Future-proof AI solution: How a step back can help to go beyond

The first step was a detailed review of the existing PoC. It quickly became clear that the technical architecture used was not suitable for meeting STAEDEAN's requirements in the long term. At the same time, it became apparent that the original PoC had been developed without a clear definition of use cases. Neither roles nor user needs nor specific problems had been described. As a result, the solution path was unclear, functions were ineffective, and the project lost its direction.

A central element of the reorientation was therefore consistent work with customer use cases. Together with STAEDEAN, the following questions were developed:

• Who uses the Connectivity Studio and what role does each feature play in everyday work?

• What specific problem needs to be solved?

• How should the solution be integrated into the existing product?

• Which tasks are really relevant to AI – and which can be solved more easily via user experience or classic logic?

This focus on users was a turning point for the project. It helped to outline where AI promised real added value – and where other solutions made more sense.

Based on this analysis, codecentric and STAEDEAN decided to redesign the PoC. Instead of only focusing on limited data, a RAG (Retrieval Augmented Generation) system was used. The AI model is not retrained with all data, but instead accesses relevant information in a specially optimized database for each query. This is faster, more flexible, and much easier to scale.

This approach was crucial for STAEDEAN because a large number of tables and objects have to be processed. In addition, data changes over time. The solution therefore had to be robust, expandable, and high-performing.

The project was implemented by a two-person codecentric team:

• An infrastructure consultant responsible for architecture, deployment, and operations in Microsoft Azure

• A GenAI consultant who took care of implementation, use case definition, UX consulting, and technical management – including sparring for the STAEDEAN team

A three-person team was involved on the client side. The collaboration quickly developed into a trusting partnership. Joint prioritization and planning proved particularly valuable: instead of building a large solution all at once, the team worked step by step on clearly defined use cases, each of which delivered concrete added value.

On a technical level, the team used the client's existing Azure cloud environment as a scalable basis. A lightweight framework (Streamlit) was deliberately used for the first prototypes in order to quickly test the core idea and iteratively refine the logic. Once this was working reliably, the functions were transferred to a Microsoft Copilot-based solution, which both improves user guidance and provides a compelling showcase for internal and external stakeholders.

It took just under four months from the initial realignment to the finished, stable, and scalable PoC. During this time, the following was achieved:

• the architecture was completely rebuilt,

• use cases were clearly defined,

• UX prototypes were analyzed and optimized,

• the RAG system was developed in line with production requirements,

• performance was significantly improved, and

• the solution was fully integrated into Azure.

The codecentric team not only acted as developers, but also as methodological guides – especially for UX issues, prioritization, and product strategy.