ENHANCING TAX PREPARATION THROUGH LARGE LANGUAGE MODELS: A USER-CENTRIC FRAMEWORK

The complexity and opacity of modern tax systems present significant challenges for individuals and small businesses during tax preparation. In recent years, large language models (LLMs) have demonstrated the potential to understand, interpret, and generate human-like language at scale. This paper proposes a user-centric framework that leverages LLMs to enhance tax preparation through personalized assistance, error detection, regulatory compliance guidance, and intelligent document analysis. We analyze the capabilities and limitations of current LLMs, present a system architecture for integrating these models into tax platforms, and evaluate their performance using simulated taxpayer scenarios. The framework emphasizes explainability, privacy protection, and real-time adaptability to user input. Results indicate that LLMs significantly reduce user burden, improve accuracy, and foster greater financial literacy. The findings highlight the transformative potential of language-based AI to democratize access to complex tax knowledge and reduce dependency on traditional, costly tax advisory services.