I am delighted with the success of our collaboration! The integration of LLMs into SLAMD makes materials research in the construction sector even more accessible and opens up new avenues for both our students and our partners. We’re receiving lots of enquiries about sharing this tool, and I’m keen to see what further opportunities this interdisciplinary collaboration will bring.
DISCOVERING NEW MATERIALS WITH GENERATIVE AI
Technical University of Berlin: A chat assistant for environmentally friendly concrete mixes.
How Generative AI is transforming science
The major benefit of Generative Artificial Intelligence (Generative AI, or GenAI for short) is that new content can be created quickly and without any specialist prior knowledge, simply by following straightforward instructions. This technology is being discussed across various sectors and is finding particular application in areas such as customer support and product personalisation.
Often overlooked, but no less relevant, is GenAI’s contribution to science, for example in materials research. Together with TU Berlin, we have developed a construction chemistry assistant based on language models (LLMs): this tool enables the discovery of new materials without the need for in-depth prior knowledge of construction chemistry.
Expanding the user base
The production of concrete accounts for 8 per cent of all CO₂ emissions worldwide. To develop environmentally friendly concrete mixes efficiently, we have developed the SLAMD application in collaboration with the Federal Institute for Materials Research and Testing (see short video: SLAMD Explained). This tool is based on machine learning and helps research staff to develop alternative concretes more quickly and, as a result, more cost-effectively. Read the Champion Story
The Technical University of Berlin has set itself the challenge of making this application available to a wider user group. Whereas chemical knowledge was previously required, SLAMD should now be usable even by people without specialist expertise.
Machine Learning Meets Generative AI
In its original form, SLAMD is based on Sequential Learning (SL), a specific type of machine learning. SL is a method that can operate even with very small amounts of data and, step by step, derives predictions for new concrete mixes by building on practical results from laboratory tests.
The result is the Design Assistant, an intuitive chatbot that allows users to find environmentally friendly concrete mixes even without specialist knowledge.
Where traditional machine learning methods fall short, GenAI opens up new possibilities for application:
- The application is designed to be used without specialist knowledge. This knowledge is generated by Generative AI in the first place.
- The method of in-context learning is utilised: with the help of prompts, the AI within the application is enriched with domain-specific context – that is, the specialist knowledge generated previously.
- To make this knowledge available in a user-friendly way, a chatbot has been implemented. This guides the user step by step through the process and requests the necessary information. Based on this information, the relevant specialist knowledge is then generated at the user’s request. This specialist knowledge is ultimately used to send a prompt for formula generation. As a result, the user receives a formula that can be produced in the laboratory.
Your contact
Do you have a specific request or questions about possible AI and data analytics projects for your company? Send a request and we will get back to you.
Dr. Felix Böhmer, Director Al & Data Analytics
