TUP develops its own AI ecosystem

In the myTUP project, the company is developing internal solutions that enable employees to leverage generative AI and automation in a targeted manner. The company-wide toolkit enables intuitive and, above all, secure interaction with internal and external AI services. The ecosystem already includes a variety of workflows and agents that offer direct and scalable added value: from the automation of repetitive administrative processes to drafting important documents as well as assistance systems in development and project management. This article provides an overview and shows that AI is not just for large companies.

Generative AI, especially LLMs (large language models) such as ChatGPT, is a tool used by many people on a daily basis. This is also shown by the first ChatGPT usage survey. It provides interesting insights:

• ChatGPT penetrates almost all demographic groups.
• The biggest application scenarios are instructions, answers to questions, and text production.
• When it comes to texts, users tend to ask for adjustments and feedback on existing content.

As stated in the OpenAI study, LLMs such as ChatGPT enhance judgment and productivity, especially in knowledge-intensive professions. In addition, the agentic use of AI is increasingly coming to the fore: the automation of repetitive and time-consuming tasks reduces the workload on employees. They can focus on strategic and creative activities, thereby increasing efficiency. It is no coincidence that “AI in business” has quickly become a buzzword that numerous service providers and platforms have adopted.

How should AI and automation best be implemented within a company? That is the key question. However, there is no one-size-fits-all answer: to quote TUP’s own slogan, “Software follows function,” it all depends on the company itself, as is the case with warehouse management systems. TUP has already chosen a path and, as in the past, is happy to lead by example.

TUP has extensive IT expertise in its industry. In addition, the company also uses powerful technical infrastructures to operate its intralogistics software systems. These factors lead to synergy effects and the possibility of pursuing more flexible implementation strategies for an in-house AI solution. CIO Eduard Wagner says that the company is in a favorable position to adapt the technology and turn it into a genuine in-house solution. Even with this basis, the company is also committing the necessary financial resources to its TUP.AI division.

The TUP.AI division is responsible for structuring, bundling, and standardizing all AI and BI initiatives within the company. The goal is to develop intelligent tools that improve operational processes, simplify access to data and knowledge within the company, and thereby also accelerate decision-making. TUP can already point to various experiences and projects in this area – from industrial AI applications to smart data analysis. In addition, activities related to generative AI solutions, including automation, are now also part of the division.

TUP strives to achieve these efficiency gains by applying a sound, future-proof approach in which knowledge and sovereignty remain within the company. On the way to the myTUP AI ecosystem, the first step was therefore to create a suitable framework for implementing generative AI solutions.

The decision to create an in-house AI ecosystem is also influenced by factors such as data protection, security, and sovereignty. Companies not only invest in technologies, they are also subject to legal requirements. With regard to compliance with data protection regulations, the General Data Protection Regulation (GDPR) is particularly noteworthy in Europe; however, regulations such as the EU’s Artificial Intelligence Regulation must also be taken into account in the future – in German, this is known as the KI-Verordnung (KI-VO), while in the European context, it is referred to as the AI Act.

The Conference of Independent Data Protection Supervisory Authorities of the Federal Government and the States recommends in its guidance on AI that data protection management routines be established in order to regularly adapt internal guidelines to technical developments and potential risks. An internal AI system does not pose any obstacles in this regard and thus enables a lower-risk handling of sensitive information. Another argument in favor of a company-owned system is the possibility of developing a customized usage variant that allows the use of external solutions without losing one’s own insights or the opportunity to deepen internal expertise.

Independence from APIs and specific model or LLM providers ensures both security and greater resilience for the company. This digital sovereignty makes it possible to flexibly integrate various technical solutions without being tied to specific providers in the long term. This allows the company to react quickly to market changes and adapt to technological developments while retaining control over the technologies and data used: The system is designed so that the underlying LLM model, connected via API, can be flexibly exchanged. This not only increases freedom of choice, but also boosts the company’s innovative potential.

The extent to which AI solutions can actually be implemented in-house by companies is ultimately a question of technical expertise. TUP has its own technical teams from the ITOPS, DEVOPS, data engineering, and development areas, allowing for viable internal implementation. Utilizing existing technical resources has additional advantages, ranging from independence in further development to improved acceptance among employees.

The structure of the AI ecosystem at TUP comprises the following layers:

• Self-hosted instances
  • Reactive provisioning keeps the services available and starts them when the user begins workflows or other triggers are activated.
  • Managed services are AI solutions customized to the company.
  • For critical internal processes, for example specification drafts with client data, secure, company-owned LLMs can be used.
• External providers are restricted by interfaces and legally binding account management, so that no critical data can leak out.

The system enables employees (users) to interact intuitively and securely with internal and external AI services. The solutions are integrated into existing front ends, such as the company chat. Backend and workflow automation is implemented via the open-source platform n8n. There, internal and external tools from a wide range of areas – ERP, administration, project management, knowledge management, GitHub, communication such as chat or email, calendars – are brought together. So-called AI-based orchestrators link the tools in intelligent work processes and ensure active knowledge exchange between the systems. TUP uses the LangChain AI framework for this purpose.

An MCP server (Model Context Protocol) enables dynamic exchange between tools such as mail clients or GitLab and the various AI agents. This allows additional company tools to be linked to agents quickly and securely, without having to design a completely new control logic each time. The server checks incoming requests and dynamically routes them to an AI model that is adequate in terms of data protection:

• internal models
• private cloud LLM instances, isolated from the public internet
• public LLM services, such as those from OpenAI or Anthropic.

As a first step in the myTUP AI ecosystem, the company provides its employees with its own framework that enables them to use AI in a controlled and secure manner. What makes this special is that it can be used as a basis for setting up a wide variety of services and, above all, for continuously optimizing them – from automation solutions and chatbots to business-specific approaches.

Actively shaping the topic also provides an incentive to learn about the associated opportunities and make extensive use of them. The latter is additionally supported by comprehensive documentation and internal presentations on operational use.
The company-wide knowledge chatbot makes internal knowledge centrally and quickly accessible. Users no longer have to search through different platforms, but receive all answers bundled in one place in the chat. The tool currently already accesses various internal sources; the integration of project-specific information is still in preparation, but is already foreseeable.

Among other things, the service also supports various application models that are predefined for a specific purpose, similar to the so-called GPTs in ChatGPT. When it launches, myTUP will offer application models for GDPR-compliant pseudonymization, file translation, use case creation, and support for prompting itself, with a focus on employee support.

In terms of automation, myTUP will continue to take over various work processe. The following were handled directly after its initial launch:

• an automatic newsroom in the company chat, where current articles are summarized briefly and concisely as soon as they are published,
• automated processing and calendar entry of sick leave and absence notifications submitted by email, as well as notes on home office times,
• automatic booking of meeting rooms for internal meetings and customer appointments.

Further fields of application and use cases will be gradually developed and added.

The identification of use cases, especially in the area of automation, must be considered a dedicated task, especially at the beginning. TUP has also set aside resources specifically for this purpose. Employees of the TUP.AI team analyze, in direct consultation with the various divisions of the company, which processes can be automated in a particularly profitable manner.

The goal—as emphasized—is to develop and adapt the technology specifically for our colleagues. Last but not least, part of the implementation strategy is to share identified use cases with the entire workforce in order to raise awareness regarding helpful scenarios or to stimulate further ideas.

Back to home page
To our other blog posts