Finite element analysis is widely used in structural engineering, but many simulation workflows still depend on manual preparation, specialized expertise, and time-consuming result interpretation. Recent advances in large language models (LLMs) suggest that conversational systems can support engineering workflows, although their use in simulation environments raises important concerns regarding reliability, controllability, and engineer oversight. This thesis presents a human-in-the-loop conversational framework that integrates a large language model agent with the Code_Aster finite element solver through a Python-based orchestration system. The framework allows engineers to describe structural beam analysis problems in natural language while the agent collects missing parameters, validates inputs, and confirms all simulation settings before execution.