What drove the decision to develop an AI-based use case?
The development of machine learning models involves a great deal of manual effort, fueled by alack of transparency and (future) complex regulatory requirements (e.g. EU AI Act). A transparent and well-documented development process will not only be a regulatory necessity, but is also crucial for trust in AI systems, especially in sensitive areas such as finance and healthcare. However, this documentation effort is time-consuming and often unpopular and neglected with technical teams. This is where the need for an AI-based solution comes in. The use of Large Language Models(LLMs) can significantly reduce the amount of manual documentation required in the development process. LLMs make it possible to ensure efficient, transparent and secure AI systems and at the same time create the basis for AI governance in order to be prepared for future AI regulations.

What were your expectations or requirements for the AI application?
AI-generated documentation aims to significantly speed up the development of AI systems through strong automation, thereby saving time. Additionally, internal and external governance processes are integrated into the development process through automated recommendations. The application is designed to deliver not only fast but also highly accurate, qualitative results. A central aspect of the application is to demonstrate the explainability of development decisions. The resulting documentation comes in various formats and levels of detail to simplify communication among interdisciplinary project stakeholders. Examples include comprehensive tech documentation (AI team), audit documentation (compliance/legal), and business reports(management). Overall, trail pursues a holistic approach that emphasizes not only technical performance but also user-friendliness and accessibility.

What frameworks/methods did you rely on for development?
trail's documentation engine is based on a current LLM hosted in Switzerland, which provides a  robust and flexible foundation. The governance component is based on established frameworks from research, in particular the work of the Fraunhofer Institute, the standards of the IEEE and the draft of the EU AI Act. This combination of leading AI technology and recognised governance guidelines makes it possible to generate future-oriented and trustworthy documentation.

The use of AI solutions must also be accompanied by a change in competences. How have you dealt with this challenge in respect of your employees?
Following the development of application-oriented AI solutions, governance and trust are (often belatedly) coming into focus - areas in which both technical and compliance stakeholders often still have knowledge gaps. To overcome this challenge, trail offers educational content and a user-friendly software interface. These measures facilitate the acquisition of knowledge regarding AI governance and framework requirements. trail's approach is to work closely with machine learning teams while providing insights and understanding to non-technical project stakeholders. trail thus directly addresses the problem ofa lack of interdisciplinary competences.

In what timeframe did they implement the AI solution in your company/your client‘s company?
trail's application is flexible and can be integrated into a project regardless of the maturity level of the AI system. It is suitable for use at the start of AI projects as well as for subsequent integration into existing projects, and can be used throughout the entire life cycle of the AI system. Ideally, trail is implemented at the start of a new development project so that governance factors can be taken into account directly.

A conclusion to your AI solution: This application represents a significant advance in the documentation of AI development. It enables large time savings - instead of spending up to 40 hours on documentation for a project, the effort is reduced to just one hour. In addition to the efficiency gain, trail ensures greater trust and transparency both internally and towards customers. The application also optimally prepares companies for audits by ensuring that all regulatory requirements are taken into account from the outset.

One final piece of advice/tip to other entrepreneurs looking to apply AI: For companies that produce AI, comprehensive documentation is essential for there producibility of results. Anticipating regulatory requirements at an early stage is crucial to ensure compliance in existing projects. It is also important to consider all parties in the development process and to simplify communication channels. This ensures transparency as well as technical progress.

https://trail-ml.com

Picture: trail GmbH