Incorporating large AI models into software

How can generative AI be incorporated into software? Researchers and industry partners participating in the international ELFMo project seek ways to develop new digital services on top of large foundational models (LFMs). Among the most well-known examples of these is the large language model ChatGPT. 

Large foundational models can form the foundation for a wide array of AI applications. They may be trained to handle diverse data, such as images, charts or speech.  

“I’m sure many have tried ChatGPT. Our goal is to embed similar functions in software solutions, such as information security software, building automation or e-commerce systems. This will enable the software industry to produce entirely new smart products,” says Professor of Computer Science Jukka K. Nurminen of the University of Helsinki, the project’s principal investigator.  

“In addition, large foundational models can be tailored to suit specific contexts. For example, ChatGPT can do well with general knowledge but does not know the details of a special domain. Technically these systems can be modified to be able to respond correctly in specialised contexts,” explains Nurminen. 

Preventing cybercrime and enhancing construction 

One area of application for large foundational models is information security.  

“The development of generative AI has given cyber criminals new opportunities and tools for more effective scams. Our research and development address this trend to safeguard against threats and attacks. Large foundational models allow us to develop information security technology and expertise, and to enhance the consumer’s information security experience. The models provide new avenues for developing more personalised information security products,” says Abdullah Al Mazed, Head of Protection Concept Lab at the information security company F-Secure. 

AI Lead Davor Stjelja from the construction and real estate expert group Granlund believes that large foundational models present new opportunities for using data in the built environment.   

“In this industry, data exists in various domains – ranging from text, tables, images and drawings to diagrams, BIM models, time-series data and databases,” states Stjelja, adding: “As a result, critical information about the same building is often scattered in silos. By utilising LFMs, we aim to bridge these gaps, enabling a more integrated approach to information retrieval and analysis. Beyond unifying data, LFMs can assist experts by performing lighter analytical tasks, enhancing efficiency and supporting informed decision-making.” 

Researchers note that large foundational models challenge the entire software industry.  

“Foundational models are being integrated into many IT products and services, and technical expertise in such integration is becoming a key competitive factor for the software industry,” says University Researcher Mikko Raatikainen of the University of Helsinki, who leads the European research consortium behind the ELFMo project. 

The University of Helsinki–coordinated project is part of a larger framework, managed by the international software innovation network ITEA, which involves several industry-led research projects. The University of Helsinki will spend €2.2 million on the project, mostly provided by Business Finland. The project will last until September 2027.  

This news item was originally published on the University of Helsinki website on 10.3.2025