Thesis author: Christian Bick
Legacy system modernization carries significant risks ranging from technical (data loss, integration failures, performance degradation) to organizational (resistance to change, knowledge gaps) and financial (budget overruns, service disruption). These risks are scattered across heterogeneous sources and lack a machine-readable structure suitable for integration with modern AI-based decision support tools. Existing risk catalogues in software engineering are typically unstructured prose embedded in academic papers, making them difficult for LLMs to reliably extract and reason over. There is no established methodology for transforming software engineering domain knowledge into LLM-optimized formats. Risk identification in modernization is typically ad hoc and project-specific, and the intersection of legacy modernization risk management and LLM-based tooling is virtually unexplored. With the emergence of LLMs as powerful tools for knowledge retrieval and advisory, there is an opportunity to prepare modernization risk knowledge in an “LLM-acceptable form”: structured, self-contained, taxonomically organized, and prompt-engineerable for effective retrieval-augmented generation (RAG). The aim of this bachelor thesis is to systematically gather and structure the risks associated with holistic legacy system modernization into a knowledge base specifically designed for consumption by Large Language Models (LLMs). By extracting risks from academic and grey literature, categorizing them across technical, organizational, business, and process dimensions, and formatting them as structured risk cards, this thesis creates an LLM-ingestible risk catalogue with associated prompt templates for risk querying and assessment.
