This paper describes our participation in SemEval 2025 Task 8 (Tabular Question Answering). We developed a zero-shot pipeline that leverages large-scale language models to generate functional code that extracts relevant information from tabular data based on input questions. The approach is a modular pipeline centered around a core code generation module, with additional components that improve extraction accuracy by identifying relevant columns and analyzing data types. If the generated code fails, an iterative improvement process is implemented, incorporating error feedback into new generation prompts to enhance robustness. We demonstrate that zero-shot code generation is a viable approach for tabular question answering without task-specific fine-tuning, ranking 33rd out of 53 participating teams during the testing phase.
