TTZ Research Project Automates the Processing of Complex Documents

The focus was on developing an end-to-end process that automatically recognizes, processes, and prepares printed, digital, and handwritten documents—including contracts and bank statements—for further use.

The project combined various AI methods for document analysis and report generation:

• Specialized OCR pipelines for the German language extract information from complex handwriting and reduce the need for manual processing.
• Large Language Models (LLMs) generate context-sensitive text blocks and populate report placeholders with validated information.
• An “LLM-as-a-Judge” approach automatically checks the plausibility of the generated content and supports quality assurance.
• RAG (Retrieval-Augmented Generation) technologies and the caching of embedding data enable adaptive and scalable processing without the need to continuously retrain models.

The document pipeline developed at the TTZ Günzburg addresses use cases in the regional economy where information from various document types is automatically identified and processed. The methods developed in the project can also be applied to other areas, such as the analysis of legal documents or the automated extraction of form data.

The research demonstrates how AI and OCR techniques can be used to digitize administrative processes. The application of these research findings in practice creates opportunities for companies in the region, as well as for academic teaching and future theses.