Start: 09.02.2026
End: 09.08.2026
Type: Master Thesis
Student: Benedikt Lutz
Supervisor: Prof. Dr. Andreas Harth, Sebastian Schmid
Abstract: The primary focus of this thesis is to investigate the efficacy of an AI Agent in mapping natural language requirements to a set of predefined work steps for automated effort estimation. While Large Language Models (LLMs) offer strong semantic understanding, relying solely on them poses risks regarding hallucinations, plausible but factually incorrect answers, and a lack of domain-specific context. To mitigate these limitations, this work proposes an agentic approach utilizing the ReAct (Reasoning and Acting) framework. By leveraging Tool-Use capabilities, the agent can efficiently interact with the company’s internal knowledge base. This grounds the estimation in deterministic data (standardized process catalogs) rather than relying on the LLM’s generative capabilities, thereby ensuring factual accuracy.