesses our internal documents and emails and answers questions

AI is becoming a fixed part of our daily work. However, we cannot solve all problems immediately and perfectly, but we are improving step by step. The careful selection of tools and the right approach are crucial to fully exploit the potential of AI.

For more information, please visit the original publication in german. (Paywall)