»The collaboration with Fraunhofer was very pleasant and technically convincing. The prototype developed during the Quick Check provides a good foundation on which to build further within an Exploring Project. Overall, we see great potential for further collaboration.«
Maurice Knoll
Novum Analytik GmbH
Contact at the AI Innovation Center
Chandan Kumar
AI-powered transformation of classic laboratory processes
Initial situation
Novum Analytik GmbH is an accredited specialist laboratory for microbiological examinations, supporting food manufacturers, processors, and retailers in complying with legal requirements, minimizing risks, and ensuring food safety.
A central bottleneck in the laboratory process is the manually performed order entry. Reviewing and analyzing incoming orders, as well as manually entering them into the Laboratory Information Management System (LIMS), ties up significant resources daily. The challenge: Orders are exclusively available as scans, vary greatly in structure depending on the customer, and contain relevant information in different positions – partly supplemented by handwritten notes that also need to be recognized.
Solution idea
Scanned orders are to be automatically evaluated using AI for document processing. Relevant information is to be recognized, structured, provided, and then transferred to the LIMS in a controlled manner. This is intended to reduce manual steps and form the basis for further automation in laboratory operations.
Benefit
Automated order entry processing enables daily time savings of up to 8–9 person-hours. Orders are transferred to the LIMS faster, more consistently, and with a lower error rate, which can shorten turnaround times and improve laboratory utilization. In the long term, further laboratory activities such as semi-automated test report generation can be integrated using similar AI methods, saving additional work hours and sustainably improving documentation quality. Overall, this creates a solid foundation for further automation, strengthening both efficiency and compliance and data quality in laboratory operations.
Implementation of the AI application
Several approaches were tested in the Quick Check: rule-based methods and approaches with large language models (LLMs). A prototype implemented based on Fraunhofer's AIKIDO software for robust form processing, including OCR and variable layouts, is particularly promising. The next planned steps are a pilot phase in real-world operation, the integration of user feedback, tuning of the extraction rules/models, and quality assurance through testing and monitoring.