Laboratory Knowledge Management for an Additive Specialist

Checking the Data Quality

As technical framework for the AI lab assistant, a RAG solution architecture (Retrieval Augmented Generation) was chosen. RAG is particularly characterized by its ability to effectively leverage unstructured data from text documents without requiring specialized preprocessing.

The primary effort in the project focused on the domain-specific evaluation of the laboratory knowledge in the documents, as well as fine-tuning the AI assistant and its chat logic to produce good answers.

For this, the project team first deployed [at]’s best-practice RAG architecture on an Azure native stack. In such a controlled environment, it was systematically assessed to what extent existing data would suitable for producing informative answers for lab users, and whether the technical implementation would be viable for a full rollout.

Ensuring Data Protection

Given the highly sensitive nature of the data, strict security and compliance requirements were put in place from the start. The solution was hosted exclusively within the European Union and further protected via isolation measures. This was complemented by integration with a security operations center for abuse monitoring, as well as the implementation of guardrails and rate limits to prevent bulk extraction of information and misuse.

User Acceptance Testing

Following the positive functional evaluation of the data, a broad User Acceptance Test was conducted. Here, future users successfully validated the solution in real-world laboratory usage and confirmed its functional suitability and tangible value for daily operations. In addition, a rating system enabled users to provide feedback on specific responses and recommendations generated by the AI lab assistant. 

This feedback was also evaluated after the system went live to support continuous quality improvements.