Generative Artificial Intelligence (AI) has become an integral tool across various industries. According to research done by Solita and IRO Research on the top 500 companies in Finland, about 51% of large companies said they are currently using or preparing to implement generative AI into their systems (Solita). In embracing this AI growth, the ÄLLITÄ project aims to figure out ways in which generative AI can make everyday menial tasks at work much easier.

In the first pilot we showcased the development of software for microcontrollers using Artificial Intelligence. Open-source AI models were deployed on the on-premise Nvidia DGX A100 server, which were then used to generate code for an ESP32 microcontroller. Continuing the topic of software development, our next pilot displays the use of generative AI in web development. The newest open-source model from Facebook (meta-llama/Llama-3.3-70B-Instruct) was utilized for developing a website’s frontend and backend code.

As seen in Figure 1, the website allows users to upload article files (pdf or txt) and summarize them. The frontend features of the web application were developed using React.js, while the styling was made using CSS. Once a user uploads the article, it is then sent to the backend side via an API request, where another AI model handles the summarization of the article. The summarization model that was used is Facebook’s Bart model (bart-large-cnn). This model was trained on various CNN Daily Mail articles and can summarize English text only (Hugging Face). The backend and frontend, as well as the communication between them, was fully created by the generative AI model. A visual representation of the structure of the web application can be seen in Figure 2.

As seen in the flowchart, once the summary has arrived at the front-end side, it will then be shown to the user. Figure 3 below shows a summary of an article regarding District Heating displayed on the web application.

Throughout the development of the web app, the generative AI model was also used for debugging the page, when certain aspects didn’t work. The placement of logos and other items on the page is often a difficult task, but the model had no issues handling such tasks. In Figure 4, you can see the model being prompted to fix one of the issues with the ÄLLITÄ logo at the top, and the generative AI is able to answer with multiple solutions to the problem.

The pilot aims to demonstrate how AI can reduce or eliminate time-intensive tasks. We showcase generative AI’s capability to develop fully functional web applications and create practical tools using summarization models. Generative AI supports developers in frontend/backend development and debugging, while summarization tools improve research efficiency across all company sectors.

The next pilots will aim at advancing the development of web applications, by increasing emphasis on styling. We will also display more of the debugging skills of the generative AI, by analysing code that is more complex in its nature.

References:

Hugging Face, “BART (large-sized model), fine-tuned on CNN Daily Mail” https://huggingface.co/facebook/bart-large-cnn, accessed on January 27, 2025.

Solita, “Use of Generative AI survey report: TOP 500 companies slow to adopt generative AI”,https://www.solita.fi/news/use-of-generative-ai-survey-report-top-500-companies-slow-to-adopt-generative-ai/, accessed on January 27, 2025.

Authors:

Shahbaz Baig, RDI Specialist, DigiCenter, Savonia University of Applied Sciences, shahbaz.baig@savonia.fi

Premton Canamusa, RDI Specialist, DigiCenter, Savonia University of Applied Sciences, premton.canamusa@savonia.fi

Mika Leskinen, RDI Specialist, DigiCenter, Savonia University of Applied Sciences, mika.leskinen@savonia.fi

Aki Happonen, Digital Development Manager, DigiCenter, Savonia University of Applied Sciences, aki.happonen@savonia.fi

Laura Leppänen, RDI Specialist, Project Manager, Savonia University of Applied Sciences, laura.leppanen@savonia.fi