Fraunhofer FIT launches Generative AI Lab

Many people fail to claim social benefits to which they are entitled. In addition to shame, other possible explanations for this behavior include the complexity of the application process or lack of information. Many existing counseling tools cannot be accessed easily enough, are inflexible or lack interaction capabilities. For Student BAföG, an AI-based chatbot developed by Fraunhofer FIT aims to improve this situation.

There are numerous tools on the internet that can be used to at least roughly calculate entitlements to social benefits. Unfortunately, most of these tools provide rather unspecific advice and fail in complex situations. They are therefore less suitable for counseling in more intricate individual cases. A chatbot based on a Large Language Model may offer a low-threshold service that also provides correct in-depth answers for more complex life situations. Building on our more than 40 years of expertise in the field of BAföG, Fraunhofer FIT has developed a chatbot that helps with the complex individual decision to file for Student BAföG. We hope that it can serve as a blueprint for online counseling services on other social benefits.

Instead of requiring the user to manually enter all the data required for calculating possible entitlements into a predefined input mask, our chatbot extracts this information from a natural-language dialogue with the student. This dialogue also allows for queries to clarify ambiguities, such as the exact definition of income terms, and can provide specific examples of which data is actually required.

To minimize the risk of providing incorrect information, we incorporate the latest research findings on the design and use of Large Language Models. Additionally, we use public information sources such as legal texts, information brochures, structured data and sample cases to improve quality. By combining the Large Language Model with a comprehensive and detailed calculation tool that we developed ourselves, our BAföG chatbot is able to determine a student's eligibility with a high degree of accuracy and reliability.

Besides BAföG, there are a number of other social benefits with complex application procedures that may warrant high-quality online counseling for the application process. Examples include housing benefit and parental allowance. To improve accessibility and quality of online counseling services in these fields we intend to combine our domain expertise in these fields, acquired through many years of project work, and the technological expertise gained in the development of the BAföG chatbot.

In close cooperation with Bausparkasse Schwäbisch Hall, we developed an LLM-based prototype to simplify the internal document search. In a two-week design sprint, we deepened our problem understanding, collected various solution ideas and developed a prototype in an iterative process with real user feedback. The success of the project lay not only in the development of a functioning prototype but also in the creation of a basis for future LLM-based use cases.

Togehter with Bausparkasse Schwäbisch Hall, we have developed an LLM-based prototype to improve internal document search. LLMs, such as ChatGPT, are trained to understand and generate natural language and are often used for tasks such as text generation or text summarization. The LLM-based prototype was developed as part of an intensive two-week design sprint with a heterogeneous team from various departments. Based on the development of a common understanding of the problem and objectives, various application scenarios of large language models for Bausparkasse Schwäbisch Hall were then analyzed and a variety of solution approaches were collected. Specifically, the focus was on the development of a document bot that makes company knowledge easily and intuitively accessible. Throughout the entire design sprint, particular emphasis was placed on a user-centered approach via text input. User feedback was continuously collected and used for the iterative further development of the prototype. This ensured that the prototype met the needs and expectations of the users in the best possible way and offered them real added value in their daily work.

The project was not only successful in terms of developing a functional prototype, but it also laid the foundation for future use cases of LLM technology within the company. The first prototype was presented company-wide to promote understanding and acceptance of the new technology and to identify further potential applications.

Based on the success of the initial prototype and the identified use cases, a specialized project team was formed. This team is now responsible for further implementing and operationalizing the identified use cases. This will ensure that the possibilities of LLM technology are optimally exploited and used to create value.

Large language models (LLMs) are becoming increasingly important for a rapidly growing number of organizations. Many companies are launching trials to use them in internal assistance systems, customer communication or process automation. The challenge for managers often is to fully understand this new technology and to use it effectively.

Fraunhofer FIT has developed a workshop concept that is tailored to the specific needs of managers. Our LLM workshop provides the interdisciplinary understanding and practical skills needed to fully exploit the potential of LLMs. It covers a wide range of topics, including technical fundamentals, business and strategic considerations, as well as social, ethical and legal aspects of LLMs. This comprehensive approach is designed to help participants develop a deep understanding of the preconditions, opportunities, limitations and far-reaching implications of these technologies.

The workshop offers a mixture of theoretical knowledge transfer and practical application. This enables participants to apply what they have learned directly to the context of their company. By combining academic knowledge and practical experience, managers are enabled to identify relevant areas of application for LLMs, evaluate them from different perspectives and implement them effectively in their area of responsibility.

Fraunhofer FIT is responding to the rapid evolution of generative AI by constantly updating the workshop curriculum. This ensures that the content is always at the cutting edge of technology and that participants are prepared for the challenges and opportunities of the future. This promotes innovation and competitiveness in the economy in the long term. By educating managers in the application of LLMs, the program helps to responsibly shape and lead the digital transformation in organizations.