Design and Evaluation of an LLM-Based Agentic Workflow for the (Semi-)Automation of Public Tendering Processes in Germany

Despite the digital transformation, participating in public tenders is often hindered by legacy platforms lacking modern APIs and a vast sea of unorganized documents. This thesis develops and prototypes a cutting-edge workflow using LLM agents to bridge this gap. The proposed system utilizes vision-capable agents for web navigation and advanced document processing to automate the sifting of tender documents, checking of eligibility criteria, and filling of complex calculation sheets. A central component is the development of a structured tender database that enables the AI to suggest market-aligned pricing via RAG. The evaluation focuses on the degree of automation achieved, time-savings compared to manual processing, and the error rates of various agentic architectures. This work demonstrates that while routine tasks can be automated by up to 70%, the synergy between AI efficiency and human expertise is key to navigating the legal and economic complexities of German procurement law.